Dr. Ramla Jarrar of MASS Analytics on what a retail brand should demand from a marketing mix modeling solution, and how that changes for one banner, several banners or a wholesale network.
- →Why the best MMM solution for a retail brand depends on how the business is built
- →What to demand for a single banner, a multi-banner parent and a wholesale network
- →Three drivers of retail sales that a model must include besides media
- →Six questions to ask any provider before you shortlist
Every vendor shortlist I have seen for a retail MMM starts with features. Instead, the right place to start is the structure of the business.
A buying group can serve hundreds of independent operators. A regional chain can run four banners for four different customers. A wholesaler can sell to people who then sell to the shopper. Each structure therefore asks something different of the model.
Retailers often hesitate over marketing mix modeling because they think their business is too complicated for it. However, in twenty years of building these models I have found the opposite. The structure they see as an obstacle is what gives the model something to learn from.
Four things a retail brand should demand from any MMM solution
Model at the level where you make decisions, which in retail is usually the store. Give a separate answer to each level of decision maker. Let you define the KPI that matches how you sell. And model the other things that move retail sales: distribution, weather, promotions and the effect of one category on another.
Hold every provider to those four before you look at a single feature.
Any provider can show you a feature list. Instead, ask them which store, which decision maker and which KPI the model is built around.
Why the best solution depends on how your business is built
Across our retail engagements the same three structures keep turning up. Each asks a different question of the model, so each needs something different from the solution you choose.
| Structure | Decision the model must support | What the solution must do |
|---|---|---|
| Single banner, many stores | Where to concentrate spend and where to pull back across the store estate | Model store by store, then report at banner level |
| Multi-banner parent company | How to split budget across banners, and how each banner mixes its channels | Give corporate and each banner its own answer from one model |
| Wholesale and independent network | Which operators respond to your marketing support | Measure operator purchases, segmented by operator type |
Ask which structure you are before you ask which tool to buy.
Single banner, many stores: model store by store
This is the simplest setup. One brand, one media plan, many stores. It is also where the biggest mistake happens, because a banner level model gives you averages, and averages hide the stores that matter most.
Stores serve different catchment areas and different customers. A leaflet campaign can drive strong footfall in a suburban store and do almost nothing in a dense urban one. Only a store level model can show that difference, and that difference tells you where to concentrate spend and where to pull back. The same applies to digital too. A national view of Meta spend is too generic to act on at any level that matters.
Store level modeling also gives the model more to learn from. Take 100 stores and 18 months of weekly sales. That is 7,800 data points. By contrast, three years of national weekly sales gives you 156.
The leaflet case
The leaflet case is the one I use most often, because the logic behind the cut was so reasonable. A large US retailer with hundreds of stores looked at its leaflet distribution cost by area. Some areas were expensive to reach, so they stopped distributing there and saved $1.3 million. Nobody asked how those areas responded to leaflets, only what they cost.
When the store level model landed, it attributed $14.5 million in lost revenue to that cut. In some stores leaflets barely moved sales. In others, however, they were among the highest return activities in the mix. A national average could not have told those stores apart. The lesson the client took away went beyond leaflets: no budget line gets cut again without a measured view of what the cut will cost. You can read the full story in Leaflets: What a $14.5M Mistake Teaches Us About Retail Measurement.
Demand a solution that models store by store and still reports at banner and national level. Weekly sales by store, plus your media data and outside factors, is the core input.
100 stores and 18 months of weekly data give a model 7,800 observations. Three years of national data give it 156.
Multi-banner parent company: one model, two decision makers
When a parent company runs several banners, two groups ask two different questions. Corporate asks how to split the total budget across banners. Each banner team asks which channels work for its own customers and how to spread spend across its stores.
One model should answer both. If a solution gives only one level, one of those teams is left without an answer. Or you end up with two models that disagree, and the meeting turns into a debate about the models instead of the decision. I have sat in that meeting. It never ends with a budget decision.
At the top level, the model shows which banner returns more on its media. If Banner A returns more than Banner B, that is a signal for how corporate budget should move. At banner level, each team sees what works in its own context.
Do not average premium and value stores
Banners differ in position and customer, so their response to media differs too. The same is true inside one banner. A premium urban format and a value suburban format rarely respond to the same channels in the same way. Average them and you optimize for a customer who does not exist.
A good solution lets you compare store segments. It shows which channels support your premium position and which drive volume at the value end. Those are budget decisions with direct profit consequences.
Corporate needs the split across banners. Each banner needs its own channel mix. A solution that delivers one without the other has answered half the question.
Wholesale and independent networks: a KPI that fits a B2B sale
Wholesale is where most providers go quiet, so here is the direct answer. Your marketing influences independent operators, and they make the buying decision, not the shopper. The volumes, the variability and the role of promotions differ from a chain that sells through its own tills.
Consequently, that changes what you measure. Instead of till sales or footfall, you measure the purchases of the operators in your network. The data and the KPI change. However, the modeling method stays the same. Operator type becomes a segment, such as size, format and buying pattern.
In fact, a wide mix of operators helps here. More variation between operators gives the model more to separate what drives purchases from what does not.
Test the provider early. Ask how they would define the KPI and the operator segments. If the answer is simply “we will use sales,” keep looking.
If a provider cannot tell you how operator size, format and buying pattern enter the model, they have not modeled a wholesale network before.
Three things every retail solution must model besides media
In retail, media explains the smaller share of what moves sales. A solution that sees only media will give credit to the wrong things. Three drivers matter most.
Distribution and the base
Base sales are the sales you would earn without current media. Across the brands we have modeled, recorded in the MASS Analytics Benchmark Database and adjusted for market and competitive context, they typically represent 50 to 70 percent of total sales. New stores, closures and range changes all move the base. Our Comprehensive MMM Guide covers how the base is built.
Say a chain opens many stores in a year. Sales then rise for reasons that have nothing to do with media. A model with no distribution variable hands that growth to whatever media ran at the time. Media ROI then looks better than it is, and the next budget follows the wrong signal.
The base also records what happens when media is cut for a long time. One retailer I worked with went off air for a year, came back for six weeks, then cut media again for eighteen months. They suspected the absence of media was behind the slide in sales but had no way to prove it, so the debate went round in circles. Eventually, the model settled it. It showed a 25 percent decline in sales in the mid-term and a 60 percent erosion in base demand in the long term, and tied both to the cuts. The full case is here: How We Quantified a 60% Erosion in Base Demand the Retailer Missed. A model that cannot see the base cannot warn you about this.
Weather and other outside factors
For seasonal retailers, weather is a driver in its own right. In one engagement, the model gave media credit that weather had earned. Adding temperature and seasonal variables cut the margin of error by 30 percent and corrected a 10 percent media overvaluation. You can read it in How We Helped a Retailer Reduce Margin of Error by 30% With Holistic MMM.
Similarly, promotions, pricing and competitor activity belong in the same group. Ask which outside factors a provider includes as standard.
Halo across categories
A promotion on one category can lift the baskets around it. If a customer comes in for your promoted beverages, do they also buy bakery and deli? However, a model that reads only the targeted category misses that. A solution built for retail should measure the effect of category level marketing on adjacent categories. That can change the return you see for an entire channel.
A chain that opens stores in the same year it runs heavy TV will see TV take credit for the openings, unless distribution is in the model.
Six questions to ask a provider before you shortlist
- At what level do you model my business, and can you do it store by store?
- Can one model give corporate and each banner its own answer?
- How would you define the KPI for my business, and can it change to fit how the business sells?
- Which distribution and store opening data go into the model, and how?
- Which outside factors do you include as standard: weather, promotions, pricing and competitors?
- Can you measure the effect of one category on another?
The general buying criteria still apply, such as how often the model refreshes, who owns it at the end and what it costs in total. Our 2026 buyer’s guide covers those.
Frequently asked questions
The best MMM solution for a retail brand is the one that fits how the business is built. It should model at store level, give separate answers to each level of decision maker, let you define the KPI that matches how you sell, and include distribution, weather, promotions and cross-category effects. Judge any provider against those four points before you compare features. You can see how we approach it on our marketing mix modeling for retail page.
Model at the lowest level your data allows, then report upward. A national model gives averages, and averages hide which stores and channels drive results. A store level model keeps that detail and can still roll up to banner and national views.
Yes. One model can show corporate how to split budget across banners, and show each banner its own channel mix. Do not average the banners together, because different positioning and customers produce different responses to media.
It can. The KPI changes from consumer sales to the purchases of independent operators, and operator type becomes a segment, such as size, format or buying pattern. The method stays the same, but the provider must be able to define the KPI and the segments for a B2B sale.
The core input is weekly sales by store, plus your media data, promotions, pricing and distribution changes such as store openings. Outside factors such as weather and competitor activity add accuracy. Three years of weekly data is the usual starting point, and a store level panel can often work with a shorter history.
The question to ask before you shortlist a provider
Write down which of the three structures you are. List the decision each level of your business has to make, and hold every provider to that list.
Check three things in every answer. Does the solution model at the level where you decide? Does it include the drivers beyond media? Finally, does it define the KPI the way you sell?
For the full method, see our Comprehensive MMM Guide. To see how these six answers would look on your own banners or stores, book a demo with MASS Analytics.
Before you sign with any provider, ask one question. If the model told you to cut one banner’s budget tomorrow, would you trust it enough to act?
This article is part of a six-part series for retail marketing leaders. The next one covers what it takes to get started: the roles, the data and the first 90 days. Read the full series.
- ✓A $1.3 million leaflet saving cost one retailer $14.5 million in revenue, and only a store level model could see it.
- ✓Averages hide the stores that matter most, so model store by store and report upward.
- ✓One model should serve corporate and each banner without averaging the banners together.
- ✓In a wholesale network the KPI is operator purchases and operator type is a segment.
- ✓A solution that sees only media gives credit to the wrong things, so demand distribution, weather, promotions and cross-category effects in the model.
Previous article: Reading The Signals: ROI, Response Curves, and Synergy.
Next article: Getting Ready: The Three Roles, The Data, And The First 90 Days
Continue reading the series
Six articles taking you from the measurement problem to practical readiness — written for retail marketing leaders.