Marketing mix modeling for retailers: what it measures and what it misses
Short answer: marketing mix modeling estimates how much each channel contributes to sales from two years or more of weekly data. It is built to split a budget between channels. It can’t tell you which store, which billboard or which radio station earned the money.
Retailers turn to MMM because it covers every channel, offline included, without tracking anyone. It is a good tool for one question: how to split next year’s budget between channels. This guide covers what it needs, what it costs, and the three things it can’t see when your revenue happens in stores.
What is marketing mix modeling?
Marketing mix modeling (MMM) is a regression model. It takes your weekly sales and your weekly spend by channel, adds control variables such as price, promotions, seasonality and weather, and estimates how much of the sales each channel explains.
The output is a contribution and an ROI per channel, a response curve showing where extra spend stops paying, and a budget optimizer that suggests a new split.
What MMM needs
| Requirement | Typical threshold |
|---|---|
| History | 18 to 24 months of weekly data at least. Meta’s Robyn guide asks for two years of weekly data, or 4 to 5 years if monthly. |
| Observations | 80 to 100 or more data points for stable results. |
| Variables | About ten observations per variable in the model. |
| Spend variation | Each channel must move up and down over time. A flat line can’t be measured. |
| Budget size | Under about $1 million a year in media, samples are often too small. |
| Time to first results | 4 to 16 weeks, depending on the tool and data quality. |
What MMM costs
| Option | Cost |
|---|---|
| Open source (Google Meridian, Meta Robyn) | Free license. In-house build estimated around $125,000, plus $22,000 a year to maintain. |
| Self-service SaaS | About $24,000 to $60,000 a year. |
| Hybrid platforms | About $30,000 to $150,000 a year. |
| Managed consultancy | $50,000 to $200,000 or more per engagement. |
What MMM does well
- One view of every channel. TV, radio, billboards, direct mail, search and social in the same model.
- No user tracking. It works on aggregate data, so privacy rules and cookie loss don’t break it.
- Long-term planning. Response curves show where each channel saturates, which helps set the annual split.
What MMM misses for a store network
- The store. MMM reads sales at the national or market level, week by week. It can say billboards return a given amount. It can’t say which three of your twenty boards did the work, or which stores benefited.
- Channels that move together. Local media often runs at the same time as a promotion or a seasonal peak. When spends rise and fall together, the model can’t separate them. New or small channels lack the variation to be measured at all.
- Cause and effect. MMM finds correlations. Both Meridian and Robyn documentation recommend calibrating the model with experiments before moving large budgets.
- Timing. Results arrive weeks after the data closes. By then the flight is over and the money is spent.
- Smaller chains. A regional chain with less than two years of clean data, or a modest media budget, often can’t feed a reliable model.
None of this makes MMM wrong. It answers the channel question, not the store question.
MMM vs. in-store measurement
| Marketing mix modeling | In-store measurement (Kairos) | |
|---|---|---|
| Question answered | How should I split budget across channels? | What did this campaign, panel or station earn in my stores? |
| Unit of reading | Channel, week, market | Store, panel, station, campaign |
| Data needed | 2+ years of weekly sales and spend | A sensor in each store, from the first campaign |
| When you get results | Weeks after the period closes | During the campaign |
| Method | Statistical model on aggregates | Real store visits matched to ad exposure |
| Best for | Annual budget planning | Steering spend inside each channel |
Using both
Use MMM once or twice a year to set the budget for each channel. Use in-store measurement to steer inside each channel, week by week: keep the boards, stations and placements that reach your real store visitors, and cut the ones that don’t.
The two feed each other. Store-level contribution, measured on real visits, is the calibration data an MMM lacks. On performance budgets, Kairos also runs exposed versus non-exposed tests to read incremental visits, the experiment both Meridian and Robyn recommend.
Start with your own numbers: calculate how much of your media budget reaches people who never visit your stores, or see how in-store attribution works.
Frequently asked questions
What is marketing mix modeling in simple terms?
Marketing mix modeling is a statistical model that looks at your weekly sales and your weekly spend by channel over two years or more. It estimates how much each channel contributed to sales, after removing the effect of price, promotions and seasonality.
How much data do you need for marketing mix modeling?
Plan for at least 18 to 24 months of weekly data. Meta’s Robyn guide asks for a minimum of two years of weekly data, or four to five years if you only have monthly data. A common rule is about ten weekly observations for each variable in the model.
How much does marketing mix modeling cost?
Open-source tools such as Google Meridian and Meta Robyn are free to license, but an in-house build is estimated at around $125,000 plus $22,000 a year to maintain. SaaS platforms run about $24,000 to $150,000 a year. Managed projects cost $50,000 to $200,000 or more per engagement.
Is marketing mix modeling better than multi-touch attribution?
They answer different questions. Multi-touch attribution follows clicks of individual users online. Marketing mix modeling reads aggregate sales against spend, so it includes offline media. Neither one counts the people who actually walked into your stores.
Can a small retail chain use marketing mix modeling?
Often not reliably. With less than two years of clean weekly data or less than about $1 million in annual media spend, the model has too few observations to separate one channel from another. Measuring visits store by store works from the first campaign.
Sources: data requirements from Meta Robyn and Google Meridian documentation as summarized by BlckAlpaca, Triple Whale and Improvado; pricing from Improvado’s 2026 MMM provider review and the Recast build-versus-buy calculator. Ranges are indicative.
See what your media earns in store.
Bring your current media plan. We’ll show you what each channel contributes to store revenue, store by store.