Kroger had AI score 1,934 ads before buying media. It was 81% accurate
Kroger, Vidmob and MMA Global say a model trained on a year of Kroger ads forecast online conversion with 81% accuracy before a cent of media was spent.
| Company | The Kroger Co., US grocery retailer; study run with creative-data vendor Vidmob and trade body MMA Global, announced jointly on 30 September 2026 (Business Wire release) |
|---|---|
| Data | 1,934 video and image ads from Kroger campaigns on Meta and Google's Display and Video 360; model trained on all of 2025, tested on a separate Q1 2026 dataset (Business Wire release) |
| Result | predictive creative scoring forecast e-commerce conversion with 81% accuracy, as the three parties state it; the public material does not define the accuracy measure (Business Wire release; MMA Global study page) |
| Result | ads that followed the model's recommendations averaged 4x higher conversion rates, at up to 70% lower cost per conversion (Business Wire release; MMA Global's page says up to 4x) |
| Modelled, not achieved | moving media toward higher-scoring ads could generate up to 2.2x as many conversions from the same budget (Business Wire release) |
| Finding | ads built around people sharing and enjoying food beat product-first and transactional shopping imagery for Kroger's online sales (Business Wire release) |
| Not disclosed | what the study cost, Kroger's media spend, how much budget was actually moved, and whether scoring is now a standing step in Kroger's ad approvals |
The Kroger Co. used an AI model to score 1,934 of its own video and image ads for how well they would sell online, and the model forecast e-commerce conversion with 81% accuracy before media was bought, according to a study Kroger announced on 30 September 2026 with the creative-data vendor Vidmob and the trade body MMA Global. Ads that followed the model's recommendations averaged four times the conversion rate of those that did not. Kroger has not said what the work cost or how much of its budget it has moved as a result.
Ad platforms tell you how an ad did after you have paid for it. The question Kroger's marketing team was trying to answer is the one every retailer's marketing manager faces each Monday: which of these new ads should get the money before any of it is spent?
What Kroger actually did
The three parties took a full year of Kroger's 2025 campaigns on Meta and on Google's Display and Video 360, which together produced 1,934 ads. Vidmob's software tagged each ad by what was in it: the visual treatment, the message, how the story was structured, how prominent the branding was, and whether people appeared and what they were doing. The model then learned which of those attributes went with online purchases in Kroger's own sales data, not with clicks or video views.
The important step is the next one. Rather than stop at explaining last year, the model was tested on a separate set of Q1 2026 campaigns it had never seen. That is the 81% figure: a forecast checked against later results, not a fit to old ones.
The model was judged on ads it had never seen. That is what makes the 81% worth reading at all.
What changed
According to the release, ads that matched the model's guidance averaged four times the conversion rate across both platforms, with cost per conversion up to 70% lower. The single clearest pattern was about content: for Kroger, ads showing people sharing and enjoying food sold more online than ads leading with the product or with the act of shopping.
The third figure is a projection. Moving media spend toward higher-scoring ads could, the parties say, produce up to 2.2 times as many conversions from the same budget, without making a single new ad. Nothing published says Kroger has done this at scale or what happened when it did.
What did it cost?
Nobody has said. The release gives no study budget, no Vidmob licence fee and no Kroger media spend, so neither the 70% saving nor the 2.2x can be turned into money. Vidmob sells this scoring as software to large advertisers; for a mid-sized retailer, the realistic cost is not a licence but the time of one person to tag ads and pull results, which is what the last section is about.
What the numbers do not tell you
Four things are missing, and a buyer should ask for each.
First, what 81% accuracy means. The public material does not say whether it is the share of ads correctly sorted into winners and losers, or a statistical fit. The full whitepaper sits behind a registration form; we did not obtain it.
Second, the wording moves. The release says aligned ads averaged four times the conversion rate; MMA Global's own page says up to four times. Those are different claims.
Third, cause. Ads that matched the guidance may also have run in better campaigns, to warmer audiences or at better times of year. The release describes significance testing but not how those factors were separated.
Fourth, voice. No one from Kroger is quoted in any of the documents; every quote is from Vidmob or MMA Global. Kroger put its name to the joint announcement, which is why we treat the figures as Kroger's, but the explanation of them is the vendor's.
Does this apply to a 200-person retailer?
The method does; the model does not. Kroger had nearly 2,000 ads and a full year of sales tied to each. A 200-person retailer running a few dozen ads a year has too little data for a model, and Kroger's food-sharing finding is about Kroger's shoppers, not yours. What transfers is the discipline: tag every ad by what is in it, match each to the online sales it drove, and look for the pattern before the next budget is set.
What to do this month
Export the last 12 months of ads from Meta Ads Manager with their purchases and cost per purchase. In a spreadsheet, add four columns of your own: people or product, lifestyle or price message, video or still, brand shown early or late. Sort and look. With a few dozen ads, any pattern is a hunch, not a finding, so turn the strongest one into a split test using the platform's own experiment tool before moving money. Our suggestion, not Kroger's: decide what to stop funding first, because the cheapest gain in Kroger's study came from moving existing spend, not from making more ads.