Meta has shared some new insights into its evolving ad targeting systems, and how its expanding AI processing capacity is driving better results for advertisers through improved interest matching.

And advertisers have been taking notice. More and more Meta ad partners have reported improved performance, with AI targeting helping to find customers whom they otherwise would have missed.

In its new overview, Meta provides more insight into how its system works, and how it’s driving broader performance improvements across all of Meta’s ad offerings through ongoing improvement.

As explained by Meta:

“The Generative Ads Recommendation Model (GEM) is Meta’s most advanced ads foundation model, built on an LLM-inspired paradigm and trained across thousands of GPUs.  It is the largest foundation model for recommendation systems (RecSys) in the industry, trained at the scale of large language models.”

To be clear, Meta’s been using advanced machine targeting for ads for years, with its vast troves of audience interest and engagement data enabling Meta to more accurately identify user interests, and display relevant ads accordingly.

Indeed, before the latest wave of AI tools hit the market, Meta had already been using this same LLM-based approach to targeting for many years, but the re-framing of scaled data processing as “AI” has changed the paradigm around how this is perceived.

Essentially, Meta used to be criticized for facilitating psychographic targeting, based on the data that it has on its 3 billion users, including the Pages that they like, people that they’re connected with, interests, traits, etc.

But now, all of this is not only acceptable practice, under the banner of “AI,” but Meta’s data is also considered a major advantage. And with this in mind, after weathering all of that blowback, you can see why Zuckerberg is so keen to claim the title as the leader in the AI space.

Meta says that its latest GEM model presents a significant advance in its targeting systems, by using “model scaling with advanced architecture, post-training techniques for knowledge transfer, and enhanced training infrastructure to support scalability.”

“These innovations efficiently boost ad performance, enable effective knowledge sharing across the ad model fleet, and optimize the use of thousands of GPUs for training. GEM has driven a paradigm shift in ads RecSys, transforming ad performance across the funnel – awareness, engagement, and conversion – through joint optimization of both user and advertiser objectives.”

In summary: More people click ads, more ad customers sell stuff.

In terms of performance specifics, Meta says that its updated system is now:

• 4x more efficient at driving ad performance gains for a given amount of data and compute than its original ads recommendation ranking models. 
• 2x more effective at knowledge transfer, helping to optimize broader ad performance.
• Faster and more effective based on larger compute capacity, enabling more effective scaling of ad results.

“GEM is trained on ad content and user engagement data from both ads and organic interactions. From this data, we derive features that we categorize into two groups: sequence features (such as activity history) and non-sequence features (such as user and ad attributes – e.g., age, location, ad format, and creative representation). Customized attention mechanisms are applied to each group independently, while also enabling cross-feature learning. This design improves accuracy and scales both the depth and breadth of each attention block, delivering 4× the efficiency of our previous generation of models.”

So Meta’s ad system now has more systematic capacity, enabling it to process more information, and find more correlating data signs, leading to improved ad performance.

Which is also reflected in the performance data.

Meta has previously shared that advertisers utilizing its various AI–powered ad targeting options have seen notably improved ad performance, while it’s also revealed plans to eventually automate the entire ad creation process, using these evolving systems to essentially create your ad, optimize your targeting, and manage your budget, without you needing to do anything but input your product URL.

That’s how much faith Meta has in its ad systems to drive improved performance over time.

Meta’s GEM system works in tandem with Meta’s “Lattice” architecture, and its “Andromeda” models, which all play their own role in optimizing your Meta ad targeting.

• Lattice is what Meta calls its “ad library,” which powers ad ranking, ensuring optimal placement for each campaign
• Andromeda is Meta’s personalization model, which ensures ad relevance based on each user’s engagement history and interests  

In combination, these systems ensure greater ad relevance, utilizing Meta’s ever-growing tech stack to learn more about each user’s preference, and enhance targeting accordingly.

Which, again, at Meta’s scale, means processing a heap of data points, which can lead to highly accurate, highly valuable ad results.

I mean, back in 2015, reports suggested that Facebook already had enough data to infer virtually everything about you, based on your in-app activity.

That capacity has been super-powered by the latest AI models, leading to better ad performance across the board.

It’s interesting to consider Meta’s capacity in this respect, and it could be worth trying out Meta’s evolving AI-powered ad options, via Advantage+, to see what results you get.