Understanding Meta’s Lookalike Audiences: Creation to iOS 14.5 ATT Changes and Beyond
Looking for ways to reach new customers for your business? Walled garden audience extension solutions like like Meta's Lookalike Audiences, Google's Audience Expansion and TikTok's Lookalike Audiences are often the go-to solutions for advertisers to target prospective customers.
A walled garden in the ad tech space is an advertising platform where the publisher manages all aspects of purchasing, serving, tracking, and reporting, essentially a closed system where publishers own their entire ad platform. As a closed system, there are few technical documentations available from these ad platforms on how.
But how do they work, and why have these solutions become less effective since the release of Apple’s iOS 14.5? Let's focus on the largest and broadest platform, Meta, where most advertisers have found success for years.
How does Meta create lookalike audiences?
Meta’s lookalike (LAL) audiences are some of the most powerful tools for growth marketers looking to acquire new customers for their brands.
Meta generates LAL audiences based on the data advertisers share with the platform; it could be a list of their current customers, recent website visitors or the pool of Facebook and Instagram users who interacted with advertisers’ ads. These inputs are called ‘seed audiences.’ Advertisers are telling Meta to find users who are more like these seed users.
Another level of audience compatibility is called ‘value-based lookalike’ audiences. With this method, you assign a ‘value’ to each member of the seed audience. For example, you could assign the lifetime value of your customer as the value. By doing so, you are telling Meta that some customers in the seed list are more valuable than others. Meta then tries to find prospects who are more similar to your high-value seed audience.
Behind the scenes, Meta uses powerful machine learning (ML) models to create these LAL audiences. These models take advantage of the platform's vast user activity, both within and outside of it: Advertisers send data about their users to Facebook through pixels and server-side events, allowing Facebook to gain visibility into its users' browsing and purchasing activity outside of the platform.
Let’s walk through a hypothetical example of this interaction: An advertiser is selling healthy dietary supplements for young kids. The brand’s seed list has their current customers, many of which are parents with young children. Meta can match these users’ browsing behaviors (such as clicks, likes, comments, shares and more) within the platform.
Some of these parents may be searching for local daycare in Facebook groups, liking local school’s Instagram posts, commenting on their friends posts about a nearby children’s museum, etc.
Meta has knowledge of these users' activities outside of the Meta platform. These same parents may be purchasing children's clothing and strollers from other merchandise stores that share data about their customers with Meta. This provides Meta with a near-perfect view of the consumer journey, often called the ‘click path,’ both within and outside of its platform.
Meta's ML algorithms sort through this granular data to identify patterns. In this example, the algorithms may find that customers of the kid’s dietary supplement brand also shop at certain other brands and engage with certain types of posts within Meta’s platforms.
With this understanding, the algorithm then finds the ‘nearest neighbors’ of those users, people who also may be purchasing children’s clothing online or researching about local daycare, but they aren’t already current customers of the advertiser, and thus also not on the seed list. The algorithm identifies those people as good prospects and includes them in the LAL audience.
Meta's ML finds these micro-patterns amongst thousands and millions of behaviors and purchase traits and identifies top prospects. If our advertiser is building a 5% LAL audience then Facebook identifies the top 5% of its users on the platform who share most similarities with the seed audience.
How is interaction different now after Apple’s iOS 14.5 ATT?
In April 2021, Apple released iOS 14.5, which introduced the App Tracking Transparency (ATT) prompt. This prompt requires users to provide explicit permission for apps to collect and share data that was generated outside of the app. With over 80% of US iPhone users opting out of this data sharing agreement, it has created a serious challenge for platforms like Metas in terms of ad delivery, optimization, and reporting.
Although Meta can still track the breadcrumbs of its users' activities within its own platform, it can no longer use the click path of users' activities outside of Meta to match them with their activities within the platform. This is despite advertisers and brands continuing to share their granular purchase data with the platform. Meta cannot link the activity of these users outside the platform to their activity on the platform as effectively as it was able to do before. Consequently, Meta's lookalike algorithms are making inferences on less granular data. Often this is termed as ‘signal loss’ in ad tech communities.
The Path Forward: How to Improve Performance of Your Lookalike Audiences
There are however options for advertisers to navigate this signal loss, some of which are more costly than others.
The brand can create their own LAL models outside of Meta by utilizing their valuable first-party data (customer data) and the power of machine learning to build predictive audiences off a licensable US consumer graph. This graph includes demographics, interests, psychographics, and consumer purchase insights (both in-store and online) from over 1,100 brands. Since the advertiser licenses this data and these audiences, they are not subject to Facebook's App Tracking Transparency (ATT) opt-out. In fact, these audiences don't need to use any digital identifiers such as cookies or mobile advertising IDs. This makes them future-proof for a cookieless world, and provides heightened privacy and security. These advertisers then can send their custom predictive audiences to Meta for matching and activation.
Another option for advertisers to navigate signal loss is to improve the quality of Meta’s Lookalike (LAL) audiences by providing better seed data to the platform. Instead of sending the entire customer base as seed, one could send a subset of the entire customer base filtered based on predictive power of the customer base, a ‘predictive seed.’ The filter could be high predictive lifetime value customers, or customers who were acquired most efficiently (lower customer acquisition cost), or customers with higher order values (higher ROAS). Meta’s LAL algorithm then matches these higher quality seeds’ activity within the platform to build a higher quality LAL audience.
With these tactics, marketers can regain some control over their acquisition marketing efficiency and align their audience strategy with the goals and objectives of their marketing programs. These tactics can be impactful for not just advertising in Meta, but also others like Google, TikTok, Amazon and more. However, building and maintaining such systems can be very expensive and require hard-to-find talent at the intersection of machine learning, MarTech, and identity and data orchestration.
The solution for this is a new category in the MarTech space: Prospect Data Platforms, such as Angler AI, democratize the capability for growth marketers at consumer brands of any size. With Angler, you can create predictive LAL audiences to feed to walled garden ad platforms beyond Meta for distribution. Or provide smarter seed data for the ad platforms to optimize their LAL audiences, all at a fraction of the cost of hiring a full-time data scientist or engineer.