Kroger, Vidmob and MMA Global published a rare piece of research on September 30: 1,934 live video and image ads from one retailer, scored for creative attributes and matched to ecommerce conversion data. The model was trained on a year of Kroger campaigns, then tested on ads it had not seen in Q1 2026. It sorted high and low converters with 81% accuracy, according to the whitepaper ‘Predictive Creative Scoring Leads to Real Ecommerce Outcomes’.
What the study actually found
The analysis covered 1,062 Meta ads and 872 Google DV360 ads. Vidmob coded each asset on message (brand tone, proof points, CTA) and craft (live action, talent, first-three-second attention). From 231 tested attribute levels, 22 guidelines survived: 17 for DV360 and 5 for Meta, with the Meta ones labelled directional.
Two headline numbers matter. About 50% of Kroger’s historical Meta and DV360 investment went to the weaker creative tier. Building new creative to the guidelines was associated with up to 4x higher conversion, while simply moving budget to existing high-scoring assets showed a potential 2.2x lift without producing anything new.
People beat glossy product shots
The clearest creative signal was human. Ads built around relatable human narratives consistently outperformed glossy product shots and transaction-focused creative. On Meta, 87% of assets with a relatable human scenario ranked above the median, versus 38% without one. Ads showing people reacting or enjoying the moment cleared the median 76% of the time, against 45% without.
The catch: only about a quarter of Kroger’s Meta assets contained a relatable human scenario. The strongest lever was also the least used.
Where the budget leaked
The whitepaper separates two effects. The 4x figure is what new creative built to spec could achieve; the 2.2x figure is the reallocation-only opportunity from shifting existing spend. On DV360, weaker creative also carried higher CPMs, so the low converters were more expensive to deliver per thousand impressions.
Some caveats are material. The 4x and 70% lower cost-per-conversion figures are described as both ‘up to’ and ‘average’ in different parts of the material. No confidence intervals are published, and the study is observational: high-scoring ads may have run in better placements or warmer audiences. Kroger plans a live head-to-head test to settle causation.
What to do before you kill a creative
- Audit existing Meta and DV360 assets by message and craft, not by gut feel.
- Tag whether each ad has a relatable human moment, then compare conversion rates.
- Shift spend to proven assets first, and run a holdout or split test before making new guidelines permanent.
The larger context: a Perion and EMARKETER survey found only 3.6% of marketers actively optimise creative performance. With AI making production cheap, the scarce skill is knowing which ad will actually convert.
Source: PPC Land
