Dentsu sees AI modeling reshaping retail media measurement
Agency executives want retail media measured beyond conversions. AI-powered modeling may change how brands evaluate retailer-reported results.
Dentsu commerce executive Aisha Khan is asking advertisers to view retail media as more than an efficient way to reach shoppers near checkout. At the recent Media Buying Summit in California, she argued that in-store advertising, stronger creative and more independent measurement belong in the same planning conversation. The measurement question may prove especially consequential. Khan pointed to advances in AI-powered marketing mix modeling, or MMM, as a reason advertisers could become less reliant on the conversion results reported by individual retail networks. This does not mean retailer attribution is obsolete. It means buyers may have another way to test whether media spending creates incremental demand rather than simply claims credit for sales.
Table of contents
Jump to each section:
- Why retail media needs a broader measurement lens
- Where AI-powered MMM changes the calculation
- The limits of retailer-controlled attribution
- What marketers should know about full-funnel retail media
Why retail media needs a broader measurement lens
Retail media's advantage has traditionally been proximity to the sale. Retailers can connect ad exposure with transactions inside their ecosystems, offering advertisers a level of purchase visibility that many other channels cannot match. But the same advantage can narrow the questions marketers ask, rewarding placements that are easiest to attribute rather than those that contribute to demand across a longer journey. Khan sees physical stores as a particularly underused advertising environment. She told the summit audience that physical retail represents a much larger share of sales than of retail media spending. Those percentages are her cited estimates, not independently established benchmarks.
More than 80% of sales versus less than 5% of retail media spending. Khan cited these respective figures for physical-store sales and in-store retail media, according to her remarks at the Media Buying Summit.
That imbalance, if it broadly reflects the market, suggests room for in-store screens and shopper experiences to develop. It does not establish which formats will work, which retailers can sell them profitably, or whether brands should shift budgets immediately. In-store creative still has to earn attention and fit the shopping environment. Buying closer to checkout is not the same as measuring the whole customer journey.
Where AI-powered MMM changes the calculation
Marketing mix modeling examines how different marketing investments are associated with business outcomes, often using aggregated data rather than tracing every individual conversion. Its appeal to advertisers is the ability to consider multiple channels within one framework, including media whose influence is difficult to capture in a retail network's own dashboard. Khan argued that AI has accelerated more agile MMM and incrementality approaches. Advertisers are developing measurement capabilities across retailers rather than accepting each seller's account of its own performance. Khan described the concern as retailers “having the retailers grade their own homework,” a critique of relying entirely on seller-supplied performance scores. The potential outcome is pressure on the premium networks charge for exclusive attribution data, although Khan presented that as a possibility, not an observed price decline. There is a concrete technology development behind that broader argument. In a September 2026 update, Google said its open-source Meridian MMM is gaining agentic assistance for data-quality checks, troubleshooting and model building. It also added brand-equity signals designed to examine upper-funnel investments and made its GeoX experimentation library broadly available. The tools are not a retail-media-specific solution, and Google did not claim they would replace a retailer's sales records. AI can make the modeling workflow easier to operate. It cannot automatically make the underlying evidence reliable. For marketers, the useful distinction is between automation and causality. Faster modeling can help teams revisit budget decisions more frequently, while experiments and sound input data still matter when they want to know what would have happened without an ad.
The limits of retailer-controlled attribution
Closed-loop reporting remains valuable. A network with logged-in shoppers, loyalty identifiers and transaction records can show relationships between exposures and purchases that are not visible to every advertiser or publisher. The limitation is scope: a result inside one retailer's environment may not capture category demand, later purchases elsewhere or the influence of brand campaigns on future behavior. A marketing team therefore needs to separate three questions. Did a campaign correlate with sales in a retailer's dataset? Did the campaign produce incremental sales? And did it change brand preference or shopping behavior beyond that single network? Those questions require different evidence, and one reporting interface may not answer all of them. Google's Meridian updates illustrate the direction of travel, not a completed solution. Including brand signals in an MMM can bring awareness investments into planning, while GeoX-style experiments can help test causal lift. But teams still need to scrutinize data coverage, model assumptions and the extent to which retailer transactions can be compared across channels. The wider industry is also trying to address inconsistent measurement. The Interactive Advertising Bureau has called for common definitions and more interoperable commerce-media tools. That standards work is distinct from AI modeling, but both respond to the same operational problem: retail media becomes harder to manage when every seller uses a different measurement language. Retail data can show what sold within a network. On its own, it cannot settle what would have sold without the advertising.
What marketers should know about full-funnel retail media
The immediate implication is not to cut lower-funnel retail media. It is to evaluate whether the current plan measures more than it can conveniently count. Separate attribution from incrementality. Ask which outcomes come directly from retailer transaction matching and which are supported by holdout tests, geographic experiments or independently calibrated models. Give in-store creative its own brief. Store screens and shopper messaging should have a role in the experience, with objectives that fit awareness, consideration or purchase rather than a default conversion-only metric. Compare networks through common definitions. Align time windows, sales coverage, new-customer definitions and evaluation methods before treating two retailer dashboards as directly comparable. Treat AI measurement as an analytical tool. Automated modeling workflows may reduce manual effort, but require teams to inspect source data, assumptions and the difference between modeled estimates and observed transactions. The strategic tension is that retail media has gained credibility by being measurable, yet its next phase may depend on making measurement less tied to individual media sellers. For agencies, that could turn the conversation from inventory access and reported return on ad spend toward creative strategy, experimentation and cross-channel planning. For advertisers, the aim is not to find one perfect number. It is to combine the retailer's view of transactions with evidence that can challenge it. The organizations that can connect those perspectives will be better placed to distinguish media that captures existing demand from marketing that helps create it.
