Start with the boundary of the outcome
A retail media campaign can be associated with sales of the advertised product, other products from the brand and purchases outside the retailer. A platform report may cover only part of that universe. Before comparing it with an MMM, define exactly which sales the model includes.
- Define the sales universe
- Reconcile overlap
- Separate direct and halo scope
- Test the total effect
Write down products, retailers, geographies, channels and time windows. Distinguish a same-retailer brand halo from an off-retailer effect. The measurement framework helps organise metrics with different scopes rather than forcing them into one return table.
A useful model starts with a consistent outcome and a decision it can inform. The marketing mix modelling learning path connects this idea to the assumptions, practical checks and next steps.
To work through the connected decisions, read Retail MMM: account for stockouts and distribution changes and MMM calibration: keep an experiment evidence register.
Reconcile a simple overlap example
Suppose a fictional retailer reports £80,000 of attributed product sales and £20,000 of attributed brand-halo sales. If the £80,000 already includes all brand products, adding the halo line again double-counts £20,000. If the lines are disjoint, the total may be £100,000. The labels alone do not settle the issue; the data definitions do.
Even a correctly reconciled attributed total is not automatically incremental. Customers may have bought without the ad, switched between products or moved their purchase between retailers. A total-business outcome can help capture those movements, but an observational model still needs credible assumptions.
Account for retail operations
Retail media often runs alongside discounts, merchandising, distribution changes and product availability. Keep those calendars visible. A sponsored placement during a major promotion does not provide clean independent variation merely because spend is recorded in a separate column.
Decide whether the budget question concerns the retailer, the brand or the whole portfolio. A gain for one product may partly displace another. If you model multiple outcomes, reconcile their definitions and avoid summing overlapping estimates. Reporting a range for total portfolio effect can be more useful than a precise-looking split unsupported by the data.
Use experiments to challenge the boundary
A geo or audience experiment can test a defined retail-media intervention where the design and data permit it. Specify which purchases can be observed, how cross-market shopping or other spillover is handled, and whether the outcome captures substitution. The incrementality guide explains why a credible counterfactual matters more than attribution labels.
For a proposed allocation, compare incremental contribution over the relevant sales universe, not incompatible platform ROAS figures. State what remains unmeasured, including off-retailer sales if unavailable. Do not fill that gap with a guessed halo multiplier presented as measured evidence.
The final decision record should include the outcome boundary, overlap reconciliation, promotion and availability assumptions, experiment scope and budget constraint. Those details make the next model refresh comparable to the current one.
Sources and further practice
This article’s worked example is illustrative, not a measured client result. Use the linked methodology to check the assumptions before applying it.