TikTok is, demonstrably, Meta’s most immediate competitive threat. In the company’s Q4 2021 earnings call, Meta CEO Mark Zuckerberg mentions TikTok by name six times, including in this commentary about the company’s Reels product:
The dynamic that I think is actually a little bit different with Reels than what we’ve seen with Stories and mobile feed in the past is with Reels, I would say that the teams are executing quite well and the product is growing very, very quickly. The thing that is somewhat unique here is that TikTok is so big as a competitor already and also continues to grow at quite a faster rate off of a very large base.
At the time, many analysts interpreted Meta’s emphasis on the competitive threat posed by TikTok as a response to the FTC’s updated lawsuit against the company, which had been approved by a federal judge just three weeks earlier. The re-filed suit employs logical gymnastics to differentiate the Facebook and TikTok products such that Meta’s products can be evaluated on a standalone basis.
Regardless, Meta very obviously channeled TikTok’s product design sensibilities last August when it launched Reels, a short-form video feature that is nearly indistinguishable from TikTok’s core content format. TikTok is mentioned just once in Meta’s Q1 2022 earnings call, and by the company’s CFO, Dave Wehner, while Reels is referenced nine times by Mark Zuckerberg and nine times by Sheryl Sandberg. For Q1 2022, Reels captured 20% of time spent in the Instagram app.
In June, reporting from The Verge‘s Alex Heath revealed that Meta will prioritize Reels and its underlying short-form video format throughout the product catalogue by enacting a fundamental transformation: abandoning the friend graph that dictates the content exposed to users in favor of a “discovery engine.”
Per a leaked internal memo, Meta’s “discovery engine” will allow relevant content from outside of a user’s friend graph to be surfaced to them in various content placements across the product portfolio. From the memo:
Today this is changing. Social media products – including our own – are delivering value by investing more in discovery engines that help people find and enjoy interesting content regardless of whether it was produced by someone you’re connected to or not. We see this both in research as well as in the growth of products like Reels, Watch, and In-Feed Recommendations (IFR).
Currently, 11% of content exposed to users is yielded from “unconnected” sources. Presumably that number will increase dramatically given this new initiative: the memo states that a balance will be struck between connected and unconnected content, and the integration of the Messenger app back into the core Facebook product (from which it was decoupled into an independent app in 2014) will promote communication between connected users.
Abandoning the friend graph in favor of an open graph, or a “discovery engine,” effectively means that the surface area of content sourced for exposure to any given user in their feeds and elsewhere will be enlarged beyond what is shared by connections. Ultimately, with an open graph, any content shared across the entirety of Meta’s portfolio is eligible to be surfaced to any user (with certain restrictions). As I explain in this QuantMar thread, Facebook currently ranks content within a given user’s friends scope for inclusion in their feed. The constraint of the friend connection is removed in the open graph: the entirety of aggregated user content can be ranked for relevancy and expected engagement for a user’s feed.
By adopting the open graph, Meta expresses two important ambitions:
- Encourage creators to produce more content for Facebook and Instagram because that content can be exposed to a far wider audience than simply their connections. Influencers attribute the phenomenon of going viral on TikTok to the “algorithm,” which is partially true, but really, it is the open nature of TikTok that allows a user’s content to be ranked for inclusion in the feeds of total strangers;
- Enlarge the pool of content available to be exposed to any given user during a session, which ideally increases time spent on site.
A precondition hidden in the second outcome is that the “unconnected” content surfaced to users must be relevant and interesting: surfacing random or insipid content to users may actual reduce time spent on site.
And classifying video content for relevancy is substantially more challenging than classifying text or static image content for relevancy. To that end, Facebook has announced that it requires 5x more GPU capacity in its proprietary data centers in order to parse video content into inputs to its recommendation models. Classifying text-based content is trivially straightforward. Parsing signals from video files is computationally much more expensive. Per the memo, Meta sees the creation and sharpening of this classification and recommendation system as one of the three primary pillars of its new operating strategy (“Build world class recommendations technology”).
It is from this pillar that the commercial consequence of this new strategy takes shape. There are only four means by which advertising revenue can be grown for a platform:
- Increase “ad load,” or the ratio of ads shown to each user per session relative to organic content;
- Increase reach, or the number of users that engage with a product and thus are exposed to ads;
- Increase the value generated by ads through higher-quality formats or better targeting, which improves the general price paid for ad inventory through increased bids from advertisers in the ad auction;
- Increase time spent on site, which provides more opportunities for ads to be served.
As I point out in Four risks that could slow Facebook down, published in 2016, Meta (then Facebook) succeeded in growing advertising revenue in an earlier era through a concerted effort to increase ad load, improve the relevancy of ads, and grow the reach of the News Feed. And in this article from 2017, I outline the impact of Meta’s improved ads targeting products on its revenue. The introduction of its App Event Optimization (AEO) and Value Optimization (VO) bid strategies, which I examine on QuantMar, allowed Meta to grow ARPU in the US & Canada region by between 50% and 100% relative to other regions which wouldn’t have benefited from those targeting capabilities.
This leaves tactic #4 from the above list as the last remaining option for increasing advertising revenue. And this is the purpose that the open graph is designed to serve: if user base reach, ad load, and the value of ads served across Meta’s products can’t be improved, then the only pathway to revenue growth is through increased time spent on site. The Reels product could possibly ignite growth in high-ARPU regions by arresting TikTok’s expansion and plundering some of its users, but given the existing scale of Facebook and Instagram, the potential opportunity captured in that seems small.
Meta’s goal with the open graph — and in prioritizing short-form video, which is generally more engaging and entrancing than short-form text-based content — is to expand the pool of content available to be exposed to each user. And with more content available to be shown to a given user, Meta benefits from more opportunities to score relevance and personalize the experience to that user: the user is exposed to the best and most relevant content from a larger pool of possibilities.