AI in Retail Media: Optimise the Campaign, Then Prove It Moved the Business

By Kirill Naumov · 2026-10-24

Retail media has a tempting feature: the ad and the purchase can often be observed in the same retailer’s ecosystem. Put a product beside an ad, watch the dashboard fill with clicks and sales, and it is easy to feel the campaign has explained itself. It has not. The dashboard may be accurate and still answer a narrower question than the one a finance director is asking.

AI can help teams scan large reporting packs, spot unusual changes and suggest where to investigate. It cannot turn “a purchase happened after an ad” into “the ad caused the purchase.” That distinction is where useful measurement begins.

Attribution is not incrementality

Attribution assigns credit to an eligible interaction under a chosen rule and time window. Incrementality asks a harder question: what additional outcome happened because the campaign ran, compared with a credible estimate of what would have happened without it? Both views can be useful, but they are not interchangeable. IAB and IAB Europe’s 2025 guidance highlights credible counterfactuals, bias control and separating signal from noise. Less glamorous than a colourful ROAS tile, perhaps; rather more useful when someone asks whether to invest again.

Where AI earns its place

Give AI a bounded analytical job: flag campaigns that changed materially against a relevant benchmark; group products or audiences that deserve a closer look; surface possible pacing, availability or creative issues; and summarise a reporting pack as questions for a human analyst. Each suggestion should show the data and assumptions behind it. “Spend more because attributed ROAS is high” is not an explanation.

Before acting on a recommendation, check what else changed. Was there a promotion? Did stock availability move? Were the same products exposed through another channel? Did the comparison period include a seasonal spike? A model can help organise those questions, but it cannot make missing data appear by sounding confident.

Write the measurement brief before the prompt

For one campaign, agree on five things before launch: the business decision to improve; the outcome that matters; the comparison method; the other factors that must be recorded; and the person who can challenge the result. “More sales” is not yet a measurement plan. Decide whether the question is incremental units, new-to-brand customers, margin contribution or something else—and keep delivery metrics separate from business outcomes.

Choose a comparison that fits the decision and the available data: a holdout, matched markets, a counterfactual model or another defensible design. Agree on it before results arrive. If an experiment is too small to support a confident conclusion, say so. A modest test does not need to cosplay as a national randomised trial; a large budget decision deserves more than a screenshot.

Treat the recommendation as a hypothesis

Ask the system to identify a change, show the evidence and list what a human should verify. Keep a record of the recommendation, the evidence shown, the decision and what happened later. Review the cases where the advice was wrong, unhelpful or impossible to reproduce. Until the team can explain why a budget change is justified, let AI inform the decision—not move the money on its own.

Start with one category, one clear objective and a decision that can be reversed. Compare the AI-assisted workflow with the current process: time to prepare a review, issues found, quality of the questions and whether analysts agree on what the evidence supports. Then evaluate campaign impact with a measurement design appropriate to the spend and data.

The point is not to build a data-science department before buying a sponsored placement. It is to avoid mistaking a more polished report for a better business result. In retail media, AI can help find the next question. Measurement still has to answer it.

Explore the AI for Retail workshop → · Start an AI diagnostic

Further reading: IAB and IAB Europe’s guidance on incremental measurement · IAB Europe’s 2026 measurement standards · Walmart Connect’s measurement overview · Amazon Attribution.

Start a conversation →