MMM gets described as complex, expensive, and the preserve of big brands with big budgets.
None of that is really true anymore.
Here’s a plain-language explanation of what it is, what goes into it, and what comes out the other side.
Thought Leadership - 8 min read - MASS Analytics
Marketing Mix Modeling has a branding problem. The name alone tends to conjure images of economics PhDs, six-figure consultancy engagements, and results delivered six months after anyone can act on them. Ask most marketing directors at a mid-sized retailer whether they’ve ever used MMM and you’ll often get a variation of: “We looked at it once, seemed like a lot of effort for something we couldn’t really use.“
That reputation was probably fair, once. It isn’t anymore. And given what the technique can actually do for a retail business, the gap between perception and reality is worth closing.
The Simplest Way to Think About It
MMM is a statistical method that looks at your sales data, usually at a weekly level, over two to three years, and unpicks all the different factors that drove those sales. It separates what happened because of your marketing from what happened because of price changes, seasonal patterns, competitor activity, the economy, and everything else. Then it tells you, with precision, what each factor contributed.
Think of it as the statistical truth-teller your business has always needed.
That “everything else” is what makes it different from the attribution models and dashboards you already have. Those tools measure marketing in isolation. MMM measures marketing in context: accounting for the full, complicated reality of what drives consumer behaviour.
What Goes Into the Model
The inputs fall into a few categories. Your team doesn’t need to gather all of this manually, as much of it can be connected automatically, but understanding what’s being included helps you trust what comes out:
Media
Weekly spend by channel: digital, TV, radio, leaflets, out-of-home.
Sales
Revenue or volume data, ideally at store level, week by week.
Pricing
Price changes, promotions, and discount activity over time.
External
Economic indicators, competitor spend estimates, weather where relevant.
Business events
Store openings, website outages, range changes, extraordinary events.
Seasonality
Holiday periods, calendar events, natural category cycles.
The model is built over approximately three years of history. That length matters: you need to have seen enough variation in your spend and in external conditions for the model to reliably separate causes from coincidences. A period that includes at least one significant change in marketing mix, one period of competitive activity, and the natural rhythm of seasons gives the model enough texture to work with.
How the Model is Built
Once the data is assembled, the modeling process runs in a logical sequence:
1 – Establish the base
The model first estimates what sales would have been if all marketing had been turned off, which is the baseline driven purely by existing brand equity, consumer habits, and product demand. This is the floor everything else is measured against.
2 – Decompose the drivers
Each factor: media channels, promotions, pricing, competition, seasonality, is quantified as a contribution to sales above or below that base. Price increases show up as negative contributions. A well-timed leaflet campaign shows up as a positive one.
3 – Account for lag and synergy
Not everyone who sees a TV ad buys tomorrow. The model accounts for that lag effect (the weeks or months between exposure and purchase) and for synergies between channels. A TV campaign running alongside paid search performs better than either would alone.
4 – Validate and fit
The model is tested against actual historical sales to check its accuracy. A well-built model should explain the vast majority of variance in sales, and the fit is visible as you can see the model tracking closely against the real sales curve.
5 – Run optimisations and scenarios
Once the model is built and validated, it becomes a forward-looking tool. Change the budget. Shift spend between channels. Model a competitor doubling their activity. The model tells you what would happen, before you commit a pound.
What Comes Out the Other Side
The outputs of a well-run MMM engagement are not a set of charts in a PowerPoint. They are decision-ready answers to the questions your business is asking:
ROI by channel
For every pound spent on each media channel, how much revenue did it return, and how has that changed year on year?
Sales decomposition
A week-by-week breakdown of exactly what drove your sales: base, promotions, each media channel, pricing, competition, seasonality.
Response curves
For each channel, the curve showing where you are on the spend-to-revenue relationship, and whether you’re under-investing, optimally placed, or in diminishing returns.
Scenario modelling
What-if analysis: if we cut this channel by 20%, increase that one by £500K, or hold flat, what does revenue look like under each scenario?
The Question of Speed
One of the most significant shifts in how MMM is applied today is around speed. The traditional model: data gathered manually, analysis run quarterly or annually, results packaged into a consultancy deck, is no longer fit for purpose in a world where media cycles move week by week.
The modern approach is Always-ON. Models are refreshed automatically as new data arrives, connected directly to data sources like sales platforms, media agency feeds, and external economic inputs. Measurement keeps pace with the decisions it’s supposed to inform, not lagging behind them by half a year.
A note on terminology:
- Marketing Mix Modeling, MMM, Media Mix Modeling, and econometrics all refer to the same family of techniques.
- The underlying method is regression-based statistical modelling, but you don’t need to understand the maths to act on the outputs.
- MMM is distinct from attribution modeling: attribution assigns credit to touchpoints a customer passed through; MMM explains what actually caused sales at a population level, including factors attribution can’t see at all.
What It Isn’t
MMM is not a magic box that produces answers with no input. The quality of the output depends on the quality, consistency, and completeness of the data going in. It also isn’t a replacement for human judgement; it’s a tool that makes that judgement better-informed. A model can tell you that leaflet distribution is underperforming in three specific regions; but it takes a person to decide how to respond.
It also isn’t something that should be run once and put on a shelf. The most valuable applications of MMM are continuous models that update as the business changes, and that are integrated into the regular rhythm of planning and budget decisions rather than pulled out for the annual strategy review.
The next article brings this to life with a specific example: a retailer who made a cost-cutting decision that looked entirely sensible, and what the model revealed when the numbers were properly examined.
Previous article: Why Your Marketing Data is Lying to You Next article: Leaflets: A $14,5M Retail Case Study
Continue reading the series
Six articles taking you from the measurement problem to practical readiness — written for retail marketing leaders.