Predictive Creative Scoring: What the Kroger, Vidmob and MMA Study Means
Media targeting has been heavily automated. Creative is still where most of the judgement calls happen, and where opinions often outweigh data. A study published on September 30, 2026 by Kroger, Vidmob and MMA Global tests whether that can change: can a model trained on a brand’s own creative data predict which ads will convert before any media is bought? Here is what the study found, what its limits are, and how agencies can use the idea without enterprise budgets.
What the study did
According to the announcement, the study analysed 1,934 video and image assets from Kroger campaigns on Meta and Google’s Display & Video 360. The model was trained on a full year of 2025 campaign data and then validated against a separate Q1 2026 dataset. That second step matters: it tests whether the model can predict future performance, rather than just explain past results.
The partners evaluated creative variables such as visual treatment, messaging, narrative structure, branding and on-screen human interaction. Kroger supplied live campaign data, Vidmob supplied its creative data platform, and MMA Global supplied its industry research standards.
The headline findings
- Predictive creative scoring forecast e-commerce conversion with 81% accuracy.
- Creative aligned with the model’s recommendations averaged 4x higher conversion rates across Meta and DV360, at up to 70% lower cost per conversion.
- Moving media toward higher-scoring creative could generate up to 2.2x as many conversions from the same budget.
The white paper adds useful context. The 81% figure refers to accuracy in classifying an asset as a low or high performer on the e-commerce conversion metric. The team tested 231 attribute levels, and 105 passed standard statistical testing. A guideline was only kept if three tests agreed: it moved real conversions, it held up in live creative, and the predictive model agreed. That process produced 22 validated guidelines: 17 for DV360 and five directional guidelines for Meta.
How to read these numbers
It is one retailer’s data
The model learned what works for Kroger’s audience, products and brand. The point of the study is that a brand’s own creative data can be predictive, not that Kroger’s rules will work for your client.
“Could” is doing some work
The 2.2x figure is a modelled estimate of what reallocation could achieve, not a measured result from a live reallocation.
Meta guidance was weaker
Only five Meta guidelines were classed as directional, against 17 validated for DV360. Results will vary by platform, and probably by how much the platform’s own algorithm already optimises delivery.
A vendor is involved
Vidmob sells creative analytics, so it has an interest in the result. MMA Global’s involvement and the out-of-sample validation add credibility, but treat it as a strong case study rather than an independent benchmark. Marketing Dive’s coverage frames it as part of Vidmob’s ongoing effort to show that creative quality can be measured.
Why agencies should care
The main lesson is methodological. Creative can be broken into attributes, those attributes can be tested against real outcomes, and the resulting rules can be checked against new data before you trust them. That turns creative feedback from “the client didn’t like it” into “assets with these features converted better for this account”.
It also changes the conversation about AI-generated creative. When production becomes cheap and fast, the bottleneck is choosing what to run. A scoring system, even a simple one, helps decide which variants get budget first.
A lightweight version for smaller accounts
You don’t need 1,934 assets or a data platform to borrow the approach:
- Tag every ad by a short list of attributes: product shown in the first seconds or not, person on screen or not, offer stated or not, brand shown early or late, format and length.
- Pull results for a meaningful period, using the conversion metric the business cares about, not just CTR.
- Look for patterns with enough volume behind them. Ignore attributes with tiny sample sizes.
- Turn patterns into draft guidelines, then test them on the next batch of creative before treating them as rules.
- Revisit quarterly, because audiences and platforms change.
What to do this week
- Pick one client with steady conversion volume on Meta or DV360.
- Export the last 6–12 months of ad-level results with conversion data.
- Create a simple attribute tagging sheet and tag the top and bottom performers first.
- Write three draft creative hypotheses and build the next round of ads to test them.
- Agree with the client which conversion metric decides the winner.
The bottom line
The Kroger, Vidmob and MMA Global study is one of the clearer public examples of creative being predicted, not just reported. The specific numbers belong to Kroger. The method of attribute tagging, testing against outcomes and validating on fresh data is something any agency can start using now.
Want a creative testing and scoring process for your Meta and Google campaigns? Book a free strategy call and we’ll show you how we would set it up for your account.
