Meta Platforms

We are excited to share with you our latest research report on Meta! As this piece is our longest yet, we decided to release it concurrently with a shorter summary. If you are short on time, we recommend starting there. Please email us at info@speedwellresearch.com for any questions, concerns, or comments. See our full disclaimer on our website.

While many of our readers are familiar with Meta, we would rather err on providing too much info than too little, thus this report generally assumes little prior knowledge of the business. The first 1/3^rd of the report is on business history. The Founding History goes to the start of the company and the Background History goes until about 2019, where we pick up with the rest of the history throughout the rest of the report. Thank you for reading and we hope you enjoy it!

Founding History.

Born in a New York suburb to a dentist and a psychiatrist, Mark Zuckerberg was drawn to two things: programming and empire building games. At just 10 years old, he threw a fit demanding to be taken to a Barnes & Noble to get a C++ coding textbook. However, Mark was disappointed in the superficiality of the book “for dummies”, and his parents ultimately hired a private programming tutor for him. For 5 hours every day after school, and entire weekends, Mark would live at his computer, coding all sorts of projects and games on his IBM. One of the first games he digitized was Risk, a board game of global domination.

His other projects commonly included “hacking” AOL’s AIM messenger, which was the most popular way for teens to talk since launching in 1997. He became so natural on the computer that he would later remark that “It reached a point where it went into my intuition. I wasn’t really thinking that much about it consciously”. Hoping to bolster Mark’s chances of getting into an Ivy league and provide him with a more challenging schooling environment, his parents sent him to Phillips Exeter Academy. Despite being a very rigorous school, he was discouraged from taking AP Computer Science because he was told he knew it all already.

Early AOL Instant Messenger, which is where the acronym AIM came from.

His senior project at Phillips Exeter Academy would be his first project that translated into commercial potential. With his friend Adam D’Angelo, Zuckerberg created a music recommendation program called Synapse.ai. He would work on this project a bit after graduation, ultimately receiving interest from AOL and Microsoft, which materialized into a $1mn+ offer. But the contingency that he work for them for 3 years made it easy to turn down. Mark would later recall, “We knew that we could do something better”. Incredibly, foreshadowing Zuckerberg’s future, another classmate at Phillips Exeter created a searchable online directory of all the students at the school with basic information for his senior project: he called it “Exeter student Face Book”. (A Face Book is a physical book of all the students that were at a school—like a yearbook, but issued at the beginning of the year for students to meet the rest of the class).

Phillips Exeter Academy is a highly selective boarding school in New Hampshire.

In 2002, Zuckerberg started university at Harvard College, but he still spent most of his time on miscellaneous coding projects. One of which, a gig he found on Craigslist from a Buffalo businessman to create a website, would later get him sued for claims on his future creation (the court threw out the case before prosecuting the businessman for forgery). It was typical for him to put his coding schemes ahead of his classes or coursework. 

In 2003, Friendster, a social networking site positioned as “an online community that connects people through networks of friends for dating or making new friends”, was launched. Within a year, they would have 3mn registered users, including Mark. That summer, D’Angelo worked on a project called Buddy Zoo, which allowed AIM users to upload their “buddy list” (a list of people they added on their AIM messenger) to a site which would then tell you who your common friends were, as well as calculate popularity, detect “cliques”, and see degrees of separation. It spread quickly, with hundreds of thousands of users uploading their friends list. The virality of it was in part spurred by users who would upload their buddy list and then implore friends to do the same. This was Mark’s first up close experience in seeing how quickly a service can gain viral adoption and learning how the best marketing strategy is user word of mouth. A strong network grows itself. Friendster and Buddy Zoo would leave a lasting impression on Mark.

On the left is a Friendster profile that shows Friends, profile info, and “Testimonials”. Friendster would later update their site, which is shown on the right with Mark Zuckerberg’s own profile. Friendster was the first social network to gain wide popularity with usage peaking at over >100mn users.

One of his first projects his sophomore year was a program he created called “Course Match” that would tell students what classes other students were taking. This tool allowed people to take the same class as their crush or take courses with friends, and it was enthusiastically adopted until it crashed Mark’s computer. From this, he gathered that “people have this deep thirst to understand what’s going on around them”.

Harvard University.

The next Zuckerberg project would ostensibly be inspired by rejection from a girl. After having a couple of beers, he went to work on creating Harvard Facemash (Oct. 2003). This program would take students’ Face Book photos from various final clubs and residential houses and show a random pair, side by side. The user would then pick the better looking of the two, a take on the popular site “Hot or Not” where users ranked people on a scale of 1 to 10. While Mark only sent a test link to a few friends, it leaked and went viral. Many were deeply offended (while others were quietly jubilant that they had ranked higher than certain peers). He had to answer to Harvard’s advisory board and was put on probation. Any future infractions would be met with increased punishment; however, Mark was already plotting another project.

The Harvard school paper commented on the Facemash ordeal. Some thought Mark’s next project was motivated by a desire to help clear his name.

The Winklevoss twins, athletic rowers who were in the upper tiers of Harvard’s social circles, reached out to Mark after the Facemash debacle. They had been thinking of creating a website that was a sort of Friendster, but only for Harvard students, called Harvard Connect (many university websites were popping up for students, sometimes created by the administration themselves. Columbia had CUCommunity, Yale had YaleStation, and Stanford had Club Nexus).

The Winklevoss twins lacked technical ability and had trouble with past programmers following through, so the site had remained unbuilt for over a year. They wanted Mark to do all of the coding to build the site in return for a vague promise that if it worked out, they would make money together. Mark had taken on various ad hoc projects over the past year and showed interest. There were various email communications over the next couple months with Mark usually indicating that he was too busy to work on Harvard Connect. (A series of unearthed AIM messages from Mark would later show that he was likely stalling working on their “dating site”, but also none of their ideas were unique–the website they described seemed like a mix between Match.com and Friendster for Harvard only with a focus on meeting people to date and get discounts to nightclubs. Of course, the decision to wait and place the entire creation of their website on a single part-time, unpaid student was their decision). Exactly what happened though would be litigated for years.

Mark Zuckerberg coding in his Harvard dorm room.

Around the same time, Zuckerberg reached out to Aaron Greenspan to learn more about his houseSYSTEM program that launched at the beginning of the school year (Aug. 2003). His app was loaded with features, including the ability to exchange textbooks with peers and even a “Face Book” feature that was similar to what Zuckerberg’s classmate at Phillips Exeter Academy had made. However, Aaron made a critical error by requiring users to sign up with their Harvard ID and the same password as their Harvard account (most likely for the email feature). Besides most students’ not trusting a random online site to take their real Harvard passwords, this decision would lead to him butting heads with Harvard IT Security, and thus the little publicity it got was negative. Aside from that, a clunky interface and perhaps too many features (job listings, a calendar, upcoming events, and email, among others) kept it from ever gaining traction. Zuckerberg would message Greenspan over AIM to talk about his creation and learn from his missteps. Mark described the many features as “almost overwhelming how useful it is” and opted for a more streamlined functionality for his project (he would also later be sued by Greenspan).

Greenspan relaunched houseSYSTEM with a “Facenet” extension a couple months after TheFacebook launched. But it was too little, too late. Despite getting the Crimson coverage he felt was absent on the houseSYSTEM’s original launch, it would flutter into irrelevance (but not before he attempted to revive it a 3^rd time as CommonRoom a few years later).

While Zuckerberg was working on his project, a Dec. 2003 Crimson (Harvard’s student newspaper) article seemed to anticipate his creation, saying that the “potential benefits of a comprehensive, campus-wide online facebook are plenty”.  The Harvard administration allegedly was working on one, but few had high hopes for it.

A Harvard Crimson article notes the need for an online “Face book”

We will never know exactly where Mark Zuckerberg got the inspiration for his next project, but it was likely a mix of everything from: 1) his early experiences watching users change AIM statuses (he built an app that alerted him when an AIM user’s status was updated), 2) seeing how D’Angelo’s friend graph took off, 3) observing peoples’ desires to stare at photos from Face Mash, 4) realizing how people want to know what their peers were doing from CourseMatch, and 6) witnessing the slew of social networking attempts bubble up and putter out. He didn’t just have any single run in with social networking, but many of his early projects all seemed to circle around what would in retrospect be his Magnum Opus. 

On February 4^th, 2004, Mark Zuckerberg launched TheFacebook.com. 

The original Facebook launch screen. Note the clean user interface compared to Friendster or other social networking websites like MySpace. The logo on the top left was designed by Zuckerberg’s fellow classmate, Andrew McCollum. It would be retired a few years later.

Business History.

Going Viral.

Just sending out emails to some friends was enough to have 650 students register for TheFacebook within four days of launching. Three weeks later they would have over 6,000, which was the majority of Harvard’s undergraduate population. In contrast to Facemash where his laptop served as the de facto server and easily crashed, Zuckerberg invested $1,000 in the company alongside fellow AEPi fraternity brother, Eduardo Saverin, to rent server space. With rapid growth, that would only last so long; by the end of the month, they had over 10,000 users. Zuckerberg quickly decided to extend the service to new colleges, launching each new school as standalone, closed network (Harvard students did not have the ability to look up Yale students initially).

Zuckerberg purposely chose to launch at schools that already had some kind of online Face Book site as a test to see whether or not his project was truly better. At this point, Zuckerberg was not sure if this would be an enduring creation or just another short-lived, coding fad-project like the others. Beating out Stanford’s Club Nexus, Columbia’s CUCommunity, and Yale’s YaleState gave him the confidence that this was the real deal. After seeing a rapid uptake similar to what he saw at Harvard at each of these schools, he quickly rolled it out to the rest of the Ivy League schools. Just two weeks later, TheFacebook’s user base doubled to 20,000. To keep the growth going, they decided to open up TheFacebook’s social network to cross-campus linking with the “friend request” framework: only upon mutual agreement to be friends could users see each other’s profile.

This original “About” page shows all of the original Facebook co-founders. Eduardo Saverin would later be stricken as a co-founder when he sued in response to what he considered excessive dilution (his stake was reduced over 95% from the originally agreed 30%). His co-founder status was officially reinstated after a settlement.

That summer, Mark Zuckerberg and several engineers moved to Silicon Valley to continue working on TheFacebook. They rented a house in Palo Alto and would code late into the nights. Business partner Eduardo (who originally put up the $1k) instead went to New York City for a finance internship. He set up a bank account and put in over $10k (along with Zuckerberg who also kicked in money) to fund their Palo Alto venture. While Eduardo was allegedly getting advertisers to help fund TheFacebook, he often came up shorthanded, and the Palo Alto team was more concerned with user growth than plastering low quality ads on the site anyway.

Eduardo started feeling threatened by Sean Parker (Napster co-founder and a well-known Silicon Valley figure), who was now living with them after a fortuitous encounter in Palo Alto. Sean Parker was slowly taking on more business functions in Eduardo’s absence, and Eduardo felt the encroachment on his job as Head of Business. Eduardo froze TheFacebook’s bank account after Zuckerberg and Parker didn’t acknowledge an email he sent that reiterated his total control of all business aspects for TheFacebook. Freezing the bank account endangered their ability to pay for servers and posed an existential risk to the company if the site went down. Mark scrambled to scrounge some funds together, including money his parents gave him that was originally intended to pay for his college tuition, and rushed to raise an angel round, something he was previously happy to hold off on.

819 La Jennifer Way was the home Zuckerberg and his early engineers slept and worked at until getting kicked out for noise and damage complaints. Sean Parker would happen to be staying at his girlfriend’s home nearby, which would lead to him later becoming TheFacebook’s first President. One of his first suggestions was drop the “The” from TheFacebook.

Sean Parker arranged for Zuckerberg to meet Peter Thiel (going through Reid Hoffman), in what would go down as one of the most legendary venture capital investments of all time. Peter Thiel would invest $500,000 in an angel round for what would eventually turn into a ~10% stake (it was technically structured as a convertible note contingent on reaching 1.5mn users). As part of the investment, they would reincorporate as a Delaware company with the ability to award shares according to each person’s work contribution. Eduardo Saverin would naively sign the papers thinking it was just a formality, but it would end up allowing them to dilute him down to less than a 1% stake. (He would sue and ultimately settle for a stake that was much lower than his 30%, but still worth >$2bn at the IPO).

A Facebook profile before they dropped the “the”.

Zuckerberg’s messages later showed a ruthless side, saying that “I maintain that he f***ed himself. First by completing none of his three assigned tasks. He was supposed to set up the company, get funding and make a business model. He failed at all three and took the offensive against me with no leverage. That just means he’s dumb”. (The offensive Mark is referring to is the strongly worded email coming from New York and freezing the corporate bank account. While Saverin did set up a Florida LLC, it lacked contracts, governing documents, and didn’t specify the IP it owed. It didn’t help that while they all contemplated dropping out, Saverin was clear he would return to Harvard to finish his schoolwork and was openly contemplating graduate school. Saverin felt that they were living off of his largess while being the only one concerned about actually getting advertisers, which was largely true—profit was far from a concern of Zuckerberg’s at this stage of the company. In contrast, Zuckerberg fought to give Dustin Moskovitz, who they called “the ox”, for how much work he did for TheFacebook, a 5% stake.

Peter Thiel’s $500k investment returned over 1,000x, making it one the most lucrative VC bets of all time.

Scaling Up.

The most important thing Thiel’s investment enabled was more servers. Facebook, as it was now called after Sean Parker’s suggestion to drop the “The”, was in a continual battle to ensure that their server capacity could keep up with their usage growth. Zuckerberg was obsessed with how Facebook worked technically, and they even added timers to every Facebook page in testing to ensure each proposed feature wouldn’t slow the page load time. By October 2004 they had 500,000 users, and before the end of the year, they hit 1mn. But Friendster still had over 15mn registered accounts and MySpace was growing quickly too, having just hit 5mn users. In contrast to Facebook’s controlled growth though, these platforms were expanding beyond their technical capabilities.

Facebook kept their clean aesthetic. This had the byproduct of being easier to use and computationally less intensive.

Far from a trivial factor, page load time was hindering Friendster and, retrospectively, was one of the larger factors that led to their eventual irrelevance. Users were complaining of page load times of up to 40 seconds, with 20 seconds being not unusual. Such a poor user experience was driving users away.

Zuckerberg was far more methodical than his social networking peers, controlling Facebook’s growth until their servers were ready to handle the demand. This was partly thanks to their model of launching school by school, which allowed them to temper usage in line with server capacity. Zuckerberg was keenly aware the virtues of pacing growth: poor user experience could permanently churn users away, whereas a slower roll out seemed to only be increasing demand (it was common for students to write Facebook begging for them to roll out the service at their school). Mark would quip, “I need servers just as much as I need food… if we don’t have enough servers then the site is screwed.”

Despite Friendster and MySpace towering over Facebook in terms of users, Zuckerberg never seemed to worry about them. At the time, it wasn’t clear that the use cases were really that similar: Facebook was exclusively real IDs with your real friend network. MySpace was mostly pseudonymous with most friends met online. Friendster was conceived of more as a dating site, but one where you could date more naturally as you did in real life: by meeting friends or acquaintances of others. Knowing a mutual friend provided social proof as to the mate-worthiness of a potential partner; Friendster wanted to bring this aspect online without making it explicit–something that could seem tasteless to those new to online dating (versus Match.com where strangers would meet online with the explicit intention of dating).

Whatever Friendster’s initial intention, it was wildly successful and grew to over 3mn users just months after launching (faster than Facebook). But this torrent of growth had swamped their limited systems. They had to find a new Head of Engineering, and users had to endure a site that was crippling under the load of usage while simultaneously delaying all new features and site improvements.

Friendster’s second issue was that they wanted the site to be used differently than users wanted to use it. There was a small group of accounts that were growing in popularity on the site, essentially becoming the first wave of “influencers”. However, many of them did not use their real names, which was in violation of Friendster’s operating ethos. A prominent one, Tila Tequila, was booted from Friendster 5 separate times, and each time had to rebuild her massive friend base. Rather than start over a 6^th time, she joined MySpace, giving the Friendster copy-cat some traction when she emailed her prior 40k “friends”.

MySpace, which started when ReponseBase (a subsidiary of the disjointed internet holding company eUniverse/Intermix Media), was under sales pressure from their array of questionable merchandise and borderline spyware software (one scheme allowed users to download American flag cursors for free while discretely downloading a program that spammed them with pop-up ads). Chris DeWolfe and Tom Anderson would often look at Friendster with envy, noting that it was Friendster’s users that emailed new prospects, which resulted in a very high conversion rate (in contrast to ResponseBase’s email list of tens of millions that they would have to regularly spam just to get a few sales). Anderson and DeWolfe wanted to just make another Friendster, but one with fewer restrictions on users. It was hastily built and swiftly released. Only later would they notice that the site allowed users to post HTML code directly to the web page, which allowed for virtually unlimited customization; users seem to like it, so they just left it.

MySpace’s highly customizable pages were very popular as people used the wallpaper to express themselves.

In contrast to Friendster, they welcomed pseudonyms and did not restrict a user’s friend count. Whereas users needed to be invited to sign up to Friendster, anyone could join MySpace. The customization, coupled with broad pseudonym usage and open registration made MySpace a sort of “Wild West” of the social networks, where anything went (it was most closely associated with safety issues early on). Similar to Friendster though, their growth would stress the platform. But they dealt with it differently. When the MySpace team noticed that a popular “degrees of separation” function was very compute intensive (it told you how many degrees of separation were between you and another user), they scrapped it. Figuring out who could see whose profile was also compute intensive, so they opened MySpace profiles up to anyone. Rather than recode the site’s workings though, they hacked it by simply adding “Tom” to everyone’s profile as their first friend. This linked everyone. They essentially scraped the idea of a social network for expediency.

MySpace’s “Tom” was cofounder Tom Anderson. They made him a de facto friend of everyone to scrap the social network concept and open up the site. It also had the benefit of being less compute intensive.

MySpace continued to grow rapidly and surpassed Friendster in short order. Friendster’s issues showed up in their usage figures. MySpace would usually tout active users whereas Friendster quoted registered accounts. Myspace had roughly half of all registered accounts still active, while Friendster was closer to 1/3^rd (exact figures not available). MySpace was simply more addictive, and the fewer site issues led to less users fleeing.

MySpace was aware of Facebook and had even attempted to buy them several times (Zuckerberg would receive a lot of offers to sell out and would float what he thought was an exorbitant valuation, but it seems unlikely that he would have ever actually gone through with selling). In a meeting between Zuckerberg and the MySpace founders, the MySpace founders vented about how hard it was to keep up with all the site usage and asked if Zuckerberg had the same issue. His response: “that’s the difference between a Silicon Valley Start-up and one in Los Angeles”. The quip had more than a vein of truth. MySpace would leverage the entertainment scene in LA, going to night clubs and focusing on getting bands to sign-up (they called themselves the MTV of the internet), but they always lacked technical prowess. Not only was the site slow, but they struggled to let users add more photos, figure out a working search function, or how to load videos quicker. These issues were further compounded as Zuckerberg innovated the social networking format, increasing the table stakes.

While Facebook was behind in users, the users they had were much more active than even MySpace’s. They had 65%+ of users on their site daily (some figures put that figure closer to 80% early on), and this was in an era where you had to log in through slow desktop computers.

Facebook’s differentiation of only friending people you already knew to recreate your “social graph”, or real-life social network, online was proving successful. Targeting people in college, who tend to be the most social they will be in their life, coupled with the Harvard cache and real IDs, as well as privacy from non-friends, made Facebook more trusted than any other social network. This was a critical aspect in convincing people to upload and post personal information that was never typically shared online. New features like tagging people in photos only made it all the more addictive. Facebook would also allow unlimited photo uploads versus MySpace’s capped amount. To fuel their continued school launches, they would eventually accept a $12.7mn investment from Accel Partners at a $98mn post-money valuation; what would go down in VC history as another one of the all-time greatest venture bets.

MySpace’s messy, customized profiles presented a stark contrast to Facebook’s clean and uniform profiles.

By June of 2005, they would hit 3mn users. And by October, 5mn. That was 10x their user base just a year ago and almost 1/3^rd of all US college students. As Facebook would note, fighting the Friendster curse of parabolic growth degrading user experience was a constant obsession. They would spend over $4mn just on servers and networking equipment to keep the site up, an investment and priority that other competitors fumbled. MySpace, housed within a larger internet conglomerate, was constantly fighting for more server space and often resorted to stealing it from other divisions. Even when purchase orders were approved, they came in too slowly. MySpace was being crippled by the bureaucracy of their parent company (which was only exacerbated when Fox bought them), but they were still more than 4x bigger than Facebook.

A Facebook Profile. The theory was uniformity would help promote sharing as there wasn’t a lot else to do. The simplicity of Facebook and all of the features they wouldn’t add were important to keep users focused on posting and sharing.

2006 would be a pivotal year for Facebook. After growing at a slow and calculated pace, they would open up the flood gates with “open registration”. No longer would users be required to have a .edu email at a specific school in order to get onto the site. They first started opening up to high school students, but many high school students were already on MySpace (and it required a friend request from an existing user) which led to a disappointing uptake compared to the virtual immediate acceptance when Facebook rolled out at colleges. The relatively slower response from high schoolers led to some fears that growth may have peaked, and that Facebook would never successfully expand beyond college. It was in this backdrop that Zuckerberg considered a buyout offer from Yahoo for $1bn. While he could easily reject prior offers, he was mulling this one over. However, when Yahoo’s stock took a 20% hit after announcing earnings and insisted their offer be lowered commensurately, it was easy to walk away.

A Wall Street Journal article from 2006 comments on the Yahoo acquisition attempt.

Zuckerberg had three initiatives to stoke growth: the first was an email address importer that took your email contacts and told you who wasn’t on Facebook so you could easily invite them. The second was “Newsfeed”. From watching users’ site logs, they observed that users were constantly checking all of their friends’ profiles to see what was new and if they had changed their profile pictures. Zuckerberg’s idea was to create a “newsfeed” that showed users all of these updates in a stream, ranked for what each individual would find most interesting. No longer would users have to wade through dozens of profiles to find out what’s new. There was internal backlash from the business team, who was fearful that fewer clicks would mean less opportunities to show ads. Of course, Zuckerberg won.

When Newsfeed first launched, it was immediately met with furor. Users were creating Facebook groups to express their displeasure with the change: “Students Against Facebook Newsfeed” garnered over 700,000 members (and that was just one of several protest groups). Only one in a hundred messages about Newsfeed was positive. (The irony was that the newsfeed worked so well that people found out about anti-newsfeed groups through the newsfeed).  Zuckerberg insisted that people would get used to it and eventually adopt it. He was right—Newsfeed became the main attraction to Facebook in short order. He would draw on this experience in the future to support site changes despite wide user protest. Later, he would find out that this was an insufficient model to apply to all site changes, and his persistence in keeping universally unpopular changes came off as arrogant. (Newsfeed further separated Facebook from MySpace, whose tech debt was far too great and resources far too few for them to even contemplate copying the feature).

Newsfeed was launched in 2006. Many vocal users felt it was very invasive by publicly broadcasting all of their movements.

Facebook’s third initiative would be equally controversial: no longer would they restrict usage to students. With “Open Registration”, as they were calling it, anyone could now join. At the time, it wasn’t clear if this would be what made Facebook a national (or global) phenomenon, or if it would mark the beginning of the end as they lost differentiation and the necessary user trust that enabled so much online sharing in the first place.

It worked though; aided by the launch of Open Registration, they ended the year with 12mn users (Myspace still towered over them with over 100mn). Still, there were signs that Facebook users had a stronger attachment to the site than MySpace users since a higher portion of Facebook’s users signed on every day. Friendster was no longer part of the conversation, as a poor user experience and a stalling site fed up too many visitors. The same network effects that allowed it to grow so rapidly would ultimately be what lead to it unwinding (it was getting a second life in some parts of Southeast Asia though).

Zuckerberg presenting at F8, Facebook’s annual conference. The first F8 was in 2007.

The next move would allow Facebook to pull further away from MySpace by adding 3^rd party developers. MySpace had done this earlier, but they never developed 3^rd party support and actually undermined their developers. Oddly, MySpace rolled out apps that clearly competed with their top 3^rd party apps, even though there was no apparent reason to do so as they didn’t monetize well and MySpace was very strained for engineers (perhaps viewing themselves as a “media” company meant they wanted to control all content).

Facebook would instead give 3^rd party apps ample support, a guide on how to make money, and near full access to users’ data (which in hindsight was a grave error). At the time, though, that meant a lot of developers flocked to build Facebook apps on their platform. The recently launched Newsfeed was an incredible customer acquisition engine for these developers, as every user who used these apps would have their activity broadcasted to their friend list. Facebook was the gatekeeper, kicking off any app that violated their terms. There would be over 25,000 apps in less than 6 months. They called this initiative “Facebook Platform”, a designation they would cherish while it lasted.

Zynga created FarmVille, a casual game with social features that quickly went viral on Facebook. Also note the ads on the right side of the page; these desktop ads were an important source of revenue for Facebook in the early days.

Expanding Beyond a Website.

But as 2007 progressed, Facebook started to tarnish their reputation. The majority of their revenues were coming from a mix of basic desktop display ads and partnerships with select sponsors to promote groups (Viacom would pay for how many users joined a SpongeBob group to promote the SpongeBob movie, for instance). By and large, Zuckerberg was insistent on keeping the experience uncluttered with excessive ads and would turn down executives’ requests to display ad banners on the top of the site, in friend requests, or in other invasive places. He envisioned a more social form of advertising, one that allowed your friends to advertise a product instead of a celebrity. “Beacon” was born.

Beacon can be thought of as an extension of Zuckerberg’s idea of a single, unified ID across the web. When you visit any website, they could now build in social features that would allow you to relay that back to Facebook. If you liked a food recipe, bought movie tickets, or signed up for the New York Times, that activity would be sent to Facebook for your friends to know about. Zuckerberg was an adamant believer in “radical transparency”, and Beacon was nothing less than creating a more transparent world by connecting the people they loved to their everyday activities. Or so the party line went…

The 3 lines of Beacon code could be used by 3^rd party websites to relay their user’s onsite activity back to Facebook. Whenever a customer completed a purchase for instance, it would send the information to Facebook, who would then post it to that user’s Newsfeed, letting every friend they had know about their purchase. It defaulted everyone as “opted-in”, so millions of people awoke one day to see their off-site activities published on Facebook.

Beacon would relay off-site activity back to Facebook. The example above shows relatively anodyne activities, but there were many other web activities people would want privacy for.

While the idea may have been to spawn more “effortless” sharing with the people in your life, in reality it was interpreted as ill-conceived and a clear privacy violation. In one of the more prominent cases that became emblematic of the issues, a man bought an engagement ring online to propose to his girlfriend. Facebook sent that purchase data (including a photo and the price) to every one of their friends before the man even had a chance to propose. There was a quick backlash over the privacy violations, but Zuckerberg insisted (just like with the Newsfeed debacle), everyone would eventually learn to adapt. Facebook later rolled out a half apology thinking this was still the future, but soon after totally capitulated. This was the first real instance that users had lasting issues with Facebook’s conduct, and their trust of the company suffered as a result (in the early days, they ranked very highly in consumer trust). Nevertheless, they would exceed 50mn users by the end of the year.

A user is prompted to publish their food review back to Facebook as part of the Facebook Connect initiative.

In 2008, Facebook rolled out Facebook Connect, which was a critical step towards Zuckerberg’s dream of a “social web”. Zuckerberg envisioned every website being able to feel personalized to you with Facebook as a sort of connective tissue that lets you bring your friends and share your activity back to your newsfeed. (This was similar to Beacon, with the exception of a critical difference: Beacon would send your activity unprompted from a 3^rd party site). For instance, when you wrote a review on Yelp, you had the option to cross-post it to Facebook so your friends could see what you thought about a restaurant.  If you commented on a news article, your friends could see that and join the conversation or “like” your comments. This would make websites more interactive. Facebook Connect would allow users to log into 3^rd party websites with their Facebook credentials. And later, with the Open Stream API, publishers could integrate their Facebook posts, so they also showed up on their own websites as well.

A story in the feed that was published on Yelp originally.

Facebook Connect was a popular tool for developers and reduced friction to getting users signed up. Millions of sites would add Facebook Connect logins, but the “social web” never really came to be. Despite the ubiquity of the function, most sites had little activity to share. Nevertheless, it was still highly successful because every site Facebook connected to would position it to receive more data on their users’ offsite activity to be able to better target ads.

In 2008, Sheryl Sandberg joined Facebook. She was critical to building up Facebook’s advertising capabilities. Despite their highly engaged user base, they were only making a few dollars per user annually, or about ~$150mn in total in 2007. Sandberg’s strategy was to focus their advertising efforts on creating demand for products by matching ads to users who could potentially be interested in an advertiser’s product. This is in contrast to how Google advertises; because Google users type in what they want in the search box, Google doesn’t have the same sort of guesswork. Facebook, with all of their information on their users’ interests, would be in a unique position to provide this type of “discovery” advertising.

Example of some desktop ads. Their desktop ad business originally relied on more brand ads before they made it a big focus to personalize ads.

Other digital advertisers would try to do the same thing of course (show ads that the viewer would want), but their lack of data and logged-in user profiles usually made it a total shot in the dark, despite way fewer privacy restrictions. The other issue other digital advertisers would have though, was a lack of advertisers’ ads to pick from to make sure the ad shown was always relevant.

The key to kickstarting the process was convincing advertisers that Facebook was generating incremental purchases that wouldn’t have occurred otherwise. To solve this, Facebook would create one of the most advanced AdTech systems over the ensuing years, allowing it to not only serve the most relevant ads to a user, but also track when they converted, which helped advertisers build the conviction needed to spend more heavily on the platform.

Sheryl Sandberg joined in 2008.

Pages was the first feature to showcase Facebook’s broad and unique advertising potential. Businesses could create a “Page” for free on Facebook that looked similar to a normal user’s profile. Once built, the business could post on the page, just like any other Facebook user, and they had the ability to promote the post. The fact that most business owners were familiar with how Facebook worked, since they usually had their own personal Facebook profiles, made signing up for a business page a small task. Creating a business page was effectively the beginning of the onboarding process for a business to advertise. The business didn’t mind doing it because it was easy, free, and could help with consumer outreach. Once the page was built, it was easy enough to upsell them on promoting posts for as little as $5, and once they saw that it worked, they would be open to trying other advertising products.

All of these products are “self-serve”, which means they are initiated by a user without the need to interact with any other person. Facebook pages and the easy to use, self-serve nature of the ad platform would allow their advertisers to grow commensurately with their rapid user growth. The large pool of advertisers is critical to make sure there are enough ads to show users and that each ad is relevant. (This is something other platforms struggled with, despite similar or larger user bases).

A Facebook business page from 2009.

In 2008, Myspace’s usage would reach its all-time peak at 115mn with Facebook matching that by year end. However, while Facebook was about to surpass Myspace as the largest social network, alarms went off inside of Facebook as growth was slowing and no one was sure why. Zuckerberg gave the “Growth Team” full latitude to do whatever was needed to pursue the singular goal of growth. The growth team rolled out a slew of changes, most of which only led to minor improvements, but cumulatively were meaningful. Search engine optimization, a “people you may know” feature, and email scraping were all a part of fueling growth. More meaningfully, they would change the Newsfeed algorithm to focus on engagement and timeliness.

Twitter, founded in 2006, started to rise in prominence.

With the popularity of Twitter quickly spreading, Facebook would try to acquire Twitter for $500mn. After Twitter declined their offer, Facebook acquired FriendFeed (an app that was built off of the newly released Facebook Connect/ Open Stream APIs) for $50mn and shifted their Newsfeed to update in real-time. They would also add an ability to “fan” pages and follow users you aren’t friends with to copy Twitter’s 1-way “follow” model. They would also move the comment box to the top of the newsfeed instead of nestled away in an individual’s profile page. These changes would stoke more user engagement and push up more viral content in Facebook’s algorithm’s rankings.  

Facebook newsfeed circa 2010.

The App Era.

In 2008, Apple launched the App store, allowing developers to build 3^rd party apps. While the recently launched iPhone was widely hailed as a transformational device, it’s impact on web-based applications like Facebook was initially muted. As iPhone (and later Android) phone sales grew, it became clear that an incredible amount of time would be spent on the mobile phone. Facebook was written in PHP, a language that allowed quick iterations and updated immediately when the user refreshed their browser. However, it was ill suited for a mobile experience. Having to manage a desktop, a mobile web experience, and mobile apps on iOS, Android, and Blackberry was going be a headache.

To solve for this, they decided to use HTML5, which allowed them to write code once and push it to all of these different places. This proved to be a subpar decision, as not writing native apps for Android and iOS would lead to a slower and clunkier program. They would ride out the consequences of their decision to use HTML5 instead of coding natively in iOS and Android for a couple of years, pushing out unremarkable mobile apps. While desktops were still the main means of accessing Facebook, this decision’s impact wasn’t immediately apparent. By the end of 2009 they would have 350mn users, but their lack of mobile app focus was one factor leaving room for competition.

The first two Instagram app logos from 2010.

The first social network was conceived of in 1997 by Andrew Weinreich, called Six Degrees. It was in many ways similar in function to Friendster or pre-Newsfeed Facebook, but it was too early. Not only was the internet too slow and expensive at that time, but there were no practical ways to upload images. Andrew Weinreich knew that photos would be key, though. Prior to the digital camera, few people had digital photos, so he even considered letting users mail in photos which he would then scan for them to be able to add to their profiles; he was before his time though.

Andrew Weinreich’s Six Degrees was the first social network, launched in 1997.

Facebook benefited greatly by launching their site right after digital cameras were cheap enough to gain wide prominence. The next phase in digital photos was the mobile phone. Not only was it a camera that everyone always had on their person, it could also upload the photos and share them in the same device. Since users would no longer need to plug their digital camera into a computer to import their photos, a new form of spontaneous visual sharing was now possible.

Taking advantage of these advents in form factor, with the camera and phone merged into one device, Instagram was founded as a photo sharing app in 2010 (after a brief time as Burbn, which was partly a Foursquare clone). Despite the mobile phone making taking photos much easier, they weren’t of great quality. Instagram launched with novel filters not just because they seemed artsy, but because they helped hide the low-quality pictures the early generation smart phones took. In just 10 weeks, Instagram reached 1mn users.

Instagram’s early app would have a similar color scheme to Facebook. They placed a big “camera” button right in the middle of the bottom ribbon.

Instagram was a native mobile-first app that allowed users to easily take photos, apply filters, and then share them. In contrast to Facebook, where users would upload large albums after an event, Instagram users could post a quick photo and then go back to their event. This is why they considered themselves early on as a “check-in photo app” rather than a photo sharing app; more like a photo-first Foursquare than a photo-centric social network.

Instagram filters helped the app gain adoption as users wanted that photo aesthetic. Above are four of their most popular original filters. 

Unlike Facebook, Instagram would allow pseudonyms, and people would follow strangers whose content they liked. The friending of strangers never got carried away though (like on Myspace where it became a sort of game), because users were conscious of who they followed since they would have to see their content in their Newsfeed. Also, following strangers wouldn’t lead to their follower count rising, so there was no incentive to follow to look more popular (which was common with Friending on Facebook). This meant following was more deliberate than what developed on Friendster, MySpace, or Facebook, which gave a user’s account more longevity. The byproduct of this was an app where people could do both their socializing as well as be entertained by their interests: a sort of mobile, photo-centric mesh of Twitter and Facebook.

Instagram founders Mike Krieger (left) and Kevin Systrom (right).

With Facebook’s own mobile strategy yielding mixed results as users complained of a slow app and watching the rapid rise of Instagram, Zuckerberg was prescient in seeing how quickly the newly launched app could become a real threat. Instagram was growing rapidly and reached over 25mn users by 2012. Despite Facebook now exceeding 900mn users, rather than wrestle with another social network for screen time, Zuckerberg offered an unprecedented $1bn for a phone app. On top of that, Instagram was less than 2 years old in 2012. What was widely criticized as an extravagant price at the time, in hindsight became perhaps one of the best acquisitions of all time.

Prior to the Instagram acquisition though, Facebook developed a separate mobile native app called “Camera”, which was a failed Instagram copy-cat that was pulled less than two years after launching. The failure of “Camera” was one of many separately launched Facebook apps that would disappear quickly into history.

Photos from Facebook Camera’s press release in May 2012.

While Zuckerberg now had Instagram to take advantage of mobile, their earlier decision to use HTML5 for Facebook was having dire consequences. Their core Facebook mobile app was slow and clunky when compared to apps natively produced on iOS or Android. It was becoming more obvious that mobile would be the future of Facebook with 54% of users accessing Facebook on a phone: they had no choice but to go through a painful re-platform. This decision was more than a mere technical matter though; it would fundamentally change how they shipped and launched product. On HTML5, engineers could quickly write and launch code to their millions of users, confident that any bugs could be swiftly fixed. They would no longer have that luxury as each version of the app would have to be approved by app stores, meaning a bug could sit in an app for weeks pending approval. Their prior motto’s of “Move fast, break things” and “done is better than perfect” would have to conform to this new reality.

The Facebook ethos was to “move fast and break things”.

As they worked on rewriting their apps, Facebook’s ads business was exploding under Sheryl Sandberg. With more advertiser measurement tools and tracking, Sandberg touted that Facebook “advertising is some of the most effective”. The numbers were starting to agree with her: Revenues increased from $777mn in 2009 (+185% y/y) to ~$2bn in 2010 (+154% y/y) and $3.7bn in 2011 (+88% y/y). In 2009, they showed a $262mn profit, a figure that increased to $1.75bn by 2011. They were enjoying massive operating leverage over the past few years, getting them to a 47% operating margin for 2011.

At this time, it had been over 7 years since Facebook received its first investment from Thiel, and investors had started to push for an exit. With over 3,500 employees, massive revenue growth, and GAAP profitability, Zuckerberg finally capitulated. In May of 2012, Facebook would go public in what was one of the biggest IPOs of all time. With investors clamoring for shares, the bankers raised the price multiple times. The $38 IPO price put Facebook’s valuation at $104bn, selling $16bn of shares. Zuckerberg retained 22% ownership with 57% of the voting rights.

Facebook IPOs on the Nasdaq, May 18^th, 2022.

The zeal for Facebook shares turned out to be short-lived though, and the stock fizzled in the following months, with shares coming to trade below $20. While this could be due in part to investment bankers being too aggressive with pricing, there were also investor concerns fomenting that the shift to mobile would be a net negative for Facebook. Below is an image that shows several display ads on the right. Facebook’s ever growing ad revenue was driven by impression growth, which was supported by the 5+ ads they’d placed on each page. Despite MAUs (monthly active users) surpassing 1bn for the first time, investors were concerned that the shift to mobile would eliminate this space to show ads and overall impression growth would shrink as mobile usage increased. These concerns were showing up in the numbers with growth decelerating quicker than anticipated: in 2012, revenue growth slowed 51 points to +37% y/y from +88% y/y in 2011.

Facebook desktop with a column of ads on the left. They would also start introducing ads to the newsfeed.

Their solution to monetize mobile was simple, but contentious: insert ads into the newsfeed. In order to keep them from feeling intrusive, they would personalize the ads to each individual by drawing on their impressive collection of data on each individual. The image below, from Facebook’s S-1 filing, shows how they use the information a user shares in order to tailor ads to them. This data was further augmented by Facebook Connect, where advertisers would send them their data from their websites (where they are the publisher) to Facebook. Since these ads were better targeted to the individual, it meant that it was often something the user liked seeing, which also led to better clickthrough rates (% of people who click on an ad). This helped support Sheryl Sandberg’s argument that Facebook ads were some of the best around. After introducing ads to the Newsfeed, growth reaccelerated 18 points to +55% y/y. They closed 2013 with over $7.8bn in revenue as almost 75% of DAUs (daily active users) accessed Facebook through the mobile app.

Image from Facebook’s S-1 detailing how they planned to tailor ads to users.

Facebook would lean on international growth to add hundreds of millions of more users. They would crowdsource the task of translating the site to volunteers, which had the benefit of being quick and cheap. More importantly though, each localized facsimile of Facebook became embedded in regional cultures much better than if they had hired a group of linguists to translate the site into different languages.

But despite such growth, Zuckerberg was concerned about threats to Facebook. While there had never been a social network that had reached the same scale as them, he was constantly paranoid about Facebook’s growth reversing. His biggest fear was the explosion of mobile messaging. With people regularly talking to their friends through messages, Zuckerberg was concerned that this would be a breeding ground for a future social network to bootstrap a social graph from.

In 2013, Facebook reportedly offered $3bn to buy Evan Spiegel’s Snapchat, a camera-first messaging app with disappearing messages. Snapchat was growing quickly and had a similar number of users as Instagram did when they bought them (>25mn). The app was being quickly adopted by teens at a time they were increasingly turning away from Facebook.

The Snapchat app opened to a camera (left) and then allowed you to click who you wanted to send the image to. The image then is auto-deleted seconds after you view it.

The popularity of Snapchat was showing that Zuckerberg’s vision for a more public internet with a single channel for wide-spread sharing could be wrong. Snapchat was a “private” social network where users would send images directly to a select few individuals and there was no way to comment, like, or reshare. Snapchat would launch a feature called “Stories”, which allowed users to post images or videos that would automatically be deleted after 24 hours. Only their friends could see these stories, and again, there was no way to comment or like them, removing peer judgement from the equation. Less than a year later, ~40% of all 18-year-olds in the US were on Snapchat.

What Stories capitalized on was that people felt more comfortable posting everyday content on an app where everything is ephemeral. Compared to Instagram, which had a large celebrity and influencer culture, users felt compelled to only post the glossiest and most aesthetic photos they had to showcase their life. An Instagram post was something that was carefully planned and curated, whereas users would throw open their Snapchat app and take a short video to post to stories without thinking much about it—it would be gone in a day anyway, and even though your friends saw it, it never “felt” public.

Original Snapchat stories on the left, which was later updated to a carousel users could scroll through (top of the right photo).

Facebook would try to copy Snapchat multiple times with multiple standalone apps. First with Poke in 2012, then with Slingshot in 2014, Lifestage in 2016, and then again with Instagram Direct in 2017. These copy-cat apps were just a few of the dozens of standalone apps they would produce throughout the years. None of them would ever achieve any significant success though, and all were ultimately killed off. 

Poke was launched in December 2012 to fend off Snapchat.

While Facebook hasn’t had much luck in launching standalone apps, they have found significant success in copying popular features. Snapchat’s Stories feature, in particular, was so popular that Facebook was concerned that it would suck up a lot of content from the Instagram ecosystem (whose younger demographic was closer to Snapchat users). After several failed stand-alone app attempts, the solution Facebook arrived at was simple: launch the same exact feature in Instagram and call it the same thing so there was no confusion about what it was. Kevin Systrom would plainly acknowledge that Snapchat came up with the idea, but nevertheless their superior network helped Instagram Stories become an instant sensation and stymied Snapchat’s growth in the process.

However, Stories in Instagram rolled out in August 2016, almost 3 years after Snapchat originally launched the feature. While Instagram was able to recover lost time as Stories became exceptionally popular, in the future, squandering time with stand-alone apps before bringing the core feature into their core apps would cost them. More on this later.

Instagram Stories from a 2016 blog post.

By the end of 2016, Facebook’s original site (referred to as Facebook Blue) and Instagram were clearly the top global social networks. Facebook Blue had ~1.9bn monthly users and Instagram had an estimated ~500mn, whereas Snapchat had 158mn and Twitter had 319mn. Facebook also owned WhatsApp, one of the other largest platforms, with ~1bn users by 2016, thanks to Zuckerberg’s long-term orientation and readiness to greatly outbid competitors.

Early WhatsApp UI.

Going back to 2009, when telecom services would charge users per SMS text sent, WhatsApp was founded as a cheap and easy way to message your friends. Requiring just a phone number to log in, messages were sent through the internet, circumventing the telecoms’ fees. It would reach 100mn users by 2012 and 500mn by 2014. Zuckerberg, conscious of the large time spent and engagement WhatsApp was garnering, became paranoid that that this app would begin to develop more social networking functionalities. He bought the app for $19bn in 2014, beating out Tencent and Google who also were courting WhatsApp founders Jan Koum and Brian Acton. Given what Tencent has been able to do with Weixin (WeChat), Zuckerberg’s fears were well founded.

The second move Facebook would make with messages was to split Facebook Messenger into a stand-alone app called “Messenger”. Since being crammed into the Facebook Blue app felt cramped and slowed down the app, separating them would allow the Messenger app to function quicker with a cleaner feel. 

Facebook Messenger app launched as a stand-alone app in 2014.

Messenger usage exceeded 500mn in 2014, and would be the first (and so far, only) time they created a successful stand-alone app. (They would also try to separate out a Facebook function into a separate app with Facebook Groups, Facebook Events, a notification app called Notify, a private photo sharing app called Moments, and an app for public figures to talk to their fans called Facebook Mentions, but ultimately kill them all).

Five of Facebook’s attempts to create standalone apps were unceremoniously squashed.

While Zuckerberg was protecting against the risk of a new mobile-born social network, he was becoming increasingly aware of how painful it was becoming to lose “platform” status. Their 3^rd party developer business (users spending on games, virtual items, and other digital goods) made up 16% of revenues in 2012 and at one point was a business they thought could make up 30% of the business. Instead, it dwindled down to ~3% by 2016. It wasn’t just the rapid growth of their advertising revenues that dwindled: Facebook would earn less from their “Payments and other fees” in absolute in 2016 than 4 years earlier. The main culprit: mobile phones and the app store.

Facebook Credits were used mostly for virtual purchases in games to reduce purchase friction and so you didn’t have to share credit card info to 3^rd parties. Released in 2011 and taking a 30% fee, they were discontinued in 2013 because of low adoption and the shift to mobile.

Facebook 2.0.

No longer did developers clamor to be on Facebook’s desktop platform, they shifted their efforts to building native Android and iOS mobile apps. As a consolation for Facebook, these apps would need their best-in-class ad targeting and large audience to find users, leading to app downloads being one of their largest advertising verticals. But Zuckerberg did not like being just another “app” on iOS or Android. Feeling that they missed the opportunity to shape the mobile world (they would briefly explore creating a Facebook phone, which was rolled back to just a Facebook skinned Android dubbed “Home”), Zuckerberg did not want Facebook to miss the chance to become the next platform.

Facebook bought Oculus, an early virtual reality handset maker, with just 75,000 orders for their Rift headset at the time, for $2bn. This began an ambitious and expensive bet to regain their platform status in what Facebook is hoping is the next computing transition. More on this later.

Oculus founder Lucky Palmer with an early Oculus device.

Fast forwarding to 2017, Facebook’s virtual reality initiatives were still nascent, but their Family of Apps was as big as they’ve ever been, and their ad machine was printing money. From just 2015 to 2017, advertising revenue increased from $17bn to $40bn. In just a little over a decade since being founded, they were generating ~$15bn of free cash flow annually. Their core business would prove to be very resilient, lasting multiple scandals without a blip, but users’ trust in Facebook was being rocked.

While data and privacy scandals have been a part of Facebook since its early founding with the Beacon snafu in 2007, the Terms of Service changes in 2009 (claimed ownership over user data), and the 2012 FTC Consent Decree about how they mislead consumers on how private their data was, all of that was relatively quaint compared to the accusation that they allowed over 80mn users’ data to be improperly gathered by the consulting firm Cambridge Analytica to unscrupulously target users with divisive messages. Even worse for Facebook was they were blamed for Donald Trump’s presidential victory. This was the beginning of a flood of complaints about fake news, misleading political ads, and other harmful content.

Mark Zuckerberg testifies before the Senate to answer for the Cambridge Analytica Scandal.

Facebook would have to scramble to grow their moderation team, which ballooned to 10,000 in 2018 with further plans for future increases. They would then invest billions of dollars into artificial intelligence and machine learning systems to be able to moderate content before it was even posted. This is an on-going effort, and despite drastic improvement and sophistication in their ability to block content before anyone ever sees it, it is a politically fraught area with many discordant opinions on how Facebook should proceed. Much of the legislation that governs internet platforms hasn’t been updated since the Communication Decency Act of 1996, which included Section 230 that limits internet companies’ liability for something a user publishes.

Whereas prior earnings calls and presentations would focus on the virtues of connecting people, the conversation shifted to how to mitigate against the worst that can happen when billions of people are connected. Uprisings against authoritarian regimes, humanitarian movements, as well as bloody genocides and election manipulations could all be attributed to Facebook with a vein of truth.

In 2019, Facebook accepted a $5bn fine for violating their 2012 FTC decree, the largest fine the FTC has ever imposed. With $70bn of revenues and almost $30bn of EBIT, the fine was financially tenable. But the negative sentiment that ensued amongst many consumers with Facebook (particularly in the US) would hinder their ability to hire and retain some engineers, as well as launch new products.

Facebook’s first hardware foray with the Portal device. It’s video quality and UI was liked by reviewers, but most were unwilling to let Facebook put a camera in their house after the privacy scandals.

Observing the rise of home devices, Facebook launched the Facebook Portal for consumers to easily video chat with their Facebook friends or WhatsApp contacts. The device got great reviews, with a sleek look and clean UI, but the simple fact that Facebook made it would preclude most customers from even considering it. In online reviews, it was clear that Facebook had lost consumers’ trust. However, fortunately—and surprisingly—for Facebook, the scandals haven’t materially weighed on usage stats. But they would rename to Meta, an ode to their metaverse ambitions (detailed later), and a needed rebrand to launch new products.

Facebook’s Family of Apps (the one on the far right is Oculus).

Facebook Blue still has over 2.9bn monthly users with 67% of them returning to the site every day. Their Family of Apps has 3.7bn unique users with many of them utilizing multiple apps. FY 2021, they generated $118bn in revenues with ~$40bn of net income. Their average revenue per user globally was ~$41, a figure that increases to ~$213 when looking at just North America. This is a far cry from the couple of dollars per user the most optimistic VCs would assume Facebook could monetize in their early days. But despite such monstrous financial success, Meta today is facing their largest threats ever from the rise of a vociferous competitor (ByteDance’s TikTok), Apple crippling Facebook’s ad targeting mechanism with ATT, potential anti-trust actions, and global legislation. So far though, key metrics, usage and time spent, are steady—but Meta isn’t sitting still.

With majority control of the company, Zuckerberg isn’t just navigating the current turmoil, but is trying to set Meta up for the next decade plus. With his founder’s authority, he is driving several ambitious initiatives simultaneously, from a product transition that risks alienating loyal users to the largest investment any company has ever made in an unproven technology, all the while stitching up the core business and placating a myriad of legislative bodies and governments.

We will cover all of this in the report. First, we will now move to an overview of how the business works and pick back up on competitive factors, AdTech landscape changes, geopolitical risks, and the metaverse.

Business.

Meta Platforms is a holding company with two reporting segments: 1) Family of Apps, which includes Facebook, Messenger, Instagram, and WhatsApp, and 2) Reality Labs, which houses their efforts to build virtual and augmented reality hardware and software. As shown below, virtually all revenues currently come from their Family of Apps, and just ~2% coming from Reality Labs.

Over 99% of Meta’s Family of Apps revenues is generated through advertising, with the other ~0.5% (itemized as “Other Revenue”) coming from developers using Facebook’s payment infrastructure, which is mostly a vestige of the desktop, developer business. Below is Meta’s ad revenue since 2009, when they generated just ~$750mn. In the 12 years since, they have grown ad revenue at an incredible ~50% CAGR to reach $115bn. (You might notice that LTM revenues haven’t grown compared to 2021—much more on this later).

Meta’s advertising business is driven by essentially 4 variables: 1) number of users, 2) time spent per user, 3) ad load, or how many ads are shown per minute, and 4) price per ad. Now, there are many other factors that will impact these 4 variables, but Meta’s revenue is ultimately a simple output of multiplying these four variables.

1) The first variable in the equation, number of users, is self-explanatory. But keep in mind that Facebook reports both monthly users and daily users. It is typical to start this math with monthly users (MAUs) before adjusting it with a DAU/MAU ratio (this is done because more companies report MAU than DAU). For Facebook, their DAU to MAU ratio has averaged around ~67%, which means that on average, ~2/3rds of all reported monthly users use the site daily.

2) The second factor, time spent, refers to how much time the average user spends on the site or app. This is typically calculated as a daily average of annual time spent to adjust for potential differences in weekend versus weekday usage. For our purposes, when we say time spent, we will be referring to the total average time spent per user per day. It is also worth remembering that a single day’s “time spent” can consist of dozens of separate “sessions”. A session is each discrete time a user checks the app or site. Different apps will have different user behavior and will help inform how the time spent can be monetized. For example, if you are checking an app 30 times a day (normal for a messaging app) with a lot of short sessions, then it would be hard to show a video ad, but that behavior could support banner ads.

3) The third factor, ad load, is how many ads are shown to the user in a given amount of time or other unit, i.e., per post. Ad load has the least uniform definition of any of the variables. We could think about it in terms of how many ads are shown per post or how much time is spent on ads per minute. What definition we pick will vary on the product: for Newsfeed posts, it makes sense to think of it as ads per number of user posts, but for video products, it probably makes more sense to think of it as seconds of video ads per minute. We can compare ad load across platforms and product by turning it into a % (i.e., % of posts that are ads or % of time spent watching ads). As long as whatever metric picked comports with the ad pricing units, it doesn’t matter much.

Facebook mobile ad for an app download that shows up in a user’s newsfeed.

4) Ad pricing is the price that each individual ad gets. Every time a user is shown an ad, it is called an ad impression. Ads are often sold on a “price per thousand” basis, where the thousand refers to the number of impressions. This is referred to as a CPM or cost per thousand. (A $5 CPM means that the price to buy or sell 1,000 ads is $5). Now, while ads are often sold on a CPM basis, they could also be sold on a CPC, cost per click, basis (users clicking on the ad), or any other action they want to incentivize (called cost per action), but CPM and CPC are the most popular and easiest to talk about since it’s a uniform activity (cost per action could be to incentivize an email sign-up, download an e-book, follow the company of Facebook, etc.).

We can actually take the first three variables together to simplify our revenue model further. Below, we show how multiplying the # of users by time spent per user and ad load can get us ad impressions. Ad impressions x ad price gets us revenue.

We will go into the minutia of AdTech and all the factors that impact these revenue variables, but do not forget that ultimately the advertising revenue Meta generates is a very simple equation. The tricky part is that of all of the revenue variables, Meta only explicitly gives us users (they provide figures on y/y ad impressions and ad prices, but not an average “ad price” or total ad impressions). 

Since IPO, Meta has reported how many users they have on Facebook Blue, which they continue to do today. Below, we can see that Facebook Blue users have grown at an incredible pace, increasing at a 20% CAGR since 2009, yielding them 2.9bn Facebook users (Facebook Blue and Facebook Messenger users are counted here). 

However, as their other apps grew in size, they started reporting “daily active people” (DAP) and “monthly active people”, which include users from Instagram and WhatsApp. The shift to the “Family of Apps” metric is also to help emphasize that even if a person isn’t on Facebook, chances are they use one of their other apps.  Between all of their apps, they have ~3.7bn monthly users (or monthly active people as Meta refers to them) and ~2.9bn daily active users (or daily active people).

In the Competition section, we will talk more about how users use these apps, but below we list the high-level functionality of each. If you are familiar with each app, you can skip this next section.

Family of Apps.

Facebook. Below we show the Facebook mobile app, which is how >90% of people access the service. The app picture below is from the US version, but keep in mind that they tailor the app to different markets. They also have a “lite” version for markets that have a less developed communications network in order to make the app’s data requirements lighter.

Instagram. Instagram does not really have a direct comparison in social media, which is partly why it is so popular and unique. It is the app that created the idea that social media is fake and glossy because everyone who posts usually feels a need to present only the best in their life. Celebrities and influencers tend to value Instagram popularity more so than any other platform. This is not only because it’s the only scaled platform that allows public figures to post more casual moments, as well as more “formally” with feed posts, but also because of the artistic and aesthetic “vibe” it accrued early on.

Relative to Instagram, posting on Facebook feels “cheaper”, like it isn’t quite the right fit (it’d be similar to finding a Gucci bag at Target—something feels off about it). The only other place where public figures can reach a wide audience is Twitter, but that is text-based and posting photos or video is clunky. For most celebrities and influencers, Instagram is the only good place to host their public profiles because of the visually appealing way content appears on the site, but also critically because the “follow” model allows fans to easily stay connected. For these reasons, even if an influencer gets big on TikTok, they still want to grow their Instagram.  

The idea that Instagram posting should only be reserved for the most beautiful things in your life was what made the app unique for creators, but also what made it somewhat pernicious. Many users felt that the bar to post something was too high. Instagram solved this by launching Stories, which automatically disappear in 24 hours. This is the most used feature to create content because users feel they can be a little “realer” there.

Instagram also is not a “real name” social network like Facebook, so there are many pseudonymous accounts that enabled meme accounts, pseudonymous influencers, and “Finstas” (fake Instagram accounts that are usually used by users who want to create a 2^nd Instagram with just a smaller group of friends).

Recently, Reels (a TikTok clone) and Shops (shoppable posts, discovery, and in-app purchases) have been a focus for Instagram. Reels are short videos that are under 90 seconds (and usually 30), which originally were housed in a separate Reels tab. More recently, Meta is introducing Reels to the main Instagram feed though. In contrast to the way Instagram used to operate, where you only saw content from people you followed, Reels can be from any user. This is to copy short-form video app TikTok, which pushes the most compelling content personalized to each user, rather than the follow model where the user finds and curates what content they want to consume (much more on this later).

Messenger. The picture below is a more recent version of Messenger, but keep in mind that they change it often. They temporarily had a “Discover” section that allowed users to find and message businesses, but they ultimately removed it as a separate tab icon. More recently, they have added a separate “call” button so users can voice or video call friends (this existed in prior versions, but it wasn’t in the ribbon).

Generally speaking, the advantage of Messenger is that you can message your Facebook friends, many of whom are people whose phone numbers you do not have. Messenger also has more customization, GIFs, and stickers compared to other apps. In order to use Messenger though, users have to create a Facebook account first. They can continue to use Messenger after deactivating their Facebook account, but they have to have one to set it up in the first place. This is a meaningful difference versus WhatsApp, which just requires a phone number to sign up and use.

WhatsApp. WhatsApp is a messaging app. While the app started as a pure-play text messaging app, they have added a lot of features since, including voice and video calling, groups, businesses, and in some markets, Shopping (which has been a particular focus in India). WhatsApp has a reputation of being a privacy-centric app (in contrast to Meta’s other services) since messages are end-to-end encrypted. There are over 2bn WhatsApp users globally.

These four apps constitute Meta’s Family of Apps, and over 100% of operating profits (Reality Labs is operating at a large loss). We will go into more detail later with regards to Reality Labs, but at a high level, it is Meta’s virtual reality and augmented reality division that is trying to launch the next-generation computing platform. They currently sell the Quest virtual reality device, a pair of smart glasses (that can record video/take photos), and the rebranded Meta Portal. In total, this division generated ~$2.3bn in revenues for 2021 and a similar amount LTM.

However, the combined company has been under pressure with LTM (from 3Q22) revenues unchanged compared to 2021. Their stagnant revenue, coupled with growing Family of Apps expenses and losses at Reality Labs, has resulted in consolidated margins compressing ~10 points. Their core Family of Apps business is typically very profitable with ~50% EBIT margins, but over-hiring, server buildouts, and other investments in content moderation, AI, and Discovery have kept the Family of Apps expenses high. Combined with soft macro, privacy changes limiting data availability, and competition have resulted in Meta reporting their first ever quarter with negative revenue growth in 2022. This was happening concurrently with Meta increasing their Reality Labs expenses, leading to profits dropping $12bn LTM compared to 2021.  

We will pick back up on this later, but the high-level takeaway is that Meta’s core advertising business is typically highly profitable and could either be going through a transitory headwind or facing more permanent structural issues (we will explore this later on). We will also contextualize their Reality Labs investment and what the potential opportunity is. 

First though, we will move into the advertising industry and the competitive landscape.

Advertising Industry and Competitive Landscape.

Meta’s competitors come from two distinct vectors: 1) those that compete with them for users’ time spent, and 2) those that compete with them for advertising dollars. Even if Meta successfully captures half of the world’s attention with their apps, it would not be easy to monetize it if advertisers thought other ad venues were better. A CPM can range anywhere from ~$0.50 (five 1/100^ths of a cent per ad impression) to >$40 CPMs for the best ads (4 cents per ad impression). This wide range of pricing is based on (1) how effective the advertising is and (2) the extent to which the platform can prove their advertising is effective. Competence in both is essential to achieve the highest CPMs. In this section, we will dovetail into the advertising industry, saving questions of whether Meta will be able to continue to demand users’ attention as they currently do (time spent competition) for later.

Advent of Digital Ad Industry.

Before the internet, television, radio, billboards, sponsorships, newspapers, magazines, and other print media were the main channels an advertiser could pay to reach a customer (other channels like cold calling, door to door sales, and direct mail flyers did not require paying an intermediary for placement). A television show, magazine publication, or radio broadcast was tasked with creating enough compelling content to attract a large audience so that they can then sell a portion of that audience’s attention to a business. A business would happily pay to rent a consumer’s time without having to go through the hassle of trying to find potential customers in the first place. This is the foundation of advertising that continues to this day.

These are examples of lower quality ads. The left is the worst offender: early 2000s pop-up ads which had low click through rates and enraged customers. At the bottom you can see an ad for a surveillance camera, which could very well have been from MySpace’s parent company eUniverse. On the right is an invasive Target desktop display ad with another ad on the right of the content.

The internet has not only introduced new channels to advertise in, but also an entirely new capacity to test and track advertising. Before digital ads, there was little ability to know how ads were performing and which specific advertising campaign to attribute to an increase in sales. There is an old advertising industry saying that captures this sentiment: “I know half of my advertising budget is wasted, I just don’t know which half”. Digital ads allowed advertisers to know how many times people viewed an ad, clicked on it, and completed a purchase after clicking on it. If a business had dozens of different ads (as is typical) they could now know which one was most effective and which performed the best with different demographics. However, these capabilities wouldn’t come until much later.

The digital ad ecosystem was initially characterized by very traditional advertising practices that didn’t take advantage of all the new capabilities that digital ads allowed. A website owner would directly reach out to various businesses to try to make a direct deal by selling the advertiser a portion of the website to display ads on for a fixed price. Initially, every user would receive the same ad, making it no more than a sort of digitized billboard.

Matching ad space from publishers to potential businesses that wanted to buy the ad space was a highly manual and arduous process that also required haggling over pricing in an opaque market. DoubleClick (who would later be acquired by Google in 2007) was one of the first companies to address this. They would represent a group of publishers (websites) so a business could interact with just DoubleClick to buy ads on the various properties they represented. This “ad network” was an early predecessor to a slew of functions that would later take its place, each specializing in a more specific tasks, like serving publishers, advertisers, data management, ad auctions, verification, etc.

DoubleClick would first focus on publishers to become what became known as a Sell Side Platform or SSP. An SSP is a platform that aggregates publishers’ (website owners) available ad space to sell. This is similar to any other marketplace i.e., a Farmers’ Market, where different businesses group together so buyers can purchase from multiple vendors at a single venue. DoubleClick would take this growing base of ad space they represented and go business by business to get advertisers, in addition to plugging into the one of many other ad networks that were sprouting up. 

While this model worked tolerably with large advertisers and Fortune 500 companies, it would be impossible to reach smaller businesses with such a manual process. With the growth of internet usage, there was far more ad space available than there were digital advertisers. However, many other businesses would have been willing to advertise digitally if they only knew how.

Demand side platforms (DSPs) eventually emerged, which solved these issues. A DSP would aggregate the advertising demand (businesses that wanted to buy ad space) and let a single business interact with a number of different SSPs without having to coordinate launching separate campaigns on multiple different platforms. “Auctions” mediated the decision of whose ads were placed where. The formation of DSPs was coupled with the advent of ad exchanges, which auctioned off, in real time, each ad impression to the highest bidder (there were other styles of auctions too).

The DSPs would connect advertisers to SSPs through ad exchanges, which facilitated these real-time auctions, called Real-time Bidding. Now, every time a user loads a web page, there are millions of advertisers bidding against each other for each ad impression in real time. (Ever notice why the ads sometime load slower than the content on a webpage? That is because of the time it takes to conduct the live auction). All of this could only be done digitally and represented the biggest change to the advertising industry since the prominence of TV in the 1950s.

From DoubleClick’s promotional material. DoubleClick would eventually become a fully integrated AdTech provider with a DSP, SSP, and an Ad Exchange.

Now with auctions running for each ad placement, the question of price is determined by ad efficacy (how many people click the ad), how many advertisers are bidding for the same ad slot (supply and demand), and how valuable a click is to the advertiser (clicks to a high-end beauty site are worth more than clicks to an ad-supported celebrity gossip site). 

The digital ad market has cannibalized the traditional ad market while also materially grown it. To understand this, we need to talk about the two main types of ad categories: 1) brand advertising and 2) direct response advertising.

Brand advertising is when a business markets their brand or product generally with the intention of growing awareness or creating associations for their product in hopes that the next time the consumer is in the market to buy something like their product, they will choose their brand. Duracell advertises their batteries as the #1 most trusted battery brand with fire fighters. The battery maker’s intention is to both convey superior quality versus competitors and imprint the idea in a consumer that if they want a reliable battery, they will buy Duracell. Importantly though, this sort of advertising isn’t designed to get a consumer to buy a battery today, but rather when they are in the market for a battery to go for the Duracell. What they are doing is building their “top of funnel” awareness of the brand so they can hopefully convert a sale later on.

Brand advertising can also have a “demand-generation”-like aspect to it. Coca Cola doesn’t sell soda, they sell happiness. Axe isn’t deodorant, it’s a “lady attractor”. While this sort of advertising is designed to create associations with the companies’ products so they can stand out and differentiate themselves versus the competition, it also could potentially create the desire for someone to go out and buy a can of Coca Cola or a stick of Axe. However, what distinguishes this sort of brand advertising from direct response is that the latter has a call to action.

Coca Cola advertises happiness and Axe sells the ability for the average man to “get ladies”. These ads are basically aimed at making an otherwise commodity product something unique. If we think of the Consumer’s Hierarchy of Preferences (introduced here), the basic idea is that hitting a higher order desire for a consumer is more valued by a consumer. It is easy to make a comparable soda, but if you could convince your customer that what you are really selling is happiness, then your product is in a more unique and defensible position since it is serving more Consumer Preferences. The relative unpopularity of cheaper, but similar tasting RC Cola or other rip-off sodas supports the notion that these ineffable aspects can be of critical importance to a product. However, building these associations is a slow and long process with success being unattributable to any one single ad.

The nature of brand ads is that they require a consumer to be exposed to them many times, and it’s impossible to know if any single exposure meaningfully impacted their opinions. Thus, they are not well suited to take advantage of some of the new capabilities of digital advertising like tracking the success of a single ad impression. In contrast, direct advertising would become more effective than ever before.

Prior to the digital world, direct advertising was largely limited to calling 1-800 numbers shown on television ads or responding to direct mail flyers. While these tactics were successful enough to gain prominence, they were fraught with consumer friction and lacked the ability to personalize ortarget the ads. The internet would solve both of these problems.

A typical TV direct response ad.

With computers and mobile phones, a consumer can interact with an ad in the same venue they are consuming content. Bolstered with the targeting capabilities from browsing data that enables building a user profile, ads could now be tailored to an individual, increasing the probability it’s something they actually want. While there is certainly a creep factor of targeted advertisements, there is also a clear benefit: if you have to be flooded with ads, isn’t it better that it is something you might actually want? This combo of tailoring the ad to the individual, coupled with less friction to interacting with the ad, led to a proliferation of direct response ads, or advertisements that call on the user to buy something right that instance.

A person might be reading an article or scrolling through their Facebook feed and see an advertisement for sweatpants. And it just so happens that the user wanted sweatpants. They click the ad and it takes them to a landing page to purchase the pants right there. This is even better with something like PayPal or Apple Pay where the user already has preloaded payment and shipping info and can just click one button to check out.

No longer would a business have to blanket advertise in hopes of convincing a customer to buy their product the next time they were in a position to do so; now, businesses would send the ad only to current potential customers and get them to complete a sale immediately. In fact, this form of digital-enabled advertising is so effective that it allowed a whole new generation of small businesses to be built on the backs of it, and Facebook is by far the leader here.

Allbirds was one of the first “D2C” or direct to consumer brands, which were made possible by the likes of Facebook and Instagram since they would help the brand find their customers without needing to embark on the costly and slow path of brand building before making a sale. On the bottom of the ad, there is a “Shop Now” button.

Part of Zuckerberg’s reluctance to introduce advertising to Facebook was driven by his belief it would deteriorate user experience. However, this new form of advertising was far less invasive, and its success strongly suggests the consumers actually like it. Facebook, with their incredibly granular data on a person’s interests, likes, and friends, was in the best position to offer these personalized direct response (DR) ads.

Facebook would enable a whole generation of small businesses to not only start, but thrive, allowing them to advertise directly to consumers in a cheap and effective way. Whereas before a shoe brand might have to strike expensive sponsorships or run costly television ads, Allbirds was able to cheaply advertise through the internet. They were one of many of these “D2C”, or direct to consumer companies, who would largely find their customers through social media channels.

This Adidas ad plays a video as the user scrolls by and is paired with a scrollable carousel to browse inventory. A user could click on any image of the product to get more info.

Facebook supported the creation of D2C companies and the long tail of small businesses, growing their revenues alongside the growth of a whole new category of internet-enabled commerce. Not only are ads now tailored to the individual so businesses can actually find their potential customers, but the tracking of each ad and tying it to a sale allowed them to continue to reiterate their ad copy and advertising campaigns for maximum return on ad spend (ROAS).

Now, a business could spend $1 and see $5 of sales. With a 40% gross margin, that means for every $100 they spend, they generate $100 in contribution profits (profits after variable expenses and advertising spend). This means that the next time they go to advertise, they would have $200 to spend (they received the original $100 back and made an incremental $100). High returns on advertising spend could quickly create a virtuous cycle whereby the more you spend, the more you make, and the more you have to spend again.

Return on ad spend (ROAS) is the confluence of how good an advertiser’s copy is, how accurate the targeting is, how many advertisers are bidding on the ad impression, and of course, how compelling the product or service is. (ROAS is usually either calculated as revenue over ad spend for expediency, but the technically correct method is contribution margin over ad spend).

Certain verticals like beauty and mobile gaming would have the most competition, but the economics of those businesses would support high ad prices (CPMS). This is because the ad price a business is willing to pay is a byproduct of how much they make. Beauty and mobile gaming being high margin categories enable them to spend more on advertising. This pushes up their bids in auctions and ultimately leads to higher prices.

Facebook mobile app download ads. Mobile app downloads are one of Facebook’s biggest ad verticals.

Before moving into Facebook’s advertising apparatus, we want to first touch on the advertising TAM.

Advertising TAM.

Caveating that any TAM estimate is prone to error, the global advertising market is approximately ~$800bn and has grown around 4-5% annually since 2010. While the advertising industry is usually thought of as growing at the rate of GDP, this was about 1-2 points higher than global GDP growth since 2010. Supporting this higher-than-average advertising market growth was the adoption of digital ads, which created new opportunities to reach consumers and more than offset the declines in traditional advertising. Digital advertising didn’t just cannibalize the existing traditional ad market, it also grew it.

Digital advertising in 2010 was estimated to be just ~$25-30bn, but today represents over $500bn. While these figures imply that ~$150-200bn of “traditional” advertising has gone digital, the reality is murkier as what were considered “traditional” advertising channels like newspapers, magazines, and television are digitized. Either way, digital advertising represents about 60-70% of the total ad industry today.

Meta today represents about ~13% of the total advertising industry and ~20% of the digital ad market. (They are only superseded by Google, who is ~25% of the total advertising market and 40% of the digital ad market).

Digital advertising allows for new means for advertisers to find their customers. The two big categories of advertising are “discovery” and “intentional”. As mentioned in the intro, Google commands the intentional advertising category because users will search for exactly what they want and Google just has to surface the right result (Amazon is in a similar bucket, but they are “vertical” search, since you only search items on their site, unlike Google’s general search, which could include local services, travel, medical, etc.). In contrast, Meta has to figure out what their users are interested in and suggest items to them in hopes they purchase it. While both of these are new means of advertising, it is the “discovery” channel, with the customization that digital ads enable, that is truly new to advertising.

The nature of Google Search ads is such that they just have to surface links to what you told them you wanted.

In the past, a merchant would open a physical storefront and then use newspaper ads, direct mail, or billboards to try to drive traffic into their store. In fact, the physical storefront itself served as an important advertising tool: the store sign and the window display were theirs to utilize to grab the attention of passersby. How much foot traffic and the general affluence of the local population is an important factor in the store’s rent price as it essentially includes a portion of “advertising” in the lease. The better the location the more advertising a store would get, which meant a higher rent price. As stores go online though, they lose this flow of natural foot traffic and instead need to actively try to draw customers in. While an online store saves on rent—it costs just a hundred bucks for a domain and a year of hosting—it exists just as an address among the other billion plus websites. The savings the online store owner accrues from the very low operating costs instead gets plowed into driving traffic.

This is the “CAC is the new rent” argument (CAC is customer acquisition cost). With no organic traffic, the store operator spends on digital ad space where there is user traffic in order to acquire customers. This is what underpins the strength of an internet platform with a large and captive audience: they are in a unique place to direct just a small portion of their users’ attention away to the online store owner’s advertisement.

Of course, these stores with no physical location couldn’t exist before the internet, but they also couldn’t exist before Facebook. In most cases, these users are not looking to buy anything, so they wouldn’t search on Google for said product, but instead are drawn to buy something when they see it. Facebook and Instagram are not only critical facilitators of such commerce but are really the only scaled players here. No one else even comes close.

Meta’s Advertising Machine.

Competitors such as Snapchat, Twitter, Pinterest, and TikTok all theoretically have the same opportunity, but none have successfully been able to match Meta’s success at onboarding enough advertisers or their matching of ads to users as well (and it is not for their lack of trying).

While Facebook’s first ad deals might have been struck with a normal sales cycle using ad sales reps, they quickly moved to more self-serve digital ad buying, where the average advertiser never interacts with a human and their ads’ placement is dictated algorithmically.

These self-serve digital ads were only made possible by the tens of millions of businesses that were already on Facebook and familiar with their platform. While Snapchat, Twitter, Pinterest, and TikTok are constantly “educating” advertisers on how to use their platforms and what their advertising products can do for advertisers, the average Facebook advertiser doesn’t need any of that since they already had a personal Facebook page which gave them familiarity with the platform, and they could inherently understand the value of “boosting” a post. The low price to experiment, sometimes as little as $5, empowered literally millions of businesses to try Facebook advertising. Facebook would then use all of the data they acquired from the users to best match that ad to people who it would actually resonate with. When the business owner saw a stat sheet detailing how the ad performed, with actual data on user engagement, they could tangibly see what they were buying versus a local newspaper spot where efficacy would never be clear.

Facebook has onboarded an incredible number of businesses; it is at over 200mn currently. Only a small percentage would have to advertise for Facebook to be one of the biggest ad platforms. Currently they have over 10mn advertisers on their platform, the vast majority of which are small businesses.

Having so many advertisers is crucial because it allows them to pick from a vast array of different ads to show their user base. When other ad platforms try to copy Facebook’s ad targeting capability, they have trouble not just because it is technically hard, but because they do not have enough ads in their inventory to make sure there is always an ad that is likely to resonate with a user. Without enough ad inventory, the platform has to curtail how many ads are shown (limiting revenue) or show an ad to the wrong user (reducing ad click-through rates).

This problem is hard to get out of because not having enough ad inventory makes training the machine learning algorithm tougher, so the ad platform’s matching engine suffers, which in turn makes it harder to convince other advertisers to join the platform since ad engagement is low. Whereas Meta has a natural funnel of advertisers from their Facebook and Instagram business pages, other platforms need to engage in a very labor-intensive (usually a man-powered sales force-based) effort to onboard more advertisers.

Facebook not only has the most advertisers and data, but also is in the best position to attribute ads to conversions. This is because of the “Pixel” and various other Facebook SDK’s which businesses and publishers put on their websites to send data back to Facebook. The Facebook Connect initiatives were followed up with a slew of other SDKs and ultimately the “Pixel”, which is a powerful advantage for Facebook. Facebook’s Pixel would facilitate the transfer of data from 3^rd party sites back to Facebook, so they could tie it to an individual’s Facebook profile and retarget ads to them in the newsfeed.

Facebook’s user data, coupled with 3^rd party website connectivity, made it a natural extension for them to offer publishers Facebook ads on their sites. Their 3^rd party ad network is called Facebook Audience Network, or FAN. They have had mixed success with it though, in part because the highest ROAS ads were always on Facebook and Instagram, making FAN a secondary consideration. If an advertiser gave Facebook their entire ad budget to go and find the best place to show ads, it all ended up on Facebook and Instagram. FAN is a small part of their business today and was discontinued for mobile web, but we can still see info pages for mobile apps (not clear if it extends to the open web).

FAN (or as it’s now called: Meta Audience Network) is a worse business than their core advertising though, because they have to split revenues with the web publisher or app property owner versus keeping it all for themselves on their owned properties. Additionally, as the world is increasingly focused on privacy and data concerns, this business is most susceptible. (In specific, IDFA, and ATT has rendered their Audience Network much more ineffectual, with an increasing reliance on contextual ads instead of targeted). Some estimates put their 3^rd party ad network around ~2% of revenues (net of revenue shares/traffic acquisition costs). Rather than expend efforts into this lower margin and regulatory fraught business, Meta would focus on advertising their owned properties. 

We will now go into the specific revenue drivers of their advertising business.

Ad Revenue Drivers.

Ad revenues are simply ad pricing multiplied by ad impressions. There has always been an interplay between the two as they are usually impacted inversely. Increases in impressions decrease price and decreases in impressions increase price. (But there are a few other factors we will mention momentarily). The chart below notes some of the bigger events that distort ad pricing or impressions.

Starting with the shift to mobile, ad impressions started to plummet as they traded a right-hand column of 5-7 ads for fewer ads in the mobile newsfeed. However, as Facebook learned, these newsfeed ads were more engaging and enjoyed higher click-through rates (CTR). The higher click-through rates supported higher ad pricing, as advertisers increased their bids until they hit their ROAS limit. But as Facebook continues to learn what ads work best with which people, the click-through rates continue to improve, which in turn increases advertisers’ willingness to pay, since their ROAS would go back up.

Dictating the price of an ad is simple supply and demand. The more advertisers that bid for an impression, the higher the price. The more impressions available, the more bids are spread out. Improving ad targeting works to push both price lower and impressions higher in the short run. This is because the more effective an ad is, the fewer the people that need to see it. This in turn means that there is more ad inventory available.

For instance, if an advertiser set the goal of 100 downloads of their gaming app, with poor targeting it may take showing that ad 200,000 times, but with better targeting they could do it in 10,000. That means there are now 190,000 more ad slots that are empty. The flood of ad impressions can depress prices, but it is actually a good thing—it means their machine learning algorithms are improving.

It is hard to have an opinion on whether ad pricing and impressions increasing or decreasing is a good or bad thing without broader context. Generally speaking, prices dropping is encouraging, because that can often be attributed to improvements in targeting. However, if prices were to drop because advertisers were leaving for other platforms, that would be a problem (this has never happened, though. Even the Summer 2020 Facebook protest had an immaterial impact to the number of advertisers bidding in their auctions. Despite some big-name advertisers pulling out, it was inconsequential to their overall ad revenues because there are millions of businesses bidding for every ad impression, making any single bidder—even a big one—unimportant). Other common reasons for ad prices dropping include disproportionate growth in lower income countries, which monetize at lower rates, and currency exchange changes.

Impression growth can be worrisome if it is not alongside user or time spent growth. This is because there is a limit to how many ads you can show your users. Saturating existing users with more ads is a lower quality form of growth because you can only do it so much before it can lead to user churn. If impressions are growing because more users are joining the platform or increasing their time spent, then that would be positive.

Facebook and Instagram prioritizing different products prior to monetizing them is another reason we see impressions dropping. Facebook was still growing when Stories were launched on Instagram, but we can still see impression growth dipped down in 4Q16 and trended lower for the next few quarters. This is because with the Stories product launch, Facebook didn’t show ads against them, and Stories simultaneously cannibalized the ad-laden newsfeed. We see that in early 2018 when ad load in the Stories starts to grow, Facebook called it out as weighing on ad pricing.

A Instagram feed ad (left) turned into a Stories ad (on the right). The more engaging Stories ads are vertically shot videos though.

When they first rolled out ad units for Stories, few advertisers had that format, so the bid density (amount of bids per ad slot) on Stories ads slots were light. This meant that pricing wouldn’t be that competitive. That, plus the increasing Stories ad load, kept ad pricing trending downwards for the next couple years. Investors would talk about the ad pricing difference between Newsfeed ads and Stories during that time, skeptical that Facebook would be able to reach ad pricing parity between Stories and Newsfeed. Today though, they are largely the same.

This is key to keep in mind as we see a similar dynamic currently as Meta cannibalizes usage for their new Reels product (which is not monetizing as well as Feed or Stories). In 4Q21, Meta disclosed that North American ad impressions fell 6%. Total ad impressions growth hasn’t been negative since the end of the mobile transition in 4Q15, so this was a concerning development. In the past during the Stories transition, they were still growing users and ad load enough that net impressions still grew during that period.

In 4Q21, Meta called out the transition to Reels impacting impressions, but much more concerning, losing time spent. Zuckerberg notes that the transition to Reels is similar to the prior transition from desktop ads to mobile Newsfeed ads and from Newsfeed to Stories. The idea is that they first create a product and slowly nurture it until it reaches the scale and usage that warrants advertising. However, while in the short-term it looks scary because ad revenues fall, it is the right long-term decision for the health and sustainability of the business. The issue here is: it is not entirely clear how much of the impression drop-off is self-inflicted versus the result of time spent leaving for TikTok. It is a much easier task to start showing ads against Reels than try to win back time spent, especially when the competition is as highly addicting as TikTok is.

To the positive, on the past few earnings calls, management has noted that Reels time spent is incremental to the users over all session (rather than just cannibalizing Newsfeed/ stories time). In 2Q22, they noted that a change in their AI model to recommend videos resulted in a 15% increase in Reels watch time. The fact that Reels is already time additive to time spent and they are still in the early innings of improving their AI-recommendation algorithm is encouraging. We will pick back up on this in the Competition section.

However, as it relates to ad pricing and impressions, there are two factors skewing the outputs. One is Apple’s privacy changes (IDFA and ATT), and the second is the overall macro environment. We will touch on Apple’s changes and the risks they pose to Meta in a later section, but in short, it hurts Meta’s ability to take credit for conversions and potentially their ad matching engine. Meta estimates that ATT alone caused a $10bn hit to revenue, but it is possible that it was actually more. The drop in ad pricing can partially be attributed to these privacy changes that have resulted in lower ROAS and thus less advertiser willingness to bid up the price of an impression. Convoluting this all though, is that there was an ecommerce boom during Covid, which, when combined with stimulus checks, created elevated advertiser demand and pushed up ad pricing. Some of the ad price drops could be a “hangover” effect, but how much to attribute to that is unknown (further convoluting things is that a lot of VCs were very active in investing in start-ups which in totality have huge Facebook and Instagram ad spend). 

With the general economic environment deteriorating, it’s not clear how much of the adverse results can be attributed to TikTok taking time spent, the self-inflicted Reels transition, ATT, or just general macro conditions and the aftermath of a digital ad boom. We will dissect competition and Apple privacy changes more, but the confluence of these events makes it hard to gauge which variables are more meaningful to Meta’s business deterioration.  

A Meta data center. Increasing server capacity to run more advanced algorithms is part of the solution to stitch together lost signal from ATT.

Competition.

Competition for Meta’s Family of Apps comes from many different areas. We will bucket competitors broadly into two areas: direct competitors and time spent competitors. In a moment, we will go through the 12 key factors of variation for different social media apps to show how most competition is tangential to the core value prop of their apps (specifically Facebook and Instagram. Messenger and WhatsApp have more direct competitors).

Facebook is a social network, with users interacting online with people they know in real life. Instagram also has that aspect, but users can and do also interact with pseudonymous accounts as well as influencers/celebrities. There are many social media players, but far fewer true social networks. In fact, the only real scaled social networks are Facebook, Instagram, LinkedIn, and (maybe) Twitter. However, competition doesn’t just come from other social networks, but rather from other social media apps and any app that allows sharing.

If we think of what problem a social network solves, most of those individual functions can be done elsewhere; it is just a format that lends itself best to certain types of sharing. At its core, a social network is a content repository that is integrated with a unique distribution channel that is customized by each user in a relatively frictionless way.

The key to building a social network is to have each user’s identity (whether that be real or pseudonymous) housed on the platform. When a user signs up for Facebook or Instagram, they are creating profiles right on the platform for other users to look up and follow/friend. This is critical, because without this, the platform does not have a distribution channel for users to stay connected to their follows/friends. Many social media apps do not require a searchable profile (YouTube, TikTok, Pinterest, Reddit) and because of that, they cannot create a social network.

There are three necessary conditions to becoming a social network: 1) social media, defined as user-generated content, 2) distribution, which is an on-platform sharing function, and 3) an ability to follow, connect, friend, or subscribe to an individual user. This means that competition can come from anyone who either specializes in social media orsharing. It is not necessary to be a social network to compete against a social network.

Being able to create and share content on a platform puts you close to becoming a social network, but we believe there is one more requirement: a user needs to be able to have a way to “connect” to an individual user without starting a separate communication each time. This is why email, despite allowing users to create and share in the same app, shouldn’t be considered a social network. For a similar reason, Reddit, which allows users to post content and follow groups, wouldn’t fit the definition because Reddit user’s follow interests and not individuals. Under such a definition, really only Facebook, Instagram, Twitter, Snapchat, Pinterest, YouTube, and LinkedIn can be considered social networks.

We could get more pedantic with the definition to better grapple with the spirit of a social network, but we think this definition is sufficient enough to paint the competitive environment. (In theory, we would want to exclude Pinterest as social network because of the low level of on-platform sharing and more of a focus on interests. YouTube also isn’t really a social network because there is no peer-to-peer sharing on-platform and a lot of usage that occurs isn’t even logged in. Some new apps like BeReal would also technically fall under the definition of a social network despite the low daily active users, minimal time spent, and relatively small scale).

In an effort to organize all the different ways a social platform can vary, below we show how different social platforms vary across 12 key factors.

1) Content: What sort of content is on the platform? YouTube has a lot of PUGC (professional user-generated content) as well as some professionally produced content, but TikTok is mostly user-generated content.

2) Content Delivery: How does the platform decide to show a user content? Whether content shown is dictated by a user picking what content to see (friend, follow, subscribe) or recommendations is a key distinction. TikTok was the first scaled platform to totally eschew friend recommendations and the ability to follow as a principle means of mediating what content you would see, and instead relied entirely on algorithmic recommendations.

3) Content Life: How long does the content stay up? Snapchat gained prominence because it made everything self-deleting with nothing hosted permanently.

4) Content Interaction: How do users interact with the content? Posts in a feed might merit a like if they are a close friend or upvote on Reddit, but Snaps or messaging apps typically require a timely response. The need to directly respond to someone on a messaging app (call to action) is party why messaging apps have such low churn and high engagement.

5) Main Communication Use(s): How do users communicate? Some platforms allow many people to comment on a central conversation topic (Twitter or Quora) and others allow for a more fluid conversation between many people (Discord or GroupMe).

6) Privacy: Who can see what content? Facebook requires a mutually approved “friend request” in order to gain access to content. Instagram and Twitter allow you to make your account public or private. Instagram and Twitter public accounts follow a 1-way “follow” model that enables a user to see someone else’s content without revealing their own. 

7) Identification: What are the identification rules of the platform? Facebook was the first real name social network, but Instagram’s flexibility with pseudonymous accounts meant more creative expression and a new generation of influencers.

8) Graphs: Does the platform have a social, interest, or work graph? Facebook knows all of your friends and some of your interests, whereas YouTube and TikTok focuses on learning just your interests.

9) Main Medium: What form of content is primary? Twitter popularized the 140-character (and later 280) text post, whereas Instagram was initially primarily images. TikTok was solely short-form video before increasing the video time limit.

10) Source of Connections: How do you find people in the app? Whether the app has searchable profiles for people or makes you input a phone number or platform-specific username could be key to stoking or hampering growth. It could also have privacy implications. Snapchat’s lack of central profiles to search through means it is unlikely a distant acquaintance finds you on the app, whereas Facebook is notorious for helping distant friends reconnect (especially with the People You May Know feature).

11) Purpose: What is the main purpose or “job” the app accomplishes? Users go to TikTok to be entertained and WhatsApp to message. Apps like Facebook or Instagram have multiple uses like keeping up with friends (information), connecting with people, entertainment, or even looking for cheap secondhand products in the case of Facebook Marketplace. (We list three main categories for simplicity).

12) Posting Expectations: What sorts of posts do users expect to see on each platform? For instance, Snapchat is known to be home to more unfiltered content, whereas Instagram is notorious for making people feel everything must be perfect in each post.

Below is chart that shows how we grouped answers to each of these 12 key factors of social platform variation.

We then labeled some of the most popular platforms (as well as a few that are defunct) for these 12 different variables so you can easily see where they differ. It’s interesting to note how just a few small variations on form factor can determine the success or failure of an app.

TikTok’s big innovation, for instance, was sticking to just algorithm recommendations (no friend recommendations or friend follows to dictate what content you see). Combining this with videos, which is the most engaging content, that are short, so you don’t lose attention, proved to be a winning combination. In contrast, Snapchat gained ground by making a more private social network with photo-based communication that was ephemeral, which made people very comfortable sharing more mundane moments that Instagram no longer seemed to encourage. Twitter would focus on pithy text and allow 1-way follows that were perfect for public figures, thought leaders, or organizations.

It is just along these 12 key factors of variation that we show below that most social media platforms vary.

Yik Yak was an app to leave anonymous public messages for others to browse through, aimed at a large group of people like a college. It was eventually shut down due to lack of content moderation and accountability, but it shows how even some of the more obscure mixes of the 12 key social platform factors can lead to some success.

Thinking of the “Jobs to be Done” framework, where we think of what job a consumer “hires” a product for, we see that Facebook and Instagram have very few direct competitors. (Messenger and WhatsApp have a few though—iMessage, Snapchat, Android Messages, Signal, and Telegram, among others).

It is somewhat ironic that Meta’s biggest current competitive threat isn’t coming from a social network, but rather an app that does zero of the things Facebook and Instagram do. In our table above, we note the “purpose” of each app, and you can see that despite most of the purposes differing, they are all in the same competitive set.

As mentioned, we should think of competition coming from two distinct areas: 1) direct competition, which is peers who try to compete feature for feature with Meta’s apps, and 2) time spent competition. In order to think through why Facebook and Instagram still have no direct competitors for certain of their core “jobs to be done”, we will briefly go through what people use the apps for.

Jobs to be Done.

If we think back to the college dorm room days of Facebook, it was really just a digitized repository of students’ information. There was no Wall, no Newsfeed, no messaging, no ability to post more than one photo (profile picture). The jobs this site served were mostly for college students to look up other students, gossip about them easier offline, know who was single, and experience the novelty of having their own webpage on the internet.

As they added more social functionality, it went from an inert repository of info to a live and engaging site. This helped many people socialize more and feel more attached to a community. The friend request still signified some elevated level of connection, and it wasn’t the symbolic-less gestor it is today. With email feeling too formal and in a time before the mobile smart phone existed, Facebook opened up the world to a more causal and continual form of communication. The Facebook Wall and Newsfeed would be the sole place to communicate broadly to your friend group, and critically, it was also free and required no hardware purchase.

However, after Facebook launched Open Registration, a user was increasingly pressured to accept friend requests from disparate groups of people in their life: work colleagues were mixing in with college friends, relatives, and parents. While most are familiar with Metcalfe’s Law (a network’s impact is the square of the nodes in the network, or in layman’s terms, the value of the network grows faster than the users), Reed’s Law is more insightful here. It posits that the networks that contain subgroups are even more valuable than those without subgroups because it allows more meaningful communication.

The more people that got onto Facebook, the less anyone had to say. Gone are the days of posting a full album of college party pictures for fear of someone’s parents or future employer seeing them. Facebook was acutely aware of the problem and internally people proposed creating a separate Facebook for work. Zuckerberg was an advocate of Radical Transparency though, and he felt that allowing a user to be one person in one group and another person to another group was duplicitous. He was adamant that people should be the same person with everyone. But most people have different personas with different groups of people. In fact, it would actually be seen as a form of social ineptness if someone treated their boss, parent, distant college acquaintance, and close friend the same way.

However, on Facebook, there was really only one way to communicate to all. This dramatically changed the fun of Facebook. (If Zuckerberg would have allowed multiple “IDs” on Facebook, or a post to different groups of people like Instagram has today, perhaps it wouldn’t have dropped off in popularity so much with teens. The idea that teens didn’t want to be on Facebook because their parents were there isn’t because they don’t want to do the same thing as their parents. No one says they don’t want to use iMessage because their parents use it. It’s rather that any attempt of socializing would feel monitored by their parents, which is the opposite of why a teen would use social apps in the first place). 

A TV show commented on this degradation of the friend network in South Park’s “Randy wants to be your Facebook friend”.

Zuckerberg’s staunch insistence on one single identity for everyone might have made sense to him since he literally slept in the same house as his early employees, working in the living room. But for other users, the need to bifurcate communication between different groups of people was critical. As users grew up from high school and college and spent more time in the workforce, their Facebook identities would have prior life stages calcified permanently on their walls. One of the most common reasons someone had a duplicate Facebook account was to “start over”. There is very little a user did in high school or college that they would want to tie to their permanent internet identity. As the platform grew older, this became a bigger problem, and Facebook users were increasingly holding back from posting, worried about how anything they did post could look to a potential future employer. (Google+, Alphabet’s failed social network, addressed this by letting you only post to “circles of friends”).

Google+ Circles let a user post content to different circles of friends.

This created an opening for new platforms. Snapchat was launched in 2011, taking advantage of this blind spot of Facebook’s, making every communication direct to the person you wanted to see it (no parents stumbling on your posts) and was self-deleting. These two value props were exactly what users desired and it opened up a ton of social networking. Early users were opening the app as much as 50 times a day. Evan Spiegel would also take advantage of the prominence of mobile phones with cameras built-in, creating a more visual form of communication, the likes of which weren’t possible before. Whereas Facebook grew up on text and would broadcast every one of your actions to everyone you knew, Snapchat was photo-first with every message sent only to desired recipients and self-deleting.

In 2013, Snapchat launched “Stories”, which let users post vertical videos of what they were doing to their friends, and the private nature of the app coupled with a lack of “likes” or “comments” made them the #1 place for casual and spontaneous posts. (The nature of Snapchat friends not being public for others to see meant there was no arms race to add everyone to raise your friend count. Also, the fact that you needed a phone number or Snapchat username to add someone meant social circles were kept smaller due to the higher friction of adding friends. Facebook took the opposite route with People You May Know and scraping email lists for potential friends to suggest to you). Snapchat was serving many of the needs that Facebook originally did when it was closed and kept small. In their early days, Snapchat was growing even faster than Facebook did. Evan Spiegel was fond of saying “if the 1% of your best pictures go to Instagram, I want the other 99%”.

Shapchat Stories was rolled out in 2013 as a new way of sharing that allowed users to build chains of shared content that could be viewed an unlimited number of times over a 24-hour period.

Instagram launched a year prior in 2010, and similarly to Snapchat, took advantage of Facebook’s soft spots: mobile and photos. Like Snapchat, Instagram was a mobile-native app that was focused on photos. However, whereas Snapchat was more of a messaging app that focused on connecting you to people you already knew, Instagram was ultimately about entertainment, expression, and sharing what you were doing (but not necessarily to specific people). They would create a very visual app, with each post being a large photo with room for a small comment underneath. People would share small moments and beautiful things, but you “followed”, not “friended”, people. The 1-way follow, coupled with the allowance of pseudonymous accounts, made Instagram a great venue to create content and consume it on the same platform. As the app got bigger, people would eventually follow a lot of their friends, and Instagram backed into creating a social graph without an original intention of doing so.

However, as Evan Spiegel noted, the bar to post on Instagram grew too high with a lot of “professional” Instagram creators making the average user’s photos look bad. Lucky for Instagram, Spiegel came up with the solution for their problem: Stories. Stories allowed the more spontaneous and ephemeral posting that Instagram was originally about. Instagram’s larger network also meant that they clipped Snapchat’s growth, as they now filled that need that users were leaving Instagram for.

Instagram would prove more resilient in another regard too: identity. Since Instagram was never intended to be a social network, they never had such a steadfast opinion on real name identities. People could sign up as whoever they wanted. The byproduct of this was twofold. The first consequence is that if too many loose acquaintances (or your parents) started to follow your Instagram, you could simply make another account. These second accounts, dubbed Finstas (fake Instagrams), accommodated the need for more private spaces to share with closer groups of friends. When a Facebook user encountered this problem, they simply stopped posting. An Instagram user would spin up a second account under a pseudonym and follow some close friends. The flexibility in this regard meant that users could split posting activity between accounts. They would later roll out a “post to close friends” feature to precisely address the issue of a person’s network growing too large.

A user just clicks their username at the top of the screen to switch accounts, making it easy to go in between a Finsta and main Instagram account.  

The second outcome of pseudonyms was the rise of the influencer. When just starting out, few people want to dedicate their only profile to a singular interest (i.e. beauty, cooking, health, etc.), but are happy to experiment under a pseudonym. If you recall, this was partly why MySpace initially took off so quickly. People liked having separate internet identities. Having influencers on the platform was important because it was usually the best form of entertainment. Whereas Facebook’s Newsfeed was limited to your social graph and stuff they “liked”, Instagram had a whole swath of accounts you could follow just because it met your interest and you liked the content (Facebook pages were never adopted in the same way). This meant that as Instagram started serving your social networking needs, it was also a place to be entertained, discover, and learn about things you liked. In the meantime, people were posting less on Facebook because their social networks grew too big, so there was less content to consume.

Ironically, while Facebook’s users’ social graph was becoming too big, it wasn’t open enough to compete with Twitter. Twitter was launched several years prior, in 2006, and similar to Instagram, used a 1-way follow model and allowed pseudonyms. The 140-character limitation made it more entertaining than reading long Facebook posts and 1-way follow allowed the witty users to gain large followings. The follow model also made it better suited for celebrities or organizations, and it quickly became the best and quickest way to distribute information broadly. Twitter’s early lead with this sort of communication would never be fully recovered by Facebook, whose Pages product limited friends to 5,000 until 2009. Most people posting and commenting on Tweets were pseudonymous too, so despite Facebook’s much larger user base when they removed the limit from Pages, the form function was different.

Facebook’s homepage was re-designed to make it feel more like Twitter.

After spending so much effort to expand a user’s social graph, Facebook would reemphasize Groups and Messenger, functions that are aimed at shrinking a user’s reach. Facebook’s attempts to be everything to everyone had left many users with a confused sense of what they actually needed to use Facebook for. However, Facebook is still a sort of utility for billions of people who find at least one feature useful. Despite many young adults never posting on their Facebook, they will still use Marketplace (P2P commerce), Groups, or Causes (charities/social change groups). Big life events like weddings, family vacations, and welcoming newborn babies will still get posted to Facebook (or cross-posted alongside Instagram) as these are events you are fine with everyone you know knowing about. However, active public Facebook posters are mostly people that are older in the Western world (areas in Asia and ROW are earlier in the life of Facebook, with less of a “social graph is too big” problem). As Facebook is still the only real name social network, if you want to invite a number of people to a big event, there is still no better option than Facebook when you don’t have guests’ phone numbers or emails. For similar reasons, Messenger still serves a critical function of messaging people who aren’t in your contact book.

Facebook is the closest anyone has ever gotten to creating a universal real ID, but they weren’t able to fully capitalize on it. Part of what made Facebook grow so much early on was the push to connect everyone to grow the content each user would see and benefit from the best of network effects. There was a period where it was weird to not be on Facebook, like it was to not have an email or phone number, but that period has passed. Partly because of privacy scandals, but mostly because users just didn’t want everything they did on Facebook or across the web to be broadcast out to an increasing number of people who were their “friends” permanently. (They would try to address this issue later on with Groups, post to close friends, “manage activity” to delete old posts, Stories, pseudonymous pages, apologizing for Beacon, reversing a decision to make more posts de facto public, and an ever increasingly complicated privacy control page, but the damage was done).

Since they didn’t win the spot to become a sole, universal ID for people to search up and message (like how Tencent’s Weixin/Wechat is in China today. Weixin contact info is housed in the app and not the smartphone’s contact book so other apps cannot access it without Weixin’s permission), they couldn’t close the competitive set to box out other messaging apps. Any messaging app could collect phone numbers and turn that into a user’s “Profile” with a name and profile picture and then let users socialize. Facebook would buy WhatsApp for $19bn not only because it was growing rapidly, but because such a large network of users could pose a risk to Facebook if they decided to back into a social networking (to be clear, they didn’t seem to have any ambitions to do so, but neither did Instagram when they started). The point is that when you have hundreds of millions of users with a ~80% DAU to MAU ratio and more than 30 app opens a day, you have ample opportunity to experiment and expand the app’s functionality (exactly what Tencent figured out with Weixin). More recently, WhatsApp has moved more into Groups with Communities (multiple group messages organized around a single “community”), which is one of the key draws remaining for many Facebook users, showing a potential susceptibility to losing this usage over time.

While Apple is unlikely to push more into social networking functions, you can see the risk of a messaging app clearly here. They push users to add a photo and name to share with others when they message. They could then let users just search by name (instead of number) and add other features like Stories, a wall, Groups, etc. Again, this isn’t a material fear with Apple given their reluctance to push cross-platform apps, but a potential fear if another messaging app gains widespread prominence.

Despite Twitter, Snapchat, and Instagram exposing Facebook’s weak spots, their acquisition of Instagram and later platform reformulations to copy some of the best features of Twitter and Snapchat were enough of a competitive response that no one can honestly say that Twitter or Snapchat present an existential risk. We started this section noting that competition can come from two directions 1) feature competition for the same “jobs” that Facebook/Instagram do, or 2) time spent competition. On the first point, no one can replicate what Facebook and Instagram do very well.

They are the only platforms with broad usage in the billions that enable creation and sharing on the same platform amongst a broad group of people, including most of everyone you know. Instagram may have proven to be a better format to share and create over time than Facebook, but Meta owns both anyway. Snapchat doesn’t serve public sharing and people do not have their real life friends on Twitter. The real competition is coming from those they didn’t know they were competing against.

Disruption & Orthogonal Competition.

The term disruption was originally used not to describe products that directly compete with a “core job”, but rather a product that is too subpar to serve that core job, so instead was used for a different job. As that product gets better at that unrelated job, eventually it becomes good enough to also serve that core job. One classic example is small hard drives that had inferior memory, slower speed and higher cost than the core large hard drives. The large hard drive makers were correct in saying their products were better in every way, but the smaller format hard drivers served new markets in smaller devices and eventually could get good enough to also serve the markets the traditional drives served. The small hard drive technology is “disruptive” not because it was competing against the legacy product, but precisely because it originally wasn’t.

Zuckerberg’s paranoia of a prominent messaging app isn’t because it competes with Instagram, but precisely because it doesn’t… yet. It is very hard to monitor products that don’t currently compete with you as there are so many. Facebook watched YouTube for years in case they encroached on their space, with many efforts in video to thwart a potential infringement. However, they missed that it would be a short-form video app, with zero known creators or influencers, and zero social graph that would come to be seen as their potential Achilles’ heel.

Enter TikTok.

The Chinese version of TikTok, Douyin, was launched by its parent company ByteDance in 2016, and only in China. They launched TikTok as the international version of Douyin a year later, but it didn’t take off until after merging it with Musical.ly. Muscal.ly was a lip-synch video app that would allow TikTok to seed their content library and kickstart traffic with their ~200mn users. Zuckerberg also tried to acquire Musical.ly, but he didn’t pursue it as aggressively as he did with past apps like Instagram and WhatsApp, and ByteDance moved quickly to close the deal with an offer over $800mn. The musical.ly content library combined with their large user base was enough for them to start growing rapidly: TikTok became the first Chinese-made app with broad global usage.

ByteDance acquired Musical.ly and merged it to TikTok, the Western version of their short video product Douyin.

In contrast to Facebook or Instagram, TikTok does not rely on a network of friends or user-selected “follows” to dictate what a user sees in their feed. Instead, they use AI to make recommendations based on a user’s activity. This turned out to be a soft spot for Facebook and Instagram, whose users could bore easily since content was constrained by their friends or follows. TikTok would continue to show a user content that was evergreen and personally suited to their interests, even without the user having to “follow” or “like” anything. This was both a feat of programming ingenuity and stumbling on a key reason why people reach for Facebook or Instagram: boredom. The short form video app is not only highly entertaining, its stream of content is virtually endless, leading to many teens spending upwards of 2 hours per day on the app. While many users would continue to use most social media apps, TikTok’s share of time spent was increasingly growing with a nontrivial portion (some estimate around ~60%) of Gen Z spending over 90 minutes on average on the app each day. However, it wouldn’t be true to say Facebook didn’t see the appeal of this format earlier.

From Vine promotional material. Users could post their Vines, which were 6-second videos, to other social networks and in-app.

Vine, the original short form video app from 2012, was shuttered in 2016. Twitter acquired them before Facebook had a chance to. Facebook responded to Vine’s viral growth by pulling access to a Facebook API that enabled their “find a friend” feature in 2013.

Facebook followed that up by launching a standalone collaborative short video app, Riff (for short collaborative videos with friends) and also introduced short fifteen second videos to Instagram in 2015. Riff was shut down shortly after, but luckily for Facebook, Vine was poorly managed and underfunded. Vine reached 200mn users at its peak, but Twitter management squandered the opportunity and eventually shut the app down despite it still being widely popular. Instead of Facebook just creating a copy-cat app exactly like Vine, they would move their video efforts to Instagram and Facebook Watch.

Facebook Watch was launched after observing the popularity of video in the feed. However, Zuckerberg felt that the Newsfeed should be primarily reserved for more social experiences, so he relegated videos to a separate tab. The Watch tab allows users to browse user created videos, but their recommendations are pretty mediocre, seem to border on random, and their content library pales in comparison to YouTube (YouTube’s recommendation engine was incomparably better too). Facebook would supplement the UGC (user-generated content) with several dozen short original shows they bought exclusive rights for (similar to what Quibi would later do).

Facebook not putting video recommendations in the feed though would limit its greatest advantage—access to the billions plus people who log onto Facebook every day. Creators also complained they couldn’t monetize that well (Facebook copied YouTube’s revenue share model of paying out 55%), but videos weren’t monetized until after human approval, which meant missing the ability to earn on videos when they were the most viral. While a lot of users will land on the Watch page—over 1.25bn in 2020—they have a loose definition of a video user, counting anyone who watches just 60 seconds of video. In total, their time spent there is relatively tiny. YouTube never had much to worry about from Facebook Watch.

TikTok would disrupt this balance and show that Facebook’s video strategy left it susceptible. TikTok’s powerful recommendation engine, coupled with their catchy short-form content, proved to be highly addicting. By the end of 2018, they would have over 250mn users.

Facebook released a TikTok clone, Lasso, in response to the popularity of TikTok.

Facebook responded by launching a stand-alone app, Lasso, in November 2018. Similar to Snapchat clones Poke and Slingshot, and Instagram clone Camera, Facebook would shutter this app two years later. Whereas ByteDance bought Musical.ly to merge their 200mn users with TikTok, Facebook would never use their immense Facebook and Instagram traffic to seed Lasso. This noncommittal strategy gave TikTok more time to rapidly grow and dominate the screen time of the world’s youth.

The TikTok threat became increasingly intense, reaching 1bn MAUs by 2021 with a huge swath of users watching ~2 hours a day. In an internally leaked report from March 2021, we can see how Meta was viewing their apps in light of this development.

Below is a slide that shows TikTok time spent has doubled across all users, showing they are taking an increasing portion of time spent growth. Meta estimated that teens spend 2-3x more time on TikTok than Instagram.

The slides are a little blurry, but the left shows teen time spent averages ~35 minutes a day and non-teen is closer to 30 minutes. On the right, it shows TikTok gained ~1 hour of time spent while Instagram gained 5 minutes across all DAUs (this was published 3/2021 so the pandemic is likely a factor).

In the slide below, they show that in the US, retention is starting to slip and daily session are down a bit too. Most concerning though, is that production is slowing down. The less content users produce on the platform, the less there is to consume. More reticent posting behavior and less creation is a red flag for the health of the ecosystem.

Below, we see a slight drop of 2-4% in total production on their apps. They do not attribute a reason explicitly, but the concern is that people are making TikToks or sharing more privately rather than posting on Instagram. Increasingly, there is the sentiment that Instagram is a place for influencers and celebrities with the average person’s posts not good enough to be worth posting. Stories helped squash that sentiment to an extent, but many teens today still don’t post nearly as much as earlier cohorts.

These concerns stoked a more unified and concerted effort to squash the TikTok threat.

Instagram a few years earlier tried to address video separate from Facebook’s efforts. Instagram launched Instagram TV or IGTV in June 2018. IGTV allowed users to post videos longer than one minute and up to an hour, as well as subscribe to channels, much like YouTube. It was accessible through the Instagram app as well as a stand-alone app. While it made sense for users be able to consume longer videos in Instagram and this would allow creators to serve their audience without forking them over to YouTube for their long videos, they made a critical error.

Instagram with the IGTV icon on the top right. It is notable that they gave the feature home page space versus Facebook’s typical strategy of creating totally detached, stand-alone apps.

YouTube isn’t just a tool to host long-form content, but is also a source of demand. Their discovery and recommendations were what made it the original time sink, with a stream of recommended videos auto-playing. And if that wasn’t the right video for a user, there is a long string of videos to scroll through on the right. This means that YouTube helps users find videos they want to watch, but also serves as a source of demand for creators. On top of that, YouTube would share 55% of ad revenue with creators. The value prop just didn’t make sense for creators to leave YouTube.

While it perhaps opened up a new market for those popular on Instagram but with no video channel already, IGTV’s lack of a strong recommendation engine would make it hard to gain a viewership base. In April 2020, they rebuilt the IGTV app with more robust discovery emphasis, but the timing was poor with the rise of short-form video. TikTok was simply so much more addictive and was far better at serving up entertaining personalized content to each user.

The short-form format meant that users would quickly be given another clip before getting bored, but it also meant the algorithm would have much more data to personalize videos to a user. If a user watched one 5-minute video to completion, you would know far less about them than if you gave them 5 minutes to spend flipping through an endless roll of ~20 second videos that they could quickly skip through at the moment of disinterest. Shorter-form videos just worked better to hone in on recommendations, making it much better at entertaining.

Instagram app overhauled in 2020 to focus on more discovery.

Instagram would shift their efforts to short-form videos with Reels in August 2020 and kill off IGTV in 2021. Despite Meta deciding to sunset IGTV, we could still see it having worked longer-term with more patience. But it was the wrong focus at a time a competitor was rapidly growing time spent at a rate not seen before.

When Instagram initially copied Snapchat’s Stories feature, it didn’t have the impact of churning Snapchat users, but it did slow their growth. The many users who may have joined Snapchat and become avid Stories posters instead used the feature on Instagram and never left. Reels may do a similar thing for Instagram. It is unlikely to win back TikTok users, but it reduces the risk that other users who are already on Instagram adopt the app.

After several meandering experiments with video, Meta took TikTok to be an existential risk. No longer would Meta’s video efforts be an unattached app or invested in longer form Watch or IGTV. They would create a blatant copy-cat product and put the full heft of Facebook and Instagram behind it, just like they did with Stories. However, this comparison warrants further examination because Stories was a more natural extension to Instagram than Reels is. Meta is reworking the entire Instagram app and reconceiving the purpose of Instagram in the process of fending off TikTok.

The TikTok app opens to full screen vertical videos. A user can simply swipe up to see the next. Their algorithm is very good at quickly personalizing content that each user will find compelling.

The advantage Facebook and Instagram have with your entire social network connected to you actually is a disadvantage in facing off with TikTok. A user downloads TikTok with it disconnected from any social graph. Whereas Facebook and Instagram let each user find what content to see, dictated by their friends, follows, and the “likes” of each, TikTok had none of that information to use to serve up content. Facebook and Instagram not only require the user to actively input the information to feed their recommendation engine, but it was also limited by the users’ followers/friends. Facebook and Instagram serve up content to a user based explicitly on their approval of what content to see and from their active engagement with likes and comments.

TikTok, lacking such information, would focus on gathering all of the information they could. Every time a user lingered on a video, quickly skipped it, paused it, rewatched it, and every other action is recorded and fed into an algorithm that tests various content recommendations to optimize for users’ engagement. The information they wouldn’t have—your friend network and follows—would actually be an advantage because they didn’t have to try to force that data into the algorithm as well. And as it would turn out, what your friends like doesn’t yield nearly as good of a recommendation for you as your own on-platform activity.

Meta’s efforts in video initially always relied on the users’ friends and follows, but this led them astray. Optimizing for the data they had, rather than the problem they were trying to solve, made them miss the TikTok formula. TikTok put all of the pieces together to make the most addictive app ever created: 1) gather all on-platform user data, 2) eschew all media but video because it is the most engaging, 3) limit the video length because attention spans are short, 4) let every creator compete directly for attention with the potential for every upload to be shown to any user, 5) experiment with content shown to each user to get a better sense of their interests, and 6) use advanced AI to synthesize all of this data and serve up the most potentially addictive content personalized to each user. This formula is further optimized by the huge audience that also actively creates TikToks (>80% of users have posted a video). The greater the content library, the higher the usage, and the larger the audience, the more the discovery engine can hone in on compelling recommendations for each user.

It is critical to note that the Facebook and Instagram model does not allow them to copy TikTok outright without changing their formula because the core of their platforms are not built to show random content to users in order to gauge their interest: Facebook and Instagram are built on an intentional “opt-in” to see content, not a passive and endless ream of content that is pushed to a user with them “opting-out” by swiping up. Meta shows you content from a limited subset of what a user opts-in for, whereas TikTok picks content from an unconstrained library.

Realizing this, Instagram started pushing content to users from people they didn’t follow. However, in the process, they upset users and risked changing the value prop of Instagram. In the past, Zuckerberg seemed adamant that more passive consumption of content could ruin the reason people came to FB/IG in the first place, which was connecting and interactive with other people.

Zuckerberg was referring to the rollout of Facebook Watch, which was demarcated from the Newsfeed, because he was more concerned with the quality of social interaction than increasing time spend. He reiterated his thinking again the following year.

Now though, Reels would be inserted into the main feed. This not only meant there would be more passive consumption of video, but also that for the first time, content from random people with no connection to the user would pollute a user’s feed. This change in ethos can be taken as indicative of how threatening Zuckerberg views TikTok to be. With few exceptions, Zuckerberg has never capitulated on what the product should be, let alone pivoted away from personal social connections.

Kim Kardashian protests Instagram’s changes (left) and Head of Instagram Adam Mosseri would address user complaints head on (right).

While the changes were met with user backlash, particularly from large influencers, it proved to be short lived (there is also a factor where more “discovery” content from random creators that a user doesn’t follow hurts big influencers more, since they will be competing with a broader audience for attention). The Head of Instagram, Adam Mosseri, addressed the issues in a short post that essentially reiterated that this was the direction Instagram would go in despite the criticism. Just like with Zuckerberg’s prior product feature changes that were met with user protests, he would stay the course, believing that in time users would come to adopt the new format. While so far Zuckerberg seems to be right and users are adapting, this product change has the potential to weaken the unique competitive advantage Instagram does have: the strength of the social graph.

As mentioned before, competition can either be direct or for general time spent. In order to win on time spent though, Instagram is de-emphasizing its social graph, allowing pictures from your friend’s recent vacation and your family’s Christmas party to show up alongside a video of a random teen with 400 followers bobbing their head along to some random pop song. Adding in content recommendations to the feed is effectively displacing friend and follower content for “discovery” content. It is an admittance that what makes Instagram unique for each user with their customized list of hundreds or thousands of follows may not matter as much as showing them a captivating video they didn’t know they’d like. Today though, Reels in the Instagram feed is upsetting many users, not really because it seems to violate the essence of Instagram that is highly curated and deliberate, but more because the content isn’t that good, and the recommendations are lacking in accuracy.

If we think of the purpose of Facebook and Instagram, there is definitely an important aspect of connecting and socializing with people you know, but the larger consumer preference is often just to be entertained. People reach for Facebook and Instagram dozens of times a day, not because they crave an immediate connection with a friend, but rather because they are bored and it’s a semi-impulsive behavior. If you ask the question “What app do you go to first in the morning?”, most respondents will pick a messaging app like Snapchat or Instagram or Facebook. In this survey with ~275 American college students, we can see that 42% checked Snapchat first when they woke up and 34% went to Instagram first. However, only 13% went to TikTok first. This makes sense as for a majority of college students, they want to first “check-in” with their friends and social network before going to an app strictly for entertainment. This shows the unique position that a social network has over an entertainment app, even a fantastically addicting one.

However, if you ask the question, “What app do you use first when you are bored?”, then TikTok jumps to almost ~50% for these college students and Instagram falls to ~28%, which is far better than Snapchat’s 10%.

TikTok is a far superior product to entertain and stave off boredom than Instagram or Facebook, and the TikTok time spent numbers which are 2-3x higher than their nearest competitor prove it. Critically though, Instagram and Facebook still have a unique “socializing” and “connecting with friends” aspect, which gives them a stronger competitive position than just entertainment. When fighting the time spent battle, almost anything can be competition: Netflix viewing, Call of Duty playing, or even hanging out with your friends can be swapped out for TikTok. However, if you are trying to tell all of your friends and “not-so-much friends that you just tried Hailey Bieber’s Erewhon smoothie or you are dining at RH’s aesthetic NYC rooftop restaurant, there is a much tighter competitive set for that sort of behavior; really just Snapchat, Instagram, Facebook, and messaging apps, among which Instagram is usually a top choice for users.

While the need for this sort of sharing behavior may seem silly to some, it is highly unlikely it is going away entirely, and as long as it is here, no one is better positioned for it than Instagram. Instagram, and Facebook for an older generation, are not currently threatened for this semi-public sort of sharing, and that is not for many trying from their competitors. TikTok has tried to use their immense user base and large time spent to back into a social graph with the ostensible aim of eventually hosting more social networking-like behavior, but has had very little success thus far.

In order to increase the social aspects of the platform, TikTok is pushing users to follow friends and message in-app, but beyond maybe a couple of close best friends who a user shares TikToks with, there is very little messaging behavior between friends. They are attempting to increase activity through prompts to access a user’s contacts (they only need a couple people to grant permission to get a whole node of a friend network), and also through unique codes that are embedded in TikTok links that are shared off-platform. If you share a TikTok with a user over iMessage or WhatsApp, TikTok will know who clicked on it and can match it to who sent it, linking the two people. They can then suggest you follow each other next time you open up the app.

TikTok’s efforts aimed at building more in-app social functionality in hopes of eventually launching more social apps are not unfounded fears. TikTok’s parent company ByteDance has already launched several additional social apps in China (KeSong, an Instagram-like sharing app, and messaging app Feiliao are two of their more recent efforts). The same way Zuckerberg was paranoid that a messaging app with a social graph could develop into a social network, ByteDance could try to use their 1bn+ TikTok users to encroach on Facebook and Instagram’s core socializing and sharing behavior. The irony though, is that part of the reason why TikTok is doing so well is that it is totally separate from a social network.

In our research, an interesting contradiction kept surfacing. Users would claim that TikTok felt more “private”, while acknowledging that the views or likes they would get on TikTok videos were far more than what they would get on an Instagram post or Reel. This is because despite the fact that the average TikTok is shown to vastly more people, it is seldom shown to anyone you know. The friends that people have on TikTok tend to only be a handful of close friends they would share TikToks with on-platform through messaging. This sense of privacy has led to far more TikTok users feeling comfortable creating a TiKTok than Instagram users feel creating Reels. Many users like to experiment and make a silly TikTok, usually jumping on a popular video trend that is going viral, but they do not want that to be pushed to all of their friends and family members. There is an option to not make Reels public and only share with your friend network, but there isn’t an option to only share with the public excluding your friends. This has been an advantage for TikTok, since it spawns more creations that are housed in TikTok. This is emblematic of an advantage in one domain becoming a hindrance in another.

Nevertheless, despite the platform differences, the key takeaway is that Reels are net time additive to a user’s time spent. While Reels’ content and recommendations might not be as good as TikTok’s, and a lot of videos are simply repostings of TikToks with watermarks, the strategy seems to be working. Users are staying on the platform for longer than otherwise. And many influencers are creating Reels too, in part because Reels get pushed out to their follower base in the Instagram feed, ensuring fans see it, but also to hedge their bets should TikTok be banned.

The executive order was overturned, but lawmakers still debate banning the app.

In 2020, U.S. President Donald Trump tried to ban TikTok through an executive order, but it was overturned by a judge. Instead, they worked out a deal where American data would be housed in Oracle servers in the U.S. However, there are still concerns that a China-based ByteDance could access that American data and pass it on to the CCP at their behest (Chinese law requires companies to comply with all government data requests). Just in the past month ByteDance confessed that employees in China pulled data on journalists in the U.S., shattering any illusions that ByteDance (and by proxy the CCP) can’t access any user data they want on a whim. While these employees were fired and the press releases tried to suggest this was anonymous, the truth of TikTok’s data practices are more ambiguous, which is enough to suggest that the current arrangement is insufficient to meet data protection concerns.

Perhaps more concerning to U.S. lawmakers though, is that the algorithm that dictates what the minds of tens of millions of America’s youth consumes is overseen by a China-based company that could quietly tune the algorithm or censor content to buttress the CCP’s viewpoints. In one of the more public cases, TikTok removed a viral video a teen made to spread awareness of China’s treatment of Uyghurs in Xinjiang that was disguised as a make-up tutorial to avoid censorship. Whether intentional or not, it fed into the fears of having a widely popular media property controlled by a company domiciled in a country known for censorship. More recently, TikTok has been banned on U.S. government devices. But whether there is enough support for a full ban of the app remains to be seen. Banning TikTok, of course, would be a boon to Reels, who could now operate without such a fierce competitor.

Snapchat and YouTube each rolled out their own short-form video features called Spotlight and Shorts. Snapchat’s version hosts a lot of “raw” content that has lower production quality than a Reel or TikTok. YouTube Shorts are mostly short clips from full-length YouTube videos, and have less content that was made specifically to be a YouTube Short (but that could change). It is clear that TikTok is Reels’ main competition, and short of a TikTok ban, Meta is stuck using more prosaic tactics to spawn usage.

In addition to throwing Reels in the main feed, they are also paying some creators to make Reels. Meta launched a $1bn creator fund to grow Reels usage, but for most creators, their payment for views pales in comparison to what they can make from sponsorships or selling products. The amount of money needed to incentivize creators to stick to one platform would have to be a lot more. A revenue share war breaking out, with creators flocking to the platforms that pay the most, is a business risk. However, in contrast to the payouts that YouTube does, which are for when they insert ads into the videos, Reels/TikTok inserts videos in between videos so the creator cannot take credit for the ad in the same way. Additionally, the fact that they can recommend any video and users do not search for specific videos makes it less likely there will be the same size of payouts as on YouTube. But anything is possible.

Another (blurry) leaked slide shows that teens primary messaging app is Snapchat at ~25% and iMessage is at 20% with Instagram and Messenger next at ~14% and 13%. Android Message is just below 10%.

To sum up sharing competition, Facebook and Instagram are still the only platforms you can use to widely share with family and friends, but there are other messaging apps and group chats that are taking up some of this sharing. We didn’t mention WhatsApp much, but that is because it is absolutely dominant globally outside of North America where Apple’s iMessage is leading. WhatsApp is a primary messaging app for most of South America, Africa, and Europe, with a dominating presence in India and Indonesia among many other countries. WhatsApp doesn’t just have breadth, but also deep engagement with over 100bn messages sent on the app daily. Messenger usage is broad, but generally used in addition to another core messaging app. Messenger has about half the reach of WhatsApp, but an even lower comparative engagement rate. The only other competitive messaging apps are iMessage and Snapchat. iMessage’s lack of an Android app has limited adoption in any country where Apple isn’t the leading smartphone. Snapchat’s primary focus is on disappearing photo messages, but it is still a widely popular messaging app among younger people. However, they have had limited success in aging up and few users can rely on it as their only messaging app. User behavior may shift to push more sharing that typically was on Facebook or Instagram into direct messages or group chats, but Facebook and Instagram are still the only venues to broadly share with your social network (Twitter users seldom have many friends follow them and the vast majority usually don’t post content relating to their personal life. LinkedIn users mostly only post work-related content).

For time spent competition, the competitive set it broad, including everything from TV, video games, in-person events, and socializing, but the more relevant internet media businesses are YouTube, Snapchat, and TikTok, in addition to Facebook and Instagram. Facebook’s entertainment value comes from seeing your friends’ posts and a random assortment of other content that you follow or your friends liked, but it is generally considered the lowest in terms of entertainment. Instagram ranks higher with not only your Influencer follows to entertain you and a lot of friends’ Stories, but now more in-feed recommendations and Reels too. YouTube is still an addictive service with a strong recommendation engine and is a great place to find videos and podcasts of people you follow. While their position in short-form content is currently lagging, they are still dominant in all longer-form videos. Snapchat has some curated content in their discover tab and Spotlight, but it is still largely a messaging app. TikTok is the #1 app for endless entertainment. This is the social media landscape today.

Instagram Headquarters.

Usage Metrics.

Getting accurate time spent data across our core peer set is fraught with errors as companies seldom report such data and 3^rd party estimate methodologies can be flawed. Below, we combed through several different data sources to best estimate time spent across each app. However, differences in iOS vs Android, geography, sample selection bias, and tracking methods make it hard to have total confidence in the figures. Additionally, different data sources would count users according to varying methodology and thus report time spent figures on different scales. Having said that, we think the estimates below are directionally correct enough to inform the social media landscape.

As shown, TikTok has the highest time spent per DAU and MAU, followed by YouTube, Facebook, Instagram, Snapchat, and then Twitter. These average figures, which include a mix of many power users and many periodic users can skew the actual usage patterns of each platform. For instance, it is much more common for YouTube users to binge many videos at once and then go several days without watching anything. Facebook and Instagram users tend to have much more steady usage patterns, with many sessions daily, the majority of which will only be a couple minutes or less.

However, since Facebook and YouTube still have far more users, their annual time spent on each platform swamps TikTok’s.

Note that Instagram time spent lags to Facebook, which if you consider Facebook Blue to be more at risk of becoming irrelevant, than that is a critical risk to Meta if they cannot transition Facebook Blue time spend to Instagram overtime. Keep in mind though, part of the discrepancy is because Facebook has a lot more usage in Asia-Pacific and “Rest of World” than Instagram. This is in part because Facebook is older and thus has had more time to penetrate globally, but it is also because global growth in less developed countries wasn’t a push for the Instagram team until more recently. The higher data usage of a more visual app like Instagram also limited its global growth. “Facebook Lite”, with less bandwidth requirements, was created precisely to grow the app in some of the less wealthy parts of the world (as well as Zuckerberg’s Internet.org initiative). One of the more positive take-aways from the leaked Instagram usage slides was that they felt they had not saturated the market for Instagram and could continue to grow internationally. Increasing geographic penetration could eventually allow Instagram to match Facebook’s scale. However, as of now, time spent on Facebook is still higher than Instagram.

Facebook’s Internet.org initiative aimed to bring internet to areas that lacked it.

Despite the time spent difference, the revenue generated from each app is likely much closer to parity as Instagram has high penetration in North America and Europe, where users are worth the most. North America in specific is their most valuable market by far, with an average user worth ~$200 according to their ARPU calculation. However, Meta’s ARPU calculation is misleading because it attributes the company’s total revenues to just Facebook users and doesn’t back out Instagram, Messenger, or WhatsApp revenues in the calculations shown below.

This slide shows up every quarter, but it is misleading. Facebook ARPU is not average revenue per Facebook user, but rather total revenue Meta generates across all products divided by Facebook users.

According to our time spent figures above (applied proportionally, despite Instagram time spent globally under-indexing compared to just North America), it suggest that just Facebook ARPU could be about 30% less, or $140 per user. If time spent on Instagram and Facebook within the US is more evenly split, then it could be closer to an ARPU of ~$100 each. Facebook has been monetizing for longer than Instagram though, and generally has higher ad load, especially as Instagram surfaces like Reels (and Stories until recently) monetize at lower rates. Either way, a North American user is worth ~ 2.5x times more than a European user and ~4x more than their average global user. A closing of this wide ARPU gap between geographies would be a simple, but very powerful, revenue driver.

Since the mid-2010s, Facebook started getting a reputation for being for old people whereas Instagram and Snapchat were the platforms for young people. Meta never had to worry that much because they own Instagram, and so the theory was that young users would eventually “graduate” to Facebook. This sentiment is captured by the leaked Meta slide below which shows a slowing rate of users adopting Facebook once they are 19-20. Facebook would penetrate prior cohorts of users born before 2000 at a ~90% rate by time they were 19. However, for the past several years, penetration has been dropping—a very concerning trend.

We can see that among those over 30 though, Facebook has very steady engagement with almost an hour of daily usage. However, younger users have lower and falling usage.  

Leaked Facebook slide on concerning trends

Below, the Meta researchers concluded with a somewhat positive spin. At the current trajectory, Facebook is positioned to reach saturation of those born after 2000 by the time they reach 24-26 (~5 years later than previously). However, the risk of course, is that these trends continue, and Facebook sign-ups drop off at an increasing rate as fewer peers in each cohort adopt the app.

The hope is that as each individual moves through their life stages and starts to mature, they will want to join Facebook so they can share and connect with everyone from their grandparents and parents to distant friends. The trends, though, suggest that this thesis may be idealistic and Facebook usage could continue to drop. Speaking with American teens and young adults, their impression of Facebook is even more harsh than the numbers suggest. Most of them don’t see any point to it and consider being on it borderline weird. Their impressions can of course change over time, but it is not confidence instilling. In 2018, just half of teens used Facebook versus >70% just 4 years earlier in 2014. The usage figures have continued to drop, and are an estimated ~30% today.

Having said that, it seems to be a widespread assumption among many investors that Facebook will eventually become irrelevant, and Instagram will continue to capture the younger demographic for now until the next “new thing” comes out, which many worry could be TikTok. This is why the Instagram leak, which showed negative trends of weakening production and consumption were so concerning; it supported the point that Instagram’s life may be starting to reach a decline. Instagram is supposed to be the younger, stronger platform that can pick up the slack as Facebook drags into maturity and until Meta figures out the next new app to create or acquire (of course, this is somewhat naïve as their track-record of spinning up apps is woefully poor, and it seems unlikely they will be able to acquire anything anytime soon). The leak emboldened concerns that all of these platforms have a relatively short life and something new can replace Instagram soon. However, this narrative is fallacious, because as we noted, there is no platform that is positioned to usurp Instagram or Facebook and compete on their direct functionality. The bigger concern isn’t that someone replaces Instagram or Facebook, but rather that users decide they don’t need them in the first place (or use them far less).

For now though, sentiment seems to be more negative and anecdote-driven than the actual numbers have borne out. There have been calls for Facebook’s eminent demise for the better part of a decade, but few, if any, figures support that claim. Facebook was founded in 2004, and ~20 years later it is at its all-time high in usage. Facebook adoption of the next generation is slowing in the US, but not globally. Facebook DAUs and MAUs are at all-time highs, including in North America. Time spent on Facebook for older users is steady. Instagram time spent did take a small hit from TikTok, but they are addressing that with Reels and that has led to time spent growing. Things are not perfect, and there are definitely threats to usage, but by and large Facebook and Instagram have proven to be very resilient. If the concern is that Instagram will eventually go the way Facebook has, perhaps that is a good thing as the lifetime of a social network is more than people seem to appreciate.

While competition for usage is largely steady (albeit with risks), they continue to wallop the competition for advertising dollars.

Advertising with Facebook.

The simple Meta thesis is that as long as they continue to have steady engagement, they will be able to monetize that time spent through advertising. The thesis has held up well since Sheryl Sandberg joined, with Meta growing ad revenue over 100x since 2009, from under $1bn to over $115bn in just ~12 years. This has generally led to complacency when considering Meta’s ability to monetize time spent because they have been so good at it historically. With the digital ad market much more mature now, competitors getting better, and an increasing focus on data privacy, Meta may no longer be able to monetize time spent as effectively as they have previously. In this section, we will dovetail into advertising with Facebook/Instagram and other potential advertising options. We will then conclude this section with different risks to their advertising model including recent Apple Privacy changes which have materially impacted their ability to target and track advertisements.

Instagram in-feed advertisement shown to the left with the product listings displayed when you click on it (middle). You can then checkout in-app (right), or more commonly, in a pop-up webpage that opens to the checkout screen.

Big picture, there are five classes of digital advertising: general search (Google, Bing), vertical search (Amazon, Yelp), social media (Facebook, Instagram, Snapchat, Twitter, Pinterest, YouTube, TikTok), open web/publishers (blogs, free content websites like Food.com and publishers such as NYT, Scientific American, etc.), and connected TV (Hulu, Peacock, etc.). While to a certain extent all advertising is competitive against other forms of advertising, these advertising venues are most competitive with each other.

Advertisers weigh various considerations when deciding where to allocate their ad budget. They want to place their budgets where they expect to get the best return, but that return might not always be measurable, scalable, or brand-friendly.

Measurement: Measurement pertains to the ability for an advertiser to measure what they received for their ad spend, whether that be sales, website visits, sign-ups, or whatever they are tracking. Most platforms offer their own measurement solutions, but there are also a lot of 3^rd party tools. Advertisers are typically torn because the platforms tools tend to be superior with more data and ability to give them an accurate answer, but the idea of them “grading their own homework” (common AdTech parlance) is off-putting. It is common for multiple platforms to take credit for a sale, so advertisers have trouble fully reconciling the efficacy of their spend. In the more benign scenarios, it is a result of the user having visited multiple platforms (perhaps the same user searched up a product on Google a day after clicking on it on Instagram) and more perniciously, as the result of erroneous modeling (with privacy efforts limiting ad-tracking, most conversions are modeled on a probabilistic basis instead of deterministic, opening up the door for errors).

However, these issues are more on the margin. Google’s and Meta’s advertising efficacy is (on average) obvious enough that most advertisers see the positive impact, but it could hurt their ability to convince an advertiser to grow their budget with them. But this is an even bigger problem for newer platforms who are still on an “exploratory” or “experimental” budget with the idea being that they must prove themselves first before becoming part of the “always-on” or regular advertising budget. For many of the advertisers we spoke with, Snapchat and Pinterest could never get beyond that phase. So when ad budgets were cut, they were the first to go.

As much as measurement matters to prove the advertising is working, it is nowhere near as important as the ads actually working. There were several advertisers we spoke with who pulled their Facebook/Instagram budgets in the wake of the Apple privacy changes because they thought their ad spend was deteriorating and they saw their sales plummet. So while they can’t monitor their advertising as well as before, they can’t turn it off either because it will adversely impact sales. Again though, this will make it harder for them to justify increasing ad budgets until the measurement is restored.

Scalable: This relates to the ability for an advertiser to deploy as much of their budget as they wish. For instance, an advertiser may be getting a great ROAS on their first $5,000 of spend, but when they step it up to $50,000 it quickly deteriorates. This is because a platform has to have a huge audience with a lot of usage in order to make sure the right ad is always being shown to the right person. Platforms with smaller scale like Snapchat struggle here compared to Facebook. (Also remember that usage doesn’t necessarily translate into activities that can be shown ads against. Snapchat’s main draw is messaging, which cannot easily be monetized directly with ads). Generally, advertisers do not run into this problem on Facebook until their ad spend reaches multiple millions.

Brand-Friendly. This is concerning the content that an advertisement will be shown next to. Brands do not want to be associated with negative or hateful content for fear consumers will associate their brands with that content subconsciously or that it implies their brand supports that negative message. Facebook and Instagram have the best systems to detect such negative content with the help of advanced artificial intelligence and a human moderator force of over 40,000. However, these platforms also tend to draw the most negative content and hate speech. Twitter is up there too, but with far worse moderation abilities and no sizable human moderation force (Twitter though, tends to prefer to censor anodyne content than risk not censoring something that should be). Snapchat’s closed platform, and Pinterest with mostly inspirational/aesthetic content, allows both to rank highly in brand safety. Google has to work hard to make sure the ads they show across the entire web (3^rd party ad network business) do not show up alongside unsavory content, but they commonly do nevertheless (and that’s not because they don’t have very advanced AI, but rather speaks to the complexity of the task). Thankfully for Google though, they usually don’t get the same negative publicity for it (most people don’t even know that Google placed the ad on that random 3^rd party website).

In addition to these factors, they also will have to consider the ease of using each platform as well as what their ultimate ROAS will be. Most advertisers have little issue signing up for and using Facebook ads, which is why they have over 10mn advertisers (their ad platform is self-serve, with most advertisers on-boarding themselves without assistance). TikTok may prove more complicated in this regard because in addition to figuring out a new platform, they would need to make new ads in the short form video format. Whereas a Facebook advertiser could grab some existing content from their product list page and turn that into an ad, a TikTok advertiser will have to find people to act in a video as well as make it worth watching so it doesn’t get skipped (if you do this well enough though, you don’t even need to pay for advertising as the algorithm can share your video with millions if its popular).

Facebook ad manager interface allows you to monitor all of your campaigns.

Meta’s main focus is on demand generation. Advertisers on Google Search and Amazon are paying to move up in search rankings when a user typically already has an intention to purchase something. These advertisers aren’t trying to convince the user that they need a product, but rather sell their product to a user already in need. Meta’s advertising, in contrast, is more focused on convincing someone to buy a product they had no idea they wanted (DR). While they also have an open web/publisher ad network (this is FAN), Meta has mostly focused on advertising on their own platform because the returns for advertisers are much higher there with superior engagement and click-through rates). In DR advertising, the social media companies mentioned above are Meta’s main competition. For digital brand advertising, it is the social media companies plus connected TV and the open web/publishers.

For digital ads, ROAS is ultimately what will drive advertisers’ allocations to each channel. In speaking with many D2C advertisers, Google Search and Facebook/Instagram are the top channels, often taking up to 75% of the ad spend (for online born D2C companies Facebook is a critical channel). Amazon is also getting a growing portion of budget, but many brand advertisers want to build up their business outside of Amazon because depending on them could be an existential business risk.

Meta, with their ~3bn users across the globe and vast repository of user data, is best positioned to personalize advertisements to each individual that they resonate with. Not only does Meta know what you are interested in from your follows and likes, but they also know your friends and what they are interested in as well. While the social graph might not be that helpful in figuring out what content a user would like best, it could be more useful in picking ads for that user. Meta’s ads are so accurate that people have wondered if they are listening to their conversations (they aren’t), but it is likely that if you are thinking about buying a ColourPop (popular brand on Instagram) liquid lipstick, that one of your friends already has, and Instagram knows that, so they show you that ad. Meta’s ad matching algorithm is second to none, and it is hard for any other player to come close. (It is curious that their ad recommendation engine is so superior while their content recommendation engine has historically been so mediocre).

ColourPop is a liquid lipstick brand that has found great success advertising via Instagram.

While return on ad spend (ROAS) varies by advertiser and product, for years ROAS that was in excess of 5x was common with some advertisers much higher. However, as more advertisers joined (> 10mn currently), bid density for each auction has swelled, resulting in much more temperate returns, with 2-3x seeming to be the most common (again, wildly varies by advertiser and vertical though).

Other platforms like Snapchat, TikTok, YouTube, and Pinterest do not have to match Meta’s ability to match ads to people as their ad CPMs are far cheaper, making up for the worse targeting. New ad inventory from connected TV (every streaming service seems to be increasingly relying on advertising to grow) could drastically weaken prices for brand ads across the digital ad ecosystem. TikTok is also still in its early innings of rolling out their advertising efforts and has been fairly slow in signing up advertisers. However, they have a ton of usage which means that they have a ton of potential ad slots, which could weigh on digital ad pricing when they comes online (more on this in the risk section). Meta’s advanced ad targeting has allowed them to charge much higher CPMs versus competition (3-4x higher by some counts), but competitors could take share from them just by being much cheaper. Meta’s reliance on highly-tuned ad targeting to justify high CPMs also makes them much more susceptible to anything that impacts that ability to target.

Data Privacy.

The AdTech industry is notorious for seedy practices involving creepy levels of data gathering and clumsy or nonexistent data protection practices, with many players happy to sell data to whomever pays. There were few laws that governed digital data privacy until recently, and most players seemed fine doing whatever was technically feasible in order to target ads. Facebook didn’t invent any of these practices, but their vast collection of personal user information coupled with open access for developers to share in that data certainly didn’t help give them a good reputation for privacy.

Facebook’s trove of data would help them create the most accurately targeted ads of any company ever before, but it would also cost them consumer affinity when data breaches and scandals repeatedly plagued user confidence. They would learn their lesson from their carelessness and drastically revamp their data collection operations. However, that was not before significant lapses in data policy stoked the ire of various regulators and ultimately became the impetus for privacy laws in EU and elsewhere.

The EU passed GDPR (General Data Protection Regulation) in 2018, which provides legal guidelines for collecting and processing personal information of individuals. The same year, California would pass CCPA (California Consumer Privacy Act), which would allow citizens to better control the data companies collect on them.

Despite regulation coming much later, Facebook would not escape a slew of fines for misconduct. The FTC sued Facebook for misleading data privacy practices and imposed a $5bn fine on them in 2019. They would also settle $650mn for violating the Illinois Biometric Information Privacy Act in 2020. Most recently, they were fined ~$275mn for violating GDPR and they just settled the Cambridge Analytics lawsuit for $725mn. These are just some of the larger settlements with dozens of smaller ones.

Facebook was fined a record-breaking $5bn by the FTC, almost 20x more than the largest privacy or data security penalty ever imposed worldwide.

While Facebook might be targeted more than anyone (Google is also up there), they now have one of the most secure platforms in terms of data privacy. This isn’t just because of various regulations, but also because stronger data security makes their advertising more differentiated as it gives Meta guise to refuse access to their large database of user profiles.

Meta is a “walled garden”, which means that all of the activity on their platforms stays within their company. They limit passing user-level data back to advertisers not just because they are no longer allowed to, but because it would also increase the advertiser’s reliance on Facebook. Advertisers today spend on Facebook despite having no ability to actually identify who all of their ads are being shown to (the way Facebook uses machine learning means they don’t readily know either). Advertisers will still receive any data from their own sites, of course (such as if a user inputs their email on their website or actually purchases something), but they have less ability to know who is seeing their ads. This not only makes it more privacy-friendly but makes it harder for the advertiser to leave Facebook. They cannot take the targeting that Facebook does and try to replicate that audience on a separate platform. (Ironically, these same data privacy rules would weaken Meta’s competition’s ability to compete, while they are simultaneously investigated for anti-competitive behaviors. Other regulations on content moderation could also de facto increase Meta’s competitive position as sub-scale platforms struggle to comply).

While data privacy regulation has by and large increased the moat to become a scaled social media platform because of the needs to comply with a myriad of rules, and in some cases physically house data in local servers within a country’s borders, Google’s and particularly Apple’s privacy initiatives have been very economically destructive.

Apple has embarked on a strategy of making privacy a premium product. To do that, they have been the most draconian with data permissions, but have also misrepresented the reality of current data management practices.

In 2017, Apple rolled out ITP (intelligent tracking prevention), which basically disallowed 3^rd party cookies from pretending to be 1^st party cookies. While this was uneventful to Facebook, it was just the beginning of several moves to take more control of data that is collected on an iPhone. They would make a few other privacy changes, but the next most meaningful one came years later in 2021 with the deprecation of IDFA.

IDFA would be particularly harmful to Meta’s app download business.

IDFA, short for “identification for advertiser”, was a unique, anonymous ID number that advertisers could use to track users across apps. Apple originally released it in 2012 to enable more privacy-centric advertising (versus the previous Unique Device Identifiers and MAC Address method). If you clicked a “download this gaming app” ad on Facebook, downloaded the app, and later purchased an in-app item, all of that info was tied to the IDFA. This would allow the advertiser and Facebook to know which ads worked and gather “event stream” data of in-app usage.

Event stream data is everything from whether the user completed the 1^st level of a game to how long they played to what they purchased in app to anything else they want to gather. This event stream data is particularly important for app advertising since most apps are free to download and the developer tries to sell the user on something down the line. Limiting this event stream data meant that Facebook wouldn’t know what happened after the user clicked the app ad that sent them to the app store; they wouldn’t know if they successfully downloaded and signed up for the ad, if the user played the game, or if they bought anything. This would make it much harder for Facebook to tell whether an ad was working and what an advertiser’s ROAS was.

Apple went even further with ATT in 2022, which was not just the implementation of removing the IDFA, making it “opt-in” instead of “opt-out” as it previously was, but also a broader framework for what sort of advertising would be allowed. Not only would most apps lose the ability to access the IDFA when the user opted out, but other means of tracking users across apps with emails or phone numbers were prohibited by ATT. This is despite the fact that Apple has no technical way to prevent a developer from doing so on their backend, but Apple would implicitly threaten App removal from the Apple App Store if they were caught. ATT was a big shift in the advertising landscape because Apple wasn’t just playing in the AdTech arms race to develop privacy solutions before others could figure out work arounds, they were now setting principles, the violation of which would be policed by swift removal from the app store and thus causing a developer to lose access to their most lucrative audience.

Facebook’s ATT required prompt to ask users if they want to be tracked.

If a user opts-out of tracking, then in place of the full event stream they previously received, the developer would just get limited info from Apple’s SKAd Network. Without going into too much detail, the SKAd Network was originally almost comically unhelpful. Apple would originally allow only a one 6-bit postback (data sent from the app back to the ad network), and Apple utilized an unhelpful randomized timing delay for more “privacy”. They would also limit campaign IDs and not tell you if they aggregated data across different campaigns. What all of this means, effectively, is that a lot of the optimizations advertisers embarked on for years to find their customers through constantly iterating on ad campaigns would no longer be possible. Advertisers would try a lot of work arounds to make up for the lost signal, including using look-alike audiences for those that did opt-in, but none of it was good enough to restore the lost signal.

Apple’s SKAd Network

It was originally thought that ATT would only impact app-to-app advertising, but Apple clarified that it would also regulate app-to-web campaigns as well. This means that when you click an ad in an app and it opens a website in Safari, the developer would still be expected to abide by ATT. This matters because it was originally thought that Facebook could just use their pixel (and later Conversion API, which is housed entirely on the server side) to send data back on the backend, also called server-to-server. While this is technically possible, it would violate Apple’s ATT, and so instead Meta has moved to an anonymous and aggregated means of collecting web data called Aggregated Event Measurement. This would limit the data Meta received to aggregated data sets with no ability to identify individuals. Meta now can receive any information an advertiser wishes to share (which is usually all of it), but it must be at an aggregate level. Because of this though, they still lose a lot of that event stream data at the individual level that they could have used to retarget ads to individuals.

The move to more aggregated data sets is why Meta came out saying Apple’s privacy changes would be bad for small businesses. A small business simply will not have the same level of customer traffic or ad spent to gather enough data for Meta’s algorithms to work optimally. Facebook’s and Instagram’s advertising muscle was built on the backs of 10mn small businesses, but they will now be more disadvantaged in ad auctions as their ability to measure ROAS and find their customers will be impaired.

While Zuckerberg received criticism for this being a self-serving argument, it is by and large accurate.

Our channel checks, while varying in degree, clearly showed deterioration in ROAS since ATT was implemented. Of course, confounding factors of macro and a comp including the Covid ecommerce boom with stimulus checks make it an imperfect extrapolation to gauge ATT’s damage. We have heard from larger advertisers that they were by and large able to quickly recover at least some of the lost signal and measurement, whereas many small businesses haven’t.

These changes hit Meta hard. In 4Q21, they estimated that ATT would adversely impact revenues to the tune of $10bn, or about a 9% headwind to advertising revenue. Other advertising platforms would be hurt too, but Meta was impacted the most. That is because Meta was more optimized than any platform and so the loss of data signal wrecked their precise algorithm. In contrast, Twitter, who is notorious for having struggled with direct response and targeted ads, barely registered an impact. This is because Twitter advertisers were paying ad prices that assumed poor targeting, whereas Meta advertisers were paying CPMs that assumed highly targeted ads.

Meta is working on restoring lost signal from ATT, and in addition to pushing advertisers to install server-side CAPI (conversion API), they also are rolling out other optimization tools like Advantage+ and a partnership with Shopify (Shopify Audiences helps the millions of Shopify stores to advertise more effectively on Facebook). None of these are a panacea, but each individually helps restore some of the lost signal and moves Meta closer back to where it was before in terms of targeting and measurement.

Another outcome of the privacy regulations is that it makes Meta a better advertising option relative to other options (other than Google who is potentially strengthened by these changes). Whereas TikTok and Snapchat had some momentum with direct response, ATT was a formidable step back for them and it is questionable whether they will ever be able to close the AdTech gap in terms of targeting and measurement relative to Meta. However, while Meta might be in a better position relative to other platforms, but they are still net in a much worse position than pre-ATT.

There is a good possibility that the lost data signal means their ad targeting abilities are permanently impaired. Now, that doesn’t mean that they can’t continue to improve and get to where they were pre-ATT, but just that in no way are they better off from having access to less data.

Even though they may be relatively stronger than alternatives, it doesn’t matter because advertisers look at total return on ad spend, not relative return on ad spend (lose less money advertising with us will never be a great selling point). However, it will be hard to know just how adversely impacted Meta is for some time as macro factors potentially hide the true extent of the damage.

Meta’s efforts to restore lost signal goes beyond algorithmic changes, APIs, and SDKs. They are trying to bring more activity onto their platform with Shops (enables merchants to offer checkout right on Facebook and Instagram so Meta can easily track the conversion) and embarking on a costly AI endeavor with ~$30bn in capex in 2022 alone to build out the compute needed to run complex AI models. We will pick up on their growing capex spend in the ROIC section.

The takeaway is that for virtually every advertiser, ROAS has gone down over the past year, but how much of that is attributable to ATT or macro conditions is unclear. Several initiatives like CAPI, Advantage+, and Shopify Audience, among others, are slowly repairing ROAS measurement. For more longer-term solutions, Meta is increasingly turning to powerful AI models which require spending prodigiously on new clusters of GPUs. So far, the trends are boding alright for larger advertisers with some measurement and targeting improvements in the past year since ATT, but whether the smallest advertisers will be as big of a piece of Meta’s business in the future as in the past is debatable.

Being susceptible to the latest whims of Apple (and Google) only clarified Zuckerberg’s resolve to control the platform that his applications sit on. He would strive to create the next computing platform not only so Facebook and Instagram wouldn’t become relegated to irrelevance in the shuffle (as other popular desktop services had), but so Meta can insulate itself from the capricious initiatives of the gate keepers.

Steve Job introduced the iPhone in 2007, which marked the beginning of the mobile transformation. Zuckerberg is trying to usher in the next compute paradigm shift.

Metaverse.

The first signs of Meta’s interest in virtual reality were revealed in 2014 with the acquisition of Oculus for $2bn. Oculus was just a 2-year-old start-up at the time of the acquisition, but they nevertheless had the industry-leading consumer virtual reality headset with the Oculus Rift. Zuckerberg would use the Oculus acquisition as the launching pad for his broader ambitions to develop the next big paradigm shift in computing. 

Early Oculus Rift VR Headset

The same way mobile phones changed our relationship with technology and fundamentally how we interact with the world, Zuckerberg sees virtual reality (VR) and augmented reality (AR) as being the next step in the evolution. The leaked email below is from an attempted acquisition of the game engine Unity. In it, Zuckerberg lays out his vision for the future of VR/AR. He notes that “VR/AR will be the next major computing platform after mobile in about 10 years”. He comments that it can be more economical since it can replace many other items you buy, like a TV, desktop computer, or phone, and could be even more ubiquitous than the mobile phone.

Zuckerberg’s Unity acquisition rationale laid out in a leaked email from 2015.

AR (augmented reality) combines a digital world with the real world, usually overlaying digital images on clear glasses or a live feed of the world. VR (virtual reality) in contrast is an entirely digital world. A user would put on a headset that simulates an entirely digital world.

These glasses are overlaid with digital images to “augment” reality.

With augmented reality, you are adding virtual elements layered on top of the real world. In virtual reality, you are creating entirely new worlds that only exist digitally. The Metaverse is a sort of “universal” virtual world.

For example, if you put on a device and saw just a virtual book on your desk with everything else the same, that is augmented reality. If you change the room you are in digitally to represent a 16^th century library, then that is virtual reality. If you can virtually pick up your book and walk out of that digital 16^th century library into a much bigger world connected with other people, then that is a digital world. If you can walk out of that digital world into other worlds, then that is a metaverse. A metaverse is a virtual universe that connects people and different virtual spaces together in a persistent and uniform manner. The same way Disneyland has different themed grounds within it like Adventureland, Mainstreet USA, and Star Wars Galaxy’s Edge, but all of them are still considered part of “Disneyland”, the metaverse can have a variety of different worlds. They are ultimately all interconnected in a single “Metaverse”. (Technically, the Metaverse doesn’t have to be accessed through virtual reality, but it is commonly thought of that way). 

Mark Zuckerberg demo’s the Metaverse with their virtual reality headset.

In the photo above, Mark Zuckerberg demonstrates how every person will have their own avatar(s) in the metaverse that they use to explore and play in this digital world. When a user puts on the headset, the screen enwraps the user’s full field of vision, making them feel totally engulfed in the digital world. If a user turns around, what they are shown changes to what was behind them in the digital world, making it feel like you are in this virtual world.

Meta Quest Controllers.

A user has individual controllers for each hand that they use to move in virtual reality. Infrared LEDs are used in combination with sensors and magnetometers, gyroscopes, and accelerometers in order track a person’s movement and translate it to their digital avatar. Joysticks and buttons help supplement actions like walking and picking up items, but future versions of the hardware could potentially have more natural ways to capture such actions (there is a big focus to capture the subtle movements on someone’s wrists as a means of accurately simulating hand and finger movements).

Infrared LEDs are used in combination with sensors and magnetometers, gyroscopes, and accelerometers in order track someone in virtual reality.

The advantages of the metaverse are 1) being able to create experiences that could never have existed otherwise, 2) allow more people to experience something they never could in the real world,and 3) connect people in more ways. In a totally a digital world, developers can create all sorts of experiences, like setting a philosophy lecture in an ancient Greek university class or let a user fly a spaceship through a model solar system to make education more interactive. Travel can be prohibitively expensive for many, but they could at least virtually visit Paris’ Louvre and Egypt’s Great Pyramids. If someone can’t visit home for the holidays, they could throw on their headset and talk to their parents in a digital recreation of a living room, giving them more of a sense of presence than a call or video chat. (The sense of presence that is created when you are in the same digital place—even when the graphics quality is low—is one of the big advantages of VR. It may sound gimmicky, but when you are actually with someone in VR, your mind doesn’t seem to know the difference and you feel like you are really “there” with them in a much more visceral way than how a video call feels).

Watching a movie in the metaverse (top left) could mean you feel like you are in a theatre when in reality you are at home. Someone can attend a virtual concert (bottom right) with their virtual friends as a cheaper and easier way to socialize.

While video gaming and entertainment are still the principle use cases of virtual reality today, Zuckerberg believes that will change with people working and collaborating in the Metaverse. Meta launched Horizon Workrooms and has partnered with Microsoft Teams and Zoom to connect those real-world apps into virtual reality. Meeting attendees can join a meeting per usual with a video camera or virtually with their digital avatar. The advantages of this may seem specious, but users report they are more focused in a meeting and generally feel more present in VR versus joining via a desktop app.

Microsoft Teams in Meta’s Horizon Workroom.

To help facilitate the push to more work use cases, Meta rolled out the Pro headset for $1,500 (versus $400 for the Quest 2). The Quest Pro, though, is a “mixed reality” device, which means it is designed to blend the real world and virtual world, allowing a user to be in augmented reality and then switch to virtual reality as desired.

Meta Quest Pro is a mixed reality headset.

The idea is that augmented reality would enable a worker to virtualize a more dynamic workspace, whether that be large monitors or the ability to digitally see a project. An architect can virtually create a blueprint for their building and move through it before actually putting a shovel in the ground. An industrial designer could test the proposed dimensions of their product by digitally using it.

A virtual workspace. Augmented reality on the left and virtual reality on the right.

The ultimate aim is for the metaverse to be enmeshed in our everyday lives, blurring the boundaries between the digital and physical world. With so many different experiences that could be built, Meta is first focused on building the hardware and software that will enable the next generation of developers to make their own creations. To help spawn innovation, they have rolled out several of their own apps, one of their biggest of which is Horizon Worlds. This app makes it easy for people to build games, buildings, worlds, and other experiences inside of Meta’s Horizon Worlds.

Meta Horizon Worlds.

If this all sounds costly, that’s because it is. While they only recently started breaking out the expense of their metaverse ambitions in their Reality Labs segment, we estimate they have spent around $50bn on their efforts so far (against <$8bn in revenues). Zuckerberg has noted that they expect Reality Labs losses to grow significantly year over year in 2023, but beyond 2023 it should be more contained, allowing for them to grow operating income.

If losses grow to $15bn in 2023 and they stay at that level for the next 5 years, that will amount to a total of $75bn more in Reality Lab losses. Including just the past 3 years where they started segmenting it separately, that comes out to just over $100 bn. It’s impossible to know what something like the “Metaverse” will look like in the future, but we will try to size up different potential outcomes to at least give an investor a framework for thinking about possible returns. 

Metaverse Monetization.

So far, Meta has sold an estimated ~15mn headsets, which is respectable considering they are still in the early innings of developing this form of technology. Meta has plenty of hurdles to overcome (motion sickness, latency, bandwidth, interoperability, device weight, thermal dissipation, in addition to creating compelling experiences) before it becomes mainstream, but instead of trying to troubleshoot how they can solve these issues, we will just acknowledge that there is risk they don’t figure it all out and it weighs on the experience and potentially adoption. Some investors may assume their incredible level of spend must yield answers to these problems and others may understandably be inclined to write-off the total investment until there is further evidence of their efforts bearing fruit. We will focus more on trying to frame the size of the potential opportunity and it is up to the investor to decide what they want to pay for it (more in the valuation section on this).

Meta plans to make money from the Metaverse mostly in two ways: 1) a 30% platform fee on all transactions, and 2) advertising. Similar to Apple’s app store, Meta will charge all developers 30% for all on-platform transactions. However, if a developer builds within one of Meta’s pre-existing apps, like Horizon Worlds, then Horizon charges a separate 25% platform fee. So if an a game developer sells their app for $30 in the Quest Store, they would keep $21 on it. However, if they built a game within Meta’s Horizon Worlds, then they would keep 52.5% (25% take-rate applied after the 30% Quest Fee). While this may seem excessive (especially for a company that is criticizing Apple’s App store fees), YouTube currently gives 55% of its revenues to creators and Meta’s investment to create the Metaverse swamps YouTube’s by leaps. (Having said that, the high fees have not done wonders for Meta’s PR, especially as they publicly attack Apple).

The advertising piece is more speculative as what exactly you are buying and how useful it will be to convert a sale is unclear. If the internet enabled direct response advertising, the metaverse could enable a new class of advertising as well. We can look at some existing platforms, like Roblox, to get an idea of how this could work. Roblox is an online, cross-platform video game where users play as a block-like character. A big draw of the game itself is building stuff within the Roblox platform and inviting your friends to share in your creations.

Roblox gameplay.

Roblox currently has ~50mn DAUs, most of whom are young kids. The ability to build within a shared world that is persistent across different devices gives it some metaverse-like aspects, but the conception of The Metaverse (capital M) is typically more all-encompassing. (Roblox could be one of the many different worlds to play in in The Metaverse. Some would disagree with this definition though). Either way, they were early adopters of enabling digital spaces for users to interact with and gave many people the first glimpse of what a “metaverse” could look like.

Below are several digital (usually referred to as metaverse) experiences where brands created their own spaces within the game. Typically, the brands would create special items for the users to buy and the game developer would charge a take-rate on that as well. While that is principally how the game developer charges today, we can think of other ways to monetize including charging for space in their digital world or to import a brand logo. We can also imagine an advertiser paying to advertise within the game either for the digital experience, the brand, or perhaps a real-world product.

Examples of brand in the metaverse.

We can easily imagine popular areas of a game or experience posting ads within it or sponsoring experiences. Each advertisement would be personalized to the user and could be far more interactive, like offering them the ability to win a limited edition virtual item if they play a game or let them virtually try on clothes that can then be shipped to them. (Snapchat already has this function in their app today).

Snapchat’s augmented reality capabilities lets a user try on Nike Shoes virtually.

Ultimately, it would be very hard to size up the potential number of advertisements and what the willingness to pay could be for virtual reality sponsorships. If we think about movie product promotions, like Heineken’s 2012 James Bond Skyfall deal for $45mn, they essentially paid a >$250 CPM (based on movie theatre ticket sales). Of course, many people watched Skyfall out of theatres, and more than once, and there is a social proof aspect that is valuable to advertisers, which all makes that CPM rate unrealistic, but clearly brands are looking for novel ways to get their products in front of large audiences and the metaverse is brimming with such opportunities.

Heineken paid $45mn to have James Bond drink Heineken.

If we look at estimated monetization per minute on the main social media platforms, we see that they monetize anywhere from $0.14 per hour on the high end to $0.06 on the low end. It’s not implausible to assume users spending time in a world that shows ads could be monetized at a similar rate.

In a best-case scenario, Meta is able to get both a thriving ad business with a “Meta Quest Ad Network” and also take a 30% platform take-rate on all metaverse transactions. How much current video game users spend on average varies from a hundred dollars to over a thousand. If you consider the metaverse a much larger market than just video gaming, how much each user spends could be much more. We will sensitize around various assumptions and tie it into the time spent analysis above to get revenue build in the valuation section.

In short, Meta has a huge call option with Reality Lab’s efforts to build a VR/AR business and The Metaverse, but that option carries a hefty premium to the tune of $15bn per annum. Whether Zuckerberg will ultimately be able to achieve his goals of creating The Metaverse (or even getting a positive return on invested capital) is up in the air. While the idea of virtual and augmented reality has been written about for several decades in early sci-fi novels, it seems like there is finally enough focus and resources dedicated to it that there is a good chance virtual reality devices are widespread within the decade.

Of course, whether there is a metaverse and widespread adoption of virtual reality, and whether Meta’s hardware and software are the winners, are different questions. Apple has secretly been working on two separate headset devices for virtual reality and mixed reality, but have yet to launch anything. While Apple is known for having incredible hardware and operating systems, and their ability to tie it into their Apple ecosystem gives them an edge, the market is likely big enough for more than one player.

Apple known for product development secrecy, is quietly working on their own virtual reality devices.

Apple, though, will focus on a device that fully integrates with their other Apple devices and their closed operating system has historically been complemented by a dominant open one (Windows for PC and Android for Google). Meta could potentially make that open OS (there have been reports that they disbanded their efforts to create a new OS), but for now they use a modified version of Android. (Almost definitionally, a Metaverse would have to be cross-platform and work on all sorts of devices, which could be awkward for Apple who traditionally only made apps for their iOS users. This could mean there is more than one Metaverse. But the more popular Metaverse platform, which traditionally has been the open one, would likely have a network benefit simply because more people are in it.

Meta spending >$15bn on just the Metaverse compares to Apple’s total annual R&D of ~$26bn last fiscal year across all projects. It seems highly doubtful that Apple is spending anywhere near >50% of their R&D budget on VR and AR technologies, but they don’t necessarily have to match Meta’s spend. (Google also relaunched their virtual/augmented reality initiatives, and they spend over $35bn on R&D, but their commit here doesn’t seem that big so far, especially after an early release of a device—Google Glass—in 2013 failed. For now, it seems more like another “bet” than a critical initiative).

In fact, whether you are a strong believer in the Metaverse or not, Meta’s excessive Reality Labs spend strongly suggests that they are wasting money in development. John Carmack, ex-CTO of Oculus, said as much in his goodbye Facebook post: “We have a ridiculous amount of people and resources, but we constantly self-sabotage and squander effort. There is no way to sugar coat this, I think our organization is operating at half the effectiveness that would make me happy”. The concern isn’t just that it is a waste of money in and of itself, but that the lack of a financial constraint will rob the organization of the motivation driver of “necessity”. Elon Musk’s SpaceX was able to launch rockets at one tenth the cost of NASA not in spite of their financial woes, but because of them. SpaceX needed to figure out how to reduce launch costs or they would go out of business. When you have unlimited resources, it tends to drain creativity and drive. Hopefully offsetting that is Mark Zuckerberg’s extreme focus and very public push to create the next computing paradigm.

ROIC and Capex.

Meta’s core Family of Apps (Facebook and Instagram in particular) are among the highest returning, scaled businesses ever created. While their profitability can be masked at times because of business transitions (desktop to mobile, Stories, Reels) and their investments unrelated to the core business, we can still see that Meta is a 40-50% ROIC business (excluding non-core investments and business transitions, but you could argue the latter are within the normal course of business).

The 2014-2015 ROIC dop off was due to the WhatsApp acquisition, which increased their invested capital base ~6x in a single year. If you were to adjust out that ~$17bn of goodwill from the acquisition, their ROIC would be >50% for both 2014 and 2015. Total average ROIC since 2012 is 34%, which includes Reality Labs losses which have been weighing down ROIC by about ~900bps in the last few years. If you were to exclude the WhatsApp goodwill from their invested capital base, then ROIC would be another ~25 points higher, with the impact largely backweight. LTM the WhatsApp goodwill has a ~5 point impact to ROIC. (A separate ROIC analysis we ran where we capitalize R&D and then amortize it over a 3-year period would show a ~8 point higher ROIC than the Company ROIC above. The purpose of this analysis is to better match the R&D spend over its revenue generation period, but a combined Reality Labs R&D item gives it more limited utility).

While Meta’s core Facebook and Instagram properties enjoy a strong ROIC, perhaps over 60% if parsed out alone, it has been diluted overtime with the WhatsApp acquisition and Reality Labs investments that both have yet to contribute to earnings. (The Instagram acquisition in contrast was not only their best acquisition, but is often considered one of the greatest acquisitions of all time). While their Metaverse ambitions are understandably years (or a decade) away from meaningfully contributing to earnings, WhatsApp has been lagging in monetization ability compared not just to FB/IG, but even Messenger.

Facebook click to message ad that integrates with Messenger

As of 3Q22, Meta has spun up a “click to message” ads business that is run-rating $9bn already. Click to message ads allow a business to offer a frictionless way for customers to interact with a business. The ease of being able to directly contact a business the moment an item catches your attention has helped increase conversions (and thus ROAS) for many businesses. However, Messenger is responsible for the majority of this revenue today, with only ~$1.5bn of that attributable to WhatsApp.

The trouble with monetizing WhatsApp stems from the privacy centric ethos from which it was built. Since 2016, the app has been entirely end-to-end encrypted, which has made it hard to know what happens after a user clicks on an ad. While they are slowly solving the issue, the WhatsApp team has always had its own culture and has been reticent to embrace any sort of encroachment on what they see as their privacy-centric mission. When WhatsApp was founded, they charged users just $1 a year (with the first year free) to use the app. This revenue model was removed as soon as Meta acquired them in order to increase growth and position them for a more lucrative business model down the line. However, despite buying WhatsApp almost a decade ago they have still yet to meaningful monetize the app.

In the analysis below, we roughly approximate the revenue growth that would be needed for Meta to achieve a 10-15% IRR on their acquisition of WhatsApp. WhatsApp would need to generate at least $9bn in revenues (with a 50% EBIT margin), just to not be value-destroying. If we want just a 10% IRR, WhatsApp would need to generate about $50bn (a figure that is roughly the same in all three scenarios). To achieve a 15% IRR, WhatsApp would need to generate $140bn by 2035 or $190bn by 2040, outcomes we would put at very low probabilities. (We also did not attribute any on-going costs to support WhatsApp since any assumption would be too speculative, but that would have the impact of requiring a slightly higher revenue figure to achieve each IRR). It is also worth keeping in mind that the WhatsApp acquisition could have in part been a means of keeping it out of the hands of Google, Tencent, or another tech conglomerate. 

However, WhatsApp does have some unique ways to monetize outside of click-to-message ads. WhatsApp not only has messaging ads, but also Business APIs that support messaging that is not started through ads. This allows customers to contact businesses through a channel that feels (and usually is) more responsive than alternatives. WhatsApp charges a fraction of a penny for each message depending on the business’ total volume. They also are rolling out ecommerce stores (a particularly big focus in India), where a user can find a business, search through shoppable inventory, and complete the purchase in-app with WhatsApp Pay. All of these efforts are still fairly new, with no meaningful financial information disclosed to judge their success here so far.

In theory though, given that WhatsApp has a very sticky product with over 2bn users, most of whom use the app dozens of times a day, they have multiple opportunities to make something work. The best-case scenario for Meta would be if they can turn WhatsApp into a sort of superapp with a plethora of functionality that monetizes in multiple ways including ads, commerce, business services, and payments. However, a more developed app ecosystem (compared to when Tencent’s Weixin built their superapp), makes it unclear if users even want that. They could find a niche with smaller businesses internationally, many of whom have a limited or no digital presence. We will not explicitly include the WhatsApp opportunity into the valuation, but keep in mind there is meaningful potential here.

Rendering of one of their data centers in Tennessee.

Aside from Reality Labs and acquisitions, their biggest allocation of capex has gone into servers and data centers. In fact, they are now spending almost 2.5x times on servers and data centers (capital expenditure) as they are losing with Reality Labs (operating loss), making their capex a much larger use of cash. Despite their capex having grown ~50% y/y to almost $30bn in 2022, their commentary has been sparse compared to the presentations and demos for Reality Labs.

As shown above, capex has continually outstripped D&A, most recently by a ratio over 3 to 1. It wouldn’t be possible to parse out maintenance versus growth capex, but this is suggestive that GAAP figures overstate true earnings as the capex needed to support their usage and revenue model is consistently above what they are counting in D&A. They amortize servers over a 4-year useful life, which may be how long the server are utilized, but it doesn’t account for the fact that the compute required to service a single user has increased every year as they consume more photos and high-quality video. Since maintaining the usage of a single user grows every year, you can argue that the increase in capex spend is largely maintenance and not growth capex. Either way, this has weighed on their cash conversion (free cash flow excluding SBC as a % of net income).

Whereas prior to 2016 every dollar of net income was converted into over a dollar of cash, now they barely turn it into 50 cents. How much of this capex is going to be the new “table stakes” just to service their users versus having an incremental ROIC is going to be critical to any Meta investment thesis, but it is not an easy question to answer.

David Wehner, current CSO and past CFO, expands on the nature of the capex in their 3Q22 follow up earnings call:

Recommendations are arguably “growth” capex as it increases usage beyond what it was prior. Reels is both growth and maintenance capex, as it cannibalizes time spent in the feed, but then is also net time additive to a user’s session. The last item David Wehner explicitly mentioned was “Advantage”. Advantage refers to a suite of automated tools they rolled out towards the end of 1Q22 that help address some of the performance issues that ATT created for advertisers. This suite of tools includes several product updates that improve targeting, ad placement, attribution, and campaign optimizations. The push to take as much human decision making out of the advertising process is because they are increasingly leaning on machine learning and AI to help make up ad performance gains that are lost from privacy changes. This spend is technically growth capex as it enables more sophisticated capabilities then prior, but if we think of it as a response to Apple’s privacy changes then it is a capital outlay that is necessary just to move back to where they were in a pre-ATT world.

Newly appointed CFO Susan Li had some commentary on their 3Q22 call as well:

Her commentary is framed as though this capex is ROI dependent and they will pull back if it doesn’t generate acceptable returns. However, to the extent that the spend relates to handling a competitive threat (Reels/ recommendations) and restoring lost ad performance (ATT), not spending the capex would be an admittance that they cannot do anything more to contain the threats to their business.  

The 15% time spent increase for Reels (and to a lesser extent the Advantage performance improvements) is the most concrete evidence we have that this spend will lead to revenue growth overtime. In 1Q22 Zuckerberg noted that Reels was 20% of Instagram time spent and that it grew 30% in 2Q22. If we assume ~25% of Instagram time spent is on Reels and Instagram daily time spent per DAU is 39 minutes (per earlier exhibit), that implies ~10 minutes of Reels time per Instagram DAU. Increasing that 15% or 1.5 minutes, when applied to the estimated ~1bn DAUs and taking the $0.14 Meta monetization rate (shown in a prior exhibit), that is ~$1.3bn in potential revenue when Reels is fully monetized. This is significant enough to build some confidence that these investments may indeed be ROI accretive.

Meta posted in their blog the schematics of their data center fabric network topology.

Below you can see how capex was in the  mid-teens as a % of revenue (and under 2x capex to D&A ratio) before stepping up in 2018. In 2018 they attributed the increase to network infrastructure and office facilities in addition to the typical data centers and servers. From 2018 to 2021 though they returned to their prior mid-teens capex as a % of revenue level. Despite 2020 capex being pushed to 2021, they were at capex as 16% of revenues (and a capex to D&A ratio of 2.4x). The following year, we see that capex has increased to $28bn LTM. Their outlook has now called for $32 to $34bn for the full year 2022, breaking the pattern of coming in below the low range of their capex guidance. For 2023, they expect capex to grow again, landing in between $34-39bn. 

As David Wehner commented on the 3Q22 call (below). He attributes the increase in capex to AI investments and suggests that this level of AI investment could be a short cycle if it doesn’t show measurable returns by the end of 2023. The second category of investment he mentions, the data centers, are typically 30-year investments. Their increasing usage has driven on-going data center investments, but in theory at some point data center capex should fall, but the average compute each user consumes has only gone up overtime.

As mentioned, it is very hard to know how much of this capex is just to keep their business operating as is versus providing incremental revenue. While we noted how AI can improve time spent, the more powerful driver is improving ad targeting.

Below we show an illustrative advertiser (Ouai in this example), selling a Scalp Serum. This shows two things: 1) how powerful just minor improvements an AI can be, and 2) how susceptible Meta’s revenues are to just a small deterioration in their ability to target.

Above we see the product as it shows up on Instagram after you click the ad. We are estimating the gross margins and are assuming the CPMs are $10 (which is on the low side for beauty CPMs).

Below, we have two tables. The first table is to show how much the advertiser has to spend to sell a single unit. This table is sensitized across different click-through rates (people who click the ad) and conversions (people who buy the item after clicking the ad). This top table shows that if you have a 2% click-through rate and a 2.75% conversion rate, Ouai will spend $18 per unitin advertising (1/click-through rate x conversion rate). The table below shows that at that click through rate, conversion and CPM would yield Ouai a 2.9x ROAS.   

Now, if Meta can use their AI to improve who they are showing the ad to, then perhaps they can increase the click-through rate 0.25%. This would reduce Ouai’s advertising cost per unit to $16 and increase their ROAS to 3.2x. However, what typically happens instead is that the improvements get competed away by other advertisers bidding. If Ouai was happy with their prior 2.9x ROAS, then they can now bid up to a $11.25 CPM and receive the same ROAS because of the improvement in click-through rates. Not only does Meta benefit because ad prices increase, but the improved targeting means that they will be able to show ~200 fewer ad impressions before they find someone who wants to purchase the item. This freed-up inventory can then be sold to another advertiser. If you assume this incremental inventory is sold at the same CPM, improving click-through rates by just 25bps would translate to 24% revenue growth for Meta.

In reality though, as targeting improves, inventory opens up, which first depresses prices, as auctions are less competitive. This lowers prices, which makes it even more attractive to advertisers, and more sellers like Ouai allocate more ad budget to Facebook. Then the advertisers bid up the cost of impressions again and (hopefully) Meta releases new ad targeting improvements which start the cycle again, growing revenues with each iteration while maintaining happy advertisers. When we said earlier that Meta was more optimized in terms of AdTech than other players, we meant that they went through this cycle more times than any other ad platform.

This is also a risk though, because as they improve ROAS for a business, the business grows and advertises more. However, if the ROAS is permanently impaired (and irreparable) from ATT or other future privacy actions from Apple, Google, or regulators, then this is economic activity that is destroyed. It doesn’t matter if Meta is the best relative place for advertisers to advertise because businesses are ROAS sensitive and will adjust their budgets up or down accordingly. In the most pessimistic scenario, ROAS degrades from potential future privacy actions or regulations to the point that D2C businesses go bankrupt as they can no longer effectively advertise, and thus their traffic dries up. The same way digital advertising enabled a whole new generation of internet born D2C businesses to thrive, extremely draconian restrictions on data usage could reverse their success entirely and drive them into bankruptcy.

Usually when companies have very strong ROICs, investors will typically urge them to retain capital and invest it on their behalf (this was very explicit for Constellation Software for example, where investors demanded they lower their required rate of return instead of distributing it). However, despite Meta’s strong financial track record, investors are still largely skeptical that their incremental capex will have an adequate ROIC. We can understand this hesitation, as their objectives for this capex, and its magnitude, are different than prior investments. It’s not like investing with a quick-service restaurant, where they copy and paste the same restaurant format hundreds of times with high financial predictability.

The nature of Meta’s capex is oriented towards doing things they have never done before, and while the rationale is strong and we are seeing early signs of it bearing fruit, the ultimate outcome cannot be assured. However, at the same time, it seems implausible that they yield nothing from it. And as we will show in the valuation section momentarily, only a relatively small amount of improvement is needed to support revenue growth.

We will tie in different capex scenarios and how it impacts free cash flow in our valuation section.

Valuation.

Our approach to valuation is to invert the question. Instead of asking what a business is worth today, we estimate what is the implied business return at today’s market price. We do this through a reverse DCF, where we make various assumptions on a company’s future cash flows and then figure out what discount rate would make the sum of future discounted cash flows equal to the current market price. The output of this analysis is a range of “discount rates” which can be thought of as your return on the purchase of business at today’s market price if you were to directly receive the excess cash flows (and the assumptions held). When we think of a company as under or overvalued, it is mathematically the same as saying the implied discount rate for the risk inherent in the business is too high (business is cheap) or too low (business is overpriced). In our opinion, framing the opportunity in terms of discount rates makes it easier to conceptualize the investment opportunity and answer the question of if the return adequately compensates for the risk. (It can also keep an investor from making mistakes with multiples. In theory, a multiple is shorthand for a DCF, but it is commonly used out of convenience, leading to investors being blithely unaware of the assumptions that their multiple implies. Most commonly, the multiple makes implicit assumptions investors would be uncomfortable with were they to be explicit. The Reverse DCF avoids this by making all assumptions explicit).

For Meta, there were many different scenarios that we wanted to sensitize around. However, the two critical variables are revenue growth and capex. Our cash flow projections start with separate revenues for Family of Apps (FOA) and Reality Labs (RL), and we add back all of D&A to the FOA EBIT to get EBITDA. While we typically eschew EBITDA metrics, we are doing this in order to fully burden earnings with current capex, which as we noted earlier is 2-3x higher than D&A. LTM FOA EBITDA margin is 49%, which compares to our peak estimate of ~61% and 2021 figure of 56%. This is due to expenses growing 16% LTM, but revenues growing just ~1%, leading to negative operating leveraging. We will run different scenarios, most of which assume they move back to a pre-2022 EBITDA margin level of ~56%.

The other big question is what their level of capex will be. As capex has outstripped D&A, their gross margins (where most of D&A is captured) have started to compress, from a peak of 87% (average since 2015 is 83%) to 81% in 2021 and 80% LTM. Our sensitivity analysis only shows EBIT margins, but we are taking various FOA EBITA margins and fully burdening it with the period’s capex (effectively assuming capex levels match D&A eventually). 

This ties into the prior discussion we had on maintenance versus growth capex. If the elevated level of capex is “growth capex”, then an investor would want to assume a scenario where Meta is still growing to be logically consistent. If it is maintenance capex, then you would want to pick one of the no or low growth scenarios. We think that their AI efforts will bear at least some improvements in recommendations, ad targeting, and ad attribution, which should yield some revenue growth. Of course, it is hard to know how much and where the limits of adding more GPU clusters to improve their algorithms lie. Either way, if the incremental investments prove to be a waste, it seems unlikely they will continue to invest at the elevated rate. Our sensitivity analysis assumes either 25%, 21%, or 18% capex as a % of revenue, which is reflected in the EBIT margin (since capex will approach D&A rate overtime). 

We did not sensitize for the Reality Labs (RL) losses, since they do not change the investment thesis today in our estimations. All of our models assume Meta will lose $15bn per year from RL for the next 5 years and then have no impact thereafter. Removing the RL losses is worth about 1-2% of return, but we think it is prudent to assume they continue to lose money for the next ~5 years here. Our reverse DCF analysis assumes zero value creation from RL. Of course, even if they are not very successful with their VR endeavors, this could be too draconian. We separately have a Metaverse build that shows what success could look, but we do not include it in the figures below as we believe it would convolute the analysis by putting assumptions you can have some level of confidence in with a scenario you cannot.

The table above shows our revenue growth assumptions for each period. We label each growth scenario according to the starting growth rate. So the “10%” growth rate scenario (which is how it shows up in the Reverse DCF analysis) is really 10% growth for years 1 to 5 out, then 7% for years 6 to 10, then 5% for years 11 to 20 and then 3% thereafter.

Below is the valuation at the time we ran the Reverse DCF.

Our EBIT margin range below was informed from the following assumptions shown below. For instance, the 24% EBIT margin was informed from the 49% EBITDA margin less a 25% capex as a % of revenue assumption (where capex conforms to D&A).

With 3% revenue growth, the range of free cash flow (or owner earnings) 3 years out for the 2^nd worst and 2^nd best capex scenarios are ~$19bn–26bn. With 5% revenue growth, that increases to $21bn- $28bn. With 8% revenue growth, that jumps to $24-32bn. We see that even though we cut the range of variables we are sensitizing around in this exhibit for the sake of simplicity, there still is a wide range of $19-32bn in FCF in 3 years out. Remember though, that all of these scenarios include $75bn of RL losses (and thus no value to their Metaverse), and we are not explicitly modeling any revenues from WhatsApp.

We ran our reverse DCF on the full range of variables though. The reverse DCF outputs below are for various revenue growth and EBIT margin scenarios. An investor should judge whether they deem the returns offered are adequate for the inherent risks in a Meta investment and whether they have a sufficient level of confidence in their assumptions to invest at all. Also, keep in mind the table is still only a narrow band of potential outcomes and things could be much worse (risk section to follow). Of course, Members Plus subscribers can download our excel worksheets and adjust any inputs accordingly.

Above, it shows that if an investor is comfortable assuming just GDP-like growth of 3%, that this level of capex isn’t indefinite, and that they get back some operating leverage at some point, the return is >10% (despite not considering any earnings potential of the Metaverse but including ~$75bn of losses over the next 5 years). If an investor is comfortable assuming a 5-8% growth rate, then the implied return is in the mid-teens at today’s price (given other assumptions hold).

Interestingly, assuming 0% growth and a moderate margin scenario yields a ~9% return, which is about what equity markets have returned historically. While you would want an equity risk premium for an individual stock, we can see that the market is currently roughly implying that Meta is done growing forever. For context, this is a stock that has never grown less than 20% annually for its entire decade-long public history, prior to last year (more on this shortly).

In the -5% growth scenario, an investor would consider Facebook/Instagram to be a melting ice cube and need the Metaverse to be very successful in order to rationalize an investment.

We wanted to further examine the revenue growth assumptions to see how they could achieve different levels of revenue under different scenarios. If you recall the illustration below, the main drivers of revenue can be broken down into # of users, time spent per user, ad load, and ad price.

The first three are easiest to explicitly model and have an opinion on, and the exhibit below does just that. We sensitize for three different scenarios for # of users, time spent per user, and ad load. These assumptions then yield a change in ad impressions for 5 years out from now. We assume that the # of ad impressions linearly impacts revenue (an assumption we will comment on momentarily). Then we show what the implied ad pricing improvements (or contractions) would be for Meta to hit various revenue goals. In the “adverse” scenario below, we show that with time spent headwinds of -20% cumulatively and losing -2% users annually, Meta can still grow revenues 5% by increasing ad prices 12% annually (or 76% in total over the time period). In the “favorable” scenario it shows ad prices dropping 3% annually for Meta to grow revenues only 5%. Ad prices dropping would likely spawn future revenue growth (as advertisers bid up the cheaper impressions to their prior ROAS).

This table helps break up the assumptions that different revenue growth scenarios imply. If Meta isn’t growing from ad impressions, then it must be from pricing. And if they are to maintain the same ROAS, then their ad targeting improvements need to match price increases. Said differently, if you believe ROAS does not have much room to go down, you may think of the implied ad pricing as really being the amount of ad targeting improvement that needs to occur. From this perspective, in the base case, Meta’s ads need to get just 18% more effective, over 5 years, in order for them to achieve 5% revenue growth.

Importantly, this analysis assumes a linear translation from ad impressions to revenues. However, not only is it likely that the earlier impressions in a user’s Facebook or Instagram session are worth more, but also the value of an impression varies dramatically by geography. This is both a source of opportunity but also a risk, and why the 1Q21 disclosure of North American ad impressions dropping -6% in 1Q21 was so concerning. While net impressions still increased 13%, that was mostly from lower value regions. Now, part of this ad impression contraction is a self-inflicted result of the shift to Reels, with fewer ads for now, but the leaked Teen stats show there are usage concerns for Facebook in particular, but also Instagram.

Now, the opportunity of course is that other countries start to close the gap versus the US. However, fewer new business formations, less D2C companies, and overall smaller consumer spending with the average person globally making far less than an American, could mean that closing the ARPU gap is a multi-decade process at best. Looking at Europe, which is at $65 versus >$200 for the US is also a bit confounding. The US average consumer spending per capita is about twice that of the EU’s, which can explain a good portion of the difference, but not all of it. We think other factors include fewer D2C companies, cultural differences that lead to more of a preference to buy offline and spend more on experiences, and less impulsive purchases—the US is where the idea of “consumerist culture” was born, after all.

There are multiple ways that Meta can achieve revenue growth, with the Metaverse being by far the most uncertain. As mentioned in the Metaverse section, they plan to monetize it through advertising and taking a fee on on-platform transactions. We do not explicitly model out hardware revenue and just assume it will not be a meaningful portion of profits (even if successful), but that could be wrong. Either way though, the big and more valuable sources of revenues are likely to come from advertising and transaction revenues. Below we show a range of valuations assuming 500mn DAUs, 40% EBIT margins, and capitalized at 20x.

Clearly, there is a lot of upside here if they are able to materialize their Metaverse ambitions, but it is hard to put a probability on success and even harder to figure out what to pay for that possibility today.

We will conclude this section with Mark Zuckerberg on the 3Q21 call.

Investment Narrative.

Meta has grown revenues at a 31% clip annually since 2017, but for the first time ever, revenue growth not only evaporated, but actually contracted in 2Q22 and 3Q22. Not only were they losing the next generation of users with Facebook, but for the first time ever, there were cracks showing that Instagram user health was potentially about to hit a wall and march towards a slow demise.

TikTok, the first international app to ever gain global domination with over 1bn users and most of Gen Z spending the better part of 90 minutes a day on it, was the culprit. It was so wildly engaging that some teens developed ticks from it, and it spawned copy-cat products from not just Meta, but Snapchat and YouTube too. Nothing could give more credence to the concern that the “next new thing” would be too unpredictable for the established media companies to respond to than having a previously unknown company to the western world come to dominate it within just four years.

While Meta was busy with its most material competitive threat ever faced, it was in the midst of being sued by the FTC for anti-competitive practices. They alleged Meta would buy or bury potential competitors’ products and sought to potentially unwind the Instagram and WhatsApp acquisitions. While the FTC’s case was definitely an uphill battle, as shown by the original case’s dismissal for lack of evidence, investors worried that even Meta’s legal success could be pyrrhic.

Critics drew comparisons to when the Department of Justice filed antitrust charges against Microsoft, noting that even if the court outcome was favorable, it still distracted them and built in a paranoid culture of checking in with lawyers before doing anything. How would Meta’s “Move Fast and Break Things” culture hold up?

While investors never thought of the acquisitions as being pernicious to the point of illegal, they certainly were counting on them to help Meta grow and stay relevant over time. Without the ability to acquire, they would be stuck with their own home-grown creations, which was a problem.

Other than their original Facebook concept and the spin-off Messenger app it spawned, their other successful apps were acquired. And their past efforts to create stand-alone apps, including Camera, Poke, Slingshot, Paper, LifeStage, Riff, Threads, Hobbi, Notify, Tuned, SoundBits, Audiohub, Venue, Collab, CatchUp, Moments, Mentions, were… lackluster. Meta’s original response to TikTok was to roll out another stand-alone app, Lasso. This is especially ironic given TikTok themselves didn’t reach any level of success globally until after they acquired Music.ly and their ~200mn users. If only Meta had a big audience to seed growth on Lasso…

This competitive response was eerily similar to when Instagram and Snapchat originally came out. They would try to acquire them while launching copy-cat apps (Camera and Poke, among others) that never gained prominence. While they were able to eventually buy Instagram, if Meta would have to rely on their homegrown efforts to fend off competition with acquisitions off the table, history wasn’t on their side that they would bear much success. 

While Meta appeared to be fumbling the competitive response, Zuckerberg was giving keynotes on “The Metaverse” and making comments about how they need to own the next computing platform so they can have a say in its development. This was a thinly veiled comment aimed at how Apple was swiftly and unilaterally wrecking Meta’s ability to serve targeted ads.

ATT, or App Tracking Transparency, was Apple’s latest attempt at premiumizing “privacy”. Apple would roll out commercials showing a user’s phone, constantly spewing out private information to anyone and everyone, as if the current mobile advertising environment wasn’t their creation. It didn’t matter that Apple originally created the IDFA years earlier as a more privacy-friendly way to advertise (and Google copied them); Apple would have no problems winning the PR war and convincing the vast majority of users to click “do not track” when that app prompt appears.

For those that “opt-out” of being tracked, Apple has a solution though: you may use Apple’s SKAd Network. For those that do want to advertise on Apple’s own properties (the App Store, Apple News, and Stocks, for now), you would be happy to know that Apple will not show the prompt to their users since all the data on their platform belongs to them, and people like “personalized” ads anyway.

All is fair in war though, and while it might be a bit cynical, it is also clever. Despite Zuckerberg having a legitimate gripe that Apple’s ATT would hurt small businesses’ ability to advertise since they have far less data to use aggregated measurements successfully and less sophistication to install things like CAPI, which help restore lost signal, the press was nearly universally negative for Meta and uncritical of Apple.

The press would have plenty else to criticize Meta for. The legacy media would place Facebook at the helm of the country’s political polarization and blame them for helping get Trump elected while regularly hosting hate speech and feeding teens toxic content. A whistleblower would supply the press with thousands of documents that show Facebook’s and Instagram’s negative impact on teens’ mental health, which they would happily publish on day after day.

Meta’s unprecedented contracting revenue was met with the toughest competitor they have ever faced, one that is loosening their grip on the next generation of users. This is while Apple wrings their ability to deliver accurate ads, requiring their largest capex spend ever of over $30bn to restore some signal through experimental AI technology. In the meantime, the FTC is moving in with an antitrust case while a slew of other regulatory gripes from the EU to Australia start to foment. This is all being met by a CEO who has championed losing tens of billions for an undefined amount of time on an unproven sci-fi dream of a virtual world as the economy potentially marches into a recession. This narrative was more than enough to send the stock down ~70% from its peak, pricing the company as if it was done growing forever and oblivion was only a matter of time.

But it’s a pretty bad narrative. It’s also not very accurate.

Usage has actually never been higher. Their apps have over 2.9bn people who use them every day and 3.7bn every month. The Facebook app alone still has almost 2bn daily users, and despite all of the talk of Facebook Blue sputtering into irrelevance, their daily and monthly active users in North America are at an all-time high. It’s true that there are concerning trends that the next generation of users are joining Facebook at older ages than previously, but at least for now, they are still eventually joining.

In the meantime, Meta also has Instagram, whose short-form video product Reels may have come late, but its working: Reels is net additive to a user’s time spent on Instagram. The shrinking revenue can in part be attributed to more time spent on Reels versus other Instagram products which monetize at higher rates. Just like their initial introduction of Stories, there are fewer advertisers on newer formats, so they face a headwind while they build up the new product.

People may still love TikTok, but TikTok and Instagram are not direct competitors. For Instagram’s core social and sharing functionalities, with a user staying connected to all of their friends and favorite influencers, TikTok has no answer. The only other platform that enables broad, personal sharing with your social network is Facebook. 

There was a fear that a messaging app, with its full list of your contacts, could turn into a social network over time, like Tencent’s Weixin or Snapchat, but Meta’s purchase of WhatsApp has mitigated that risk. In the US we may forget, but WhatsApp is the dominant messaging app for most of the world with well over 2bn active users.

Apple’s ATT did create material impairment for Meta’s advertising business, and it is likely the largest factor of their revenue contraction. Meta claimed it was a $10bn headwind, but it could have been more. Their immediate solution is CAPI (Conversion API), which allows Meta to receive data on the back end, server-to-server, directly from the publisher. Their new Aggregated Event Measurement tool will allow them to comply with ATT, but it won’t work well for smaller advertisers whose sales are insufficient to supply enough data. Their latest effort, “Advantage”, is a suite of AdTech products that rely on their AI abilities to help restore lost signal for their ad matching engine and repair attribution.

The problem is that most businesses are return on ad spend sensitive. They do not care if Meta is the best place to advertise relative to anywhere else; they only care that every dollar they spend on Meta is well spent. For the better part of a decade, no one questioned Meta’s abilities here, but ATT changed that. No other competitor may be better than Meta at showing an ad to someone who might want to buy something, but that isn’t good enough for them to keep growing.

If a business cannot get an adequate return on ad spend, they usually can’t go elsewhere: it is just potential economic activity that is destroyed. At its best, Meta’s ad engine would place ads to people who want to buy stuff, and in doing so, they helped grow businesses, which in turn had more money hire and contribute to the economy, which gave consumers more money to spend in the first place, after which Meta would show them an ad on what to spend that money on. It was a virtuous cycle.

While this may be a bit idealistic on the virtues of consumerism and Meta’s role, this is definitely true to some degree. 10mn businesses advertise on Facebook and Instagram and they and the economy are better off for it. You don’t have to talk to many small businesses who advertise before you hear how vital Facebook and Instagram are to their growth.

The question isn’t just whether Meta can restore return on ad spend to prior levels for all of these businesses, but whether they can keep improving it.

To address this longer-term, Meta is investing ~$30bn this year and another >$34bn next year in servers, networking equipment, and data centers to build the infrastructure that supports next generation AI capabilities with the aim of not just restoring lost targeting and attribution abilities, but improving it beyond what it was previously. Despite Apple stymying their data collection abilities, Meta still has the most data on the most people and is in the best position to improve their ad targeting abilities.

Meta is starting to see some success with their Advantage+ tools: Detail Targeting improved cost per incremental conversion by 37%, Creative lowered cost per action by 3%, Lookalike decreased median cost per action by 17%, and Shopping lowered cost to purchase a conversion by 12%. None of these will be applicable to all advertisers and nothing is a panacea, but if they can continue to make incremental gains, that is enough to return them to revenue growth.

While it’s not as bad as the market’s perception, there are still negative potential adverse risks. Further privacy restrictions, whether it be Apple’s Private Relay, Google’s rollout of their version of ATT, or something else could be another blow to their ad revenue, perhaps worse than the original impact if Meta was using the Android data sets to inform the models applied to iOS users.

There are also some concerning legislative pushes in the EU to make all personal advertising opt-in, or more bluntly, make standard digital advertising as ineffective as linear TV or radio ads. What they are solving for here is far from clear, but it could be another material headwind for Meta—especially if other markets adopt similar legislation. Publishers are also using regulators and legislation to get paid for their content that they themselves willingly publish to Facebook. While totally backwards, such legislation had some success in Australia, and similar bills are being proposed elsewhere. Even when Meta is in the right, they can’t win because of the years of scandals that robbed them of all their goodwill.

The Meta name itself was adopted in part to help repair their image. Despite all the criticism Zuckerberg takes publicly, he is extremely introspective and far thinking. Zuckerberg was planning for a lot of these issues a decade ago.

In a leaked 2015 email, he laid out his vision for the future of Facebook with VR/AR at the center of it. He noted that their aims were to restore their strategic position that they lost when the mobile transition originally happened. Prior to mobile, Facebook had a budding 3^rd party developer business and freedom to police their platform as they wished. Now though, they have been relegated to being just another app in Apple’s and Google’s app stores with all of the restrictions that that entails. Becoming a leader in the next computing platform would help them restore their strategic position.

While they were building the next-gen computing platform, Zuckerberg reasoned their tarnished brand would be helped by building tangible products that people love. Innovative companies with hit products can weather criticism and scandals much better than those who seem outdated and unimaginative.

Zuckerberg’s VR and AR efforts would enable Meta to not only ensure their social apps survived the transition from mobile to VR/AR, but would allow them to regain their strategic position. To do this, they would buy Oculus in 2014 and spend tens of billions of dollars, perhaps the most of any company ever spent on a single technology. Of course, there is plenty of risk in such a bet, but it’s not at all ill-thought out.

There are two general investing philosophies. The first is to look to downside risk first, and if you see a fair chance of impairment of capital, then you move on. Under this philosophy, if you are not sure how the future can unfold, you simply put it in the too complicated bucket and move on. Preservation of capital is primary.

A second philosophy is to focus on the risk-adjusted return of each investment. If a particular $10 stock has a 20% chance of going to $0 and 80% chance of becoming worth $15, then you would make that investment because the expected value is $12. If you have a large enough portfolio of such investments, it is statistically impossible to not make money.

Meta is interesting because it seems like there is a fair chance of impairment of capital from users flocking to another platform, regulators disallowing targeted advertising, further privacy actions by the mobile platforms, or the “next new thing” totally subsuming them. No doubt these are all real risks, but when you look into the plausibility of each of those arguments, it becomes somewhat facile.

Usage has already survived an uncountable number of scandals and regulatory action. As much as regulators are hurting Meta, they may simultaneously be increasing their competitive moats as few will be able to match their content moderation advances with AI or army of 40,000 human moderators, not to mention their legal staff to comply with a growing myriad of legislation. And while Apple caused Meta material harm, their groundbreaking privacy actions still left Meta with positive growth after adjusting for FX. And while the “next new thing” risk remains, no competitor has ever had success in building a social graph so far.

It’s challenging because the market perception of Meta is far worse than the reality, but that doesn’t mean there aren’t still meaningful risks for them. The most material of which are: 1) the implementation of Google’s version of ATT or further Apple data restrictions, 2) regulators making personalized ads opt-in or introducing data restrictions on first party data. There are other risks, but these two seem to be the most visible today. 

Having said that, it also seems implausible that their massive capex spend yields no AdTech improvements, and as we have shown, there only needs to be a minor improvement in targeting to lead to material revenue growth. For a company that was growing double digits without a hiccup for a decade, an investor only needs a low to mid-single digit growth rate to get a solid return. And that is before factoring in anything more material happening with WhatsApp, or “The Metaverse” play actually working.

The story is far from over, but the ending is also far from clear. That is why we go to lengths to show what assumptions an investor needs to assume for different returns. It is ultimately an investor’s judgment whether or not certain assumptions adequately compensates them for the risk. (This is why we called our firm Speedwell: it is named for the boat that helped deliver passengers to the Mayflower, because ultimately it is the investor, and not us, that must make the journey).  

Risks.

1) App usage falls. This can happen either because other services are more addictive, fun, or useful, which results in time spent dropping. Similarly, if a new product replaces their core socializing and messaging functionality, time spent will drop. If Meta fails to get the next generation of users on their apps, eventually the value prop for all of them would fall, and usage could spiral out of control. There is also always risk of the “next new thing” replacing some or all of the functionality of their apps.

2) Further ad signal loss. Apple’s Private Relay (which scrambles IP addresses) or their other privacy initiatives can further decrease Meta’s ability to collect data on iOS and target advertisements to users as well as know if an ad was effective. Google will roll out similar changes eventually, as they have been a slow follower to whatever privacy changes Apple makes historically. There is a concern that Meta is using the data they collect from Android users to help inform their models for iOS users, which would be made impossible in short while, so targeting and attribution would be adversely impacted again.

3) Regulation against Personalized Ads. This is a newer risk, but essentially all data privacy measures and regulations so far have been aimed at managing 3^rd party data. The EU recently fined Meta ~$400mn for breaching GDPR, under a new interpretation that claims Meta must receive explicit permission to show