Retail marketing spans TV, outdoor, digital, and in-store activation. MMM is the only measurement approach that accounts for all of these together and shows how they interact.
Retail is one of the original use cases for marketing mix modelling. FMCG and grocery brands were running MMM studies in the 1990s to understand whether TV or price promotion drove more volume through supermarkets. The fundamentals have not changed. What has changed is the media mix. Retailers now run TV alongside YouTube, outdoor alongside digital out-of-home, and national campaigns alongside hyper-local geo-targeted ads. That complexity makes proper measurement more valuable than ever.
What makes retail measurement hard
Retail operates across physical stores and online channels simultaneously. A customer might see a TV ad, search for the product online, visit the store, and buy there. That full journey is almost impossible to track at an individual level. Digital attribution tools only see the online portion, which means they systematically undervalue brand and offline media that drives in-store footfall.
Retailers also deal with competitor activity that directly affects their sales. If a competitor runs a big promotion the same week you run your campaign, your sales may dip even if your media was working well. MMM can incorporate competitor promotional data, which prevents the model from incorrectly blaming your media for the shortfall.
The typical retail media mix
Large retailers typically combine TV (including addressable and connected TV), outdoor and billboard, radio, digital (paid search, paid social, display), retail media (ads within platforms like Amazon or Criteo), and in-store promotions. Smaller retailers may focus on digital and outdoor. Whichever channels you run, the model needs spend data for all of them, including channels that feel difficult to quantify like outdoor.
- TV (linear, addressable, connected TV): strong at building brand awareness and driving short-term footfall for sale events
- Outdoor: particularly effective near store locations, best modelled using regional spend splits
- Paid search: captures demand at the bottom of the funnel, often picks up credit generated by TV
- Paid social: drives product discovery and supports new customer acquisition
- Retail media: effective for in-category conquest but often over-attributed by platform reports
- Leaflets and direct mail: still significant for grocery and DIY retailers, particularly for older demographics
Data sources for retail MMM
The ideal retail MMM dataset includes weekly sales by channel (in-store and online), weekly media spend by channel, promotional calendar (price cuts, sale events, 3-for-2 offers), price index data, distribution data (number of stores or product listings), competitor promotional data where available, and weather data for categories where it matters (barbecue season, cold and flu products, gardening).
Distribution changes are one of the most common sources of bias in retail models. If you opened 20 new stores last year, sales went up partly because of that. The model needs to know, or it will wrongly attribute that growth to media.
Retail seasonality patterns
Almost every retail category has predictable seasonal patterns. Grocery peaks at Christmas and bank holiday weekends. Fashion peaks in spring and autumn. DIY and garden peaks in spring and summer. Electronics peaks at Black Friday and Christmas. The model learns these patterns from the data, but you need to flag exceptional events separately. A particularly warm Easter or a late winter affects demand independently of media, and the model needs to know that so it does not confuse good weather with effective advertising.
Common measurement mistakes in retail
The biggest mistake is measuring online and offline independently. Digital teams often use last-click attribution for their channels, while brand teams use awareness metrics for TV. Neither gives you the full picture. MMM combines both views and shows how TV investment lifts not just brand awareness but actual sales, including in-store. Retailers who have done this comparison often find that TV and outdoor are significantly undervalued when measured in isolation.
A second common mistake is modelling at a national level only. Retailers with strong regional variation in store density or brand awareness will get better results from regional models. A national model can mask the fact that media is working very well in some regions and barely at all in others.
Using MMM for retail planning
Once you have robust model outputs, you can use them to plan media investment around your busiest periods more precisely. Instead of guessing how much TV to run in the lead-up to Christmas, you have data on the relationship between TV spend and incremental sales at different investment levels. You can also model the interaction between price promotion and media, which helps you avoid the common trap of cutting price at exactly the moment when media is already working hard.
Can MMM measure the impact of in-store activity like end-of-aisle displays?
Yes, provided you have data on it. If you can provide a weekly index of in-store promotional intensity (for example, the number of stores running a specific promotional mechanic that week), the model can estimate its contribution to sales alongside media channels.
Should we model each retail category separately?
It depends on how different the categories are. If they share media (a TV ad promoting the brand rather than a specific product), modelling them together often makes sense. If the categories have very different customer journeys and seasonality, separate models give more accurate results.
How do we account for media that runs at a local level, like outdoor near specific stores?
Regional spend splits are the most practical solution. Group your stores into regions that align with your outdoor buying regions, then run the model at that level. This takes more data and more time to set up, but it substantially improves the accuracy of outdoor and local media estimates.