Video-creators and AI-generated ads help satisfy platform algorithms’ hunger for engagement, earning more delivery at lower cost. But for big brands, a small ROAS gain that erodes brand equity or fails to drive latent sales makes no sense. Instead, Advertisers should manage to ROI, balancing delivery cost efficiency with sustained sales performance.
Platforms grade for engagement, not sales growth
Big platforms like Meta, YouTube and TikTok are highly effective at driving audience reach and holding attention at a scale no brand could buy on its own. Their advertising systems match ad creative and viewers based on what the platform can see: engagement behaviors and profile data among logged-in users. The system rewards creative diversity and heavy rotation to keep users’ attention and engagement. Admetrics reports that high-performing accounts on Meta often maintain at least 8-12 active creative variations per campaign, refreshing 25-30% of their library monthly.
Creative is the new targeting, and two things are accelerating this: Platforms are pushing short-form video and creator content, because it delivers cultural relevance and attention at scale. Generative AI has made it absurdly easy to produce twenty variants a week instead of two a quarter. The two go hand in hand.
The resulting creative surge answers algorithms’ desire for volume and diversity. Yet that surge strains longstanding principles of creative effectiveness. Brands don’t grow sustainably by chasing ROAS, or whatever delivery metric an algorithm rewards this quarter.
“Brands don’t grow sustainably by chasing ROAS, or whatever delivery metric an algorithm rewards this quarter.”
Incrementality-testing firm Haus analyzed hundreds of geo-holdout experiments and diminishing-return studies around Black Friday through Cyber Monday (BFCM) 2025. A rule emerges: sales impact from upper-funnel media arrives after the ad runs, so real-time ROAS understates it. For example, counting the period after a test ends, efficiency improved by more than 75% overall, where a $1.00 incremental ROAS became $1.75. In tests that started before BFCM, 41% of the incremental value showed up during BFCM itself. In 44% of experiments, efficiency at least doubled once that lag was measured. Lag effects during BFCM were nearly triple those in ordinary periods, which added about 26%.
Counting the lag lifted incremental ROAS by 75%
Incremental ROAS, Black Friday to Cyber Monday 2025 experiments
During the test only
Including after the test ended
What Brands Actually Need
Brands grow through relentless consistency: distinctive assets, executed in a recognizable way, for years, so they’re encoded in memory before buyers start shopping. However, recent surveys from Kantar and System 1 indicate that only about a quarter of creator content successfully links back to the brand.
“Only about a quarter of creator content successfully links back to the brand.”
One case summarized in a framework built on Byron Sharp and Jenni Romaniuk’s Ehrenberg-Bass Institute research: a $750 million brand tightened up its distinctive asset system and saw on-shelf recognition jump 16 points and branded search share climb 14%, in under a year. Les Binet and Peter Field found a similar pattern in 996 IPA-tracked campaigns: roughly 60% of budget to brand-building, 40% to activation. Overweight activation, and the gains fade quickly once spending stops. Underweight brand-building, and you keep drawing down the same pool of in-market buyers.
What Platforms Can’t See
Platforms are excellent at reporting what happened inside their walls: impressions, clicks, watch time, even attributed conversions. What’s much harder for anyone, including the platforms, to see is causality: what would have happened anyway, without the ad. This is a limit of observational data. Facebook’s own research team, with Northwestern University’s Kellogg School of Management, found that standard attribution methods can wildly overstate true lift. One randomized controlled test measured a 77% lift, while a simple exposed-versus-unexposed comparison showed 416%, over five times the true number.
Standard attribution overstated lift by 5x
Measured ad lift for the same campaign, by method
Exposed vs unexposed comparison
Randomized controlled test (true lift)
Source: Facebook and Kellogg School of Management, Northwestern University
“One randomized controlled test measured a 77% lift, while a simple exposed-versus-unexposed comparison showed 416%.”
Brands Must Manage Video Creator Content for ROI
Advertisers must own their business performance, and report results to the board, whose language is sales. Amidst the power of the platforms and surge of the Creator Economy, it’s important to stay focused on the business results. That means use platform delivery metrics to manage efficiency (ROAS). Use incrementality tests, marketing mix models and validated sales predictions to determine what content actually drives sales. Measure every creative twice: does it earn delivery, and does it build brand and sales? Good campaigns deliver not just sales, but critical learnings for the next cycle. Learnings make for better briefs to creators, better prompts for AI tools, and sharper intel for anyone with a remit to maximize ROI and business growth.