Meta has spent the last few years creating the core of what many sales teams still consider their work. Target structure, bid adjustment, placement decisions, audience segmentation. Machines are doing it more effectively now, and the gap between where most teams spend their time and what actually drives performance is widening.
I spent last week at the Meta Performance Marketing Summit, and that gap was the topic of every presentation. The central message was consistent: performance marketing is moving from manual development to AI-driven systems, and the marketer’s role is evolving accordingly. Organizations that understand this build themselves. Many are not.
The performance engine has been rebuilt from the ground up
At the conference, two systems dominated the technical discussion, and understanding them is the fastest way to understand why the platform behaves so differently now than it did even eighteen months ago.
Meta launched a major update to Lattice in February, and we wrote about it at the time with our initial thoughts on what it meant for marketers. At the conference, Meta gave a full picture yet of how it actually works. Previously, different models were developed for different purposes on their own: one for engagement, one for conversion, one for accessibility. What Lattice does is allow everyone to learn from shared behavioral data simultaneously. Shopping behavior now improves predictive engagement. Engagement signals improve conversion prediction. The whole system is smart because every part of it is learning from everything at the same time.
The strategic impact is significant. Meta systems are increasingly developing fully across the full funnel, while many marketers are still running integrated campaigns with separate KPIs and disconnected creative plans. The platform learns the cross-funnel faster than most organizations that work in reverse. That’s not a gap you can fill by working hard within the existing model.
Andromeda is sinking. Historically, retrieval systems identified relevant ads, and rating systems determined what should be displayed. With Andromeda, self-retrieval is now AI-personalized. Meta can check which ads a user is likely to be interested in before ranking. They support this with a critical infrastructure: 10,000x more computing power has been added to the retrieval systems through the Nvidia partnership. Electronic investment makes the desire clear. This is a major rebuild, not an upgrade update.
Taken together, what Lattice and Andromeda describe is a platform that no longer responds primarily to advertiser input. It makes complex decisions on its own, on top of anything the media group touches.
That change has a direct effect on how media groups should spend their time. The things that have historically alienated skilled media buyers, property targeting, bid rigging, audience segmentation, property complexity, are being automated far more effectively than humans can manage manually. New differentiators are creative quality, first-party signal quality, conversion data integrity, product feed quality, and measurement complexity. If your team spends most of its time on things that Meta has done for itself, you’re setting up the wrong layer of the stack.
Three inputs now determine whether you win
If the platform makes decisions that media groups used to make, the question becomes: what do you put in front of it? Three ideas came up frequently at the conference, and all three are under-invested by many marketers.
The first one is creative. The meta was clear: stop trying to find one winning ad and start building systems that generate and transform creative signals continuously. Their Catalog Product Video format is a very clear proof point, delivering 20% more conversions per dollar and 33% higher incremental conversions on Reels placements. Meta generative AI tools can now generate thousands of creative combinations from existing assets with limited manual effort. Changing the performance of this requirement is important. Creative strategy is no longer about periodic production cycles or heroic legacies. It is modular, repeatable, and signal driven. Organizations in the best position to win are those that treat creativity as a continuous feed into the AI process, not as the result of a short cycle generated campaign.
The second input is creator content. This was one of the most violent sections of the conference trade, and one that may have disrupted the way agencies and brands are currently structured. Meta has redesigned the Creators Marketplace to integrate directly with custom audiences, Ads Manager, and performance signals. The evaluation process for creators has moved from fan counts and engagement metrics to performance opportunities, audience overlap, and business outcomes. Affiliate Ads deliver 19% lower CPA, 13% higher CTR, and 71% improvement in brand awareness when combined with BAU campaigns. The takeaway from Meta was clear: creator content is no longer an influencer strategy that sits alongside your paid social work. It’s a core operational infrastructure, and the future requires integrated creators and paid community teams, creator scoring systems, prospecting, and growth frameworks built specifically for creator programs.
The third input is product dataand it was the most underrated theme of the conference. Your product catalog is no longer a back-end commerce infrastructure. It’s the raw material that powers AI-driven personalization, powerful creative generation, and contextual commerce experiences. Meta has described situations in which Meta AI recommends products contextually based on behavior, preferences, saved content, and previous purchases, with future capabilities in all advanced product information, category comparison, brand vs. price analysis, and automated product video creation. Feed quality directly determines referral quality, recommendation quality, and creative quality. Many marketers still treat catalog management as a technical activity. It is a strategy, and the gap between those who manage it and those who don’t will show in the results.
Why most marketers don’t see what really works
Meta was unusually specific about measurement at the conference, and it was one of the most important areas of the trade they covered. Even if you’re fine-tuning your art, your creator plan, and your product data, you may still be measuring the Meta offer incorrectly.
Their data shows that 31% of the incremental conversions driven by Meta are not extended to other channels. That number represents a structural problem in the way many organizations measure performance, with real consequences for budget allocation, channel investment, and strategic decision-making. When you’re optimizing for click-to-last ROAS, you’re making decisions based on an image that systematically underestimates what Meta does.
The reason this happens is that most of the Meta impact comes early in the journey: discovery, cultural influence, shifts in search behavior, assisted conversion. Users encounter something in Meta and convert elsewhere. Standard field reporting cannot capture this, and most measurement models are not calibrated to accurately account for it. H&M ran Conversion Lift tests to benchmark their MMM models and saw a 3x improvement in incremental ROAS across all key markets over two years. Profit at that scale is the difference between underinvesting in an active channel and really understanding what drives growth.
Meta’s recommended methodology includes growth testing, test-based measurement, Conversion Lift, Brand Lift, MMM measurement, and predictive LTV integration. A major challenge, which Meta has embraced directly, is organizational change. The tool is available. What many organizations lack is the organization of their proper use: finance and marketing work as linked activities, evaluations made instead of periodic, and leadership that corresponds to the growing business growth rather than the volume of clicks. The remaining problems are structural, and most organizations do not restructure to solve them.
Bottom line
The consistent contention of this conference was that the paid community is becoming program management, not campaign management. The strategic role of agencies and leaders is shifting to developing learning programs, integrating measurement frameworks, creating pipelines, improving signal quality, and designing AI workflows.
What will separate the organizations that win from those that fall behind is not the size of the budget or the reach of the platform. It’s the speed at which they can re-engineer the model already built by Meta. Technology is very much there. The organization’s will to use it is flexible.
