What is Marketing Mix Modelling?

If you run a brand spending a million pounds or more on marketing every year, the question of where that money should go is the most important one you ask. Most founders are answering it with attribution data, gut feel and the recommendations of whichever channel agency shouts loudest. Marketing Mix Modelling, or MMM, is the discipline that gives you a more honest answer.

This guide covers what MMM is, how it works under the hood, where it fits alongside your existing measurement, and what it can do for a growth-stage business. The numbers cited are taken from real client work and from Nielsen's industry benchmarks; if anything, they sit on the conservative side.

What MMM is

Marketing Mix Modelling is a statistical method that quantifies how each of your marketing activities contributes to sales. The term goes back to Neil Borden's 1964 paper on the "marketing mix", which gave us the four Ps. The modelling part came later, in the late sixties and seventies, when economists at companies like Procter and Gamble started applying regression analysis to spend and sales data.

The premise is straightforward. You spend money across many channels. TV, paid social, paid search, out-of-home, retail media, podcast, email, influencer, sponsorship, and all the price and promotional decisions that interact with those channels. Sales happen. Somewhere in that pile of inputs and outputs is the truth about what is making the cash register ring. MMM finds the truth by looking at patterns in your historical data.

Where attribution credits the last click before a purchase, MMM looks at every input that might have influenced demand, isolates the effect of each one, and tells you what each pound spent has returned. The output is a clean read on channel ROI, on diminishing returns, on cross-channel halo effects, and on what your budget should look like if you were starting from scratch.

The core insight is that MMM separates signal from noise in real-world data. It cannot do that perfectly. But it can do it well enough to make a much better allocation decision than you can make on intuition.

How it works

There are three stages to a build: getting the data ready, fitting the model, and interpreting what comes out.

Data

The model needs at least two years of weekly historical data. That covers your marketing spend by channel, your output metric (usually revenue, sometimes app installs or bookings or leads), and a context layer with anything that influences sales beyond marketing. Price changes, promotions, distribution shifts, weather, seasonality, competitor activity, press, stock outages. The richer the context layer, the cleaner the read, because the model has more variables to attribute residual variance to.

One of the things that makes MMM well-suited to where measurement is heading is that it uses data already inside your business. No cookies, no pixels, no user-level tracking. That means it carries on working as third-party tracking degrades, which it has been doing rapidly since Apple's ATT framework removed visibility on more than 60% of iOS users.

Model

The mathematical engine is regression analysis. In plain language, regression finds the best-fit pattern through a cloud of messy data points. Two years of weekly data gives you 104 observations. If you have, say, 15 input variables and one output variable, that is 1,664 data points and around 1.4 million unique pairwise combinations. From those combinations, the model finds the coefficients that best explain how the inputs combined to produce the observed sales.

What makes MMM more sophisticated than a basic regression is that it uses curves rather than straight lines. Real channels exhibit diminishing returns. Your first £100k on paid search might return four times the revenue. Your second £100k might return two and a half times. Your third might return one and a half. The model captures that curve for every channel, which is the part that makes the reallocation insight valuable.

It also captures lag effects (your TV burst this week is still driving search clicks next week) and adstock (an ad seen this week has a residual effect for several weeks afterwards). These are the mechanics that let MMM read brand-building work, not just performance media.

A well-built MMM will predict your historical sales within a 5 to 10 percent error band. If it cannot, something is missing from the input data and the modeller needs to go back and find it.

Outputs

What comes out of an MMM is what makes it valuable to a founder:

• Channel ROI: how much revenue each channel returned per pound spent, isolated from the noise of seasonality, price, weather, competitors and the rest.
• Diminishing returns curves: how the ROI of each channel changes as you scale spend up or down.
• Halo effects: how channels lift each other. The brand-search interaction is famously meaty, where TV or OOH activity lifts the conversion rate of paid search by 20 to 40%.
• Scenario planning: what happens to revenue if you shift £500k from one channel into another, or grow the budget by 25%.
• Optimal allocation: a clear recommendation on where each pound should go to maximise return.

The maths itself is now widely available. Meta's Robyn and Google's Meridian are open-source MMM packages that handle the modelling work. The hard part is no longer the algorithm. It is everything that surrounds it: getting clean data in, and getting trusted decisions out.

Why MMM matters now

MMM has been in and out of fashion since the seventies. Right now, it is back, and the reasons are structural.

The first is that attribution is breaking. The tracking pixels that powered a decade of digital marketing are looking through a smaller window onto reality every year. Apple's ATT removed iOS visibility. Chrome is winding down third-party cookies. Privacy-by-default is the operating model of every major browser. Attribution measurement is becoming less reliable at the moment founders need it most.

The second is that marketing has gone fully cross-channel. The brands growing fastest are running paid social, connected TV, retail media, influencer, podcast and OOH at the same time, weaving them into integrated programmes. Attribution can only see a slice of the digital paid stack. A model that captures every channel in one read is the only way to compare them on a like-for-like basis.

The third is that growth has got harder. The conditions of the last decade (cheap capital, easy attribution, scaling on a single channel) have ended. The next stage of growth means knowing where the money should go with real evidence behind it. That is a strategic allocation question, and MMM is the tool that answers it.

Who uses MMM

Historically, big FMCG dominated MMM. Unilever, Nestle, Reckitt and Procter and Gamble have all been using it for decades, because the volumes were big enough that small percentage improvements in mix translated to material profit. The discipline has now spread across most categories with serious marketing budgets.

• Drinks and spirits brands such as Diageo, Heineken and the LVMH portfolio.
• Beauty and personal care houses including Charlotte Tilbury, L'Oreal and Estee Lauder.
• Retail, ecommerce and DTC scale-ups crossing the £1m+ media spend threshold.
• Financial services, where buying journeys are long and channels overlap heavily.
• Travel, hospitality and leisure, where seasonality and weather make untangled reads difficult.
• Telecoms and automotive, both of which run multi-channel programmes with long consideration cycles.

The common thread is a marketing budget worth optimising, multiple channels in play, and sales data that can be aggregated to a weekly view. If that describes you, MMM is on the table.

What MMM is good for, and what it isn't

It is worth being clear on the boundaries. MMM is a strategic tool. It answers questions about how the budget should be allocated across channels, how to scale spend, and what the impact of brand-building work has been. It is excellent for those questions and not much help for anything else.

What MMM does well:

• Measures channel-level ROI across paid media and brand investment, on a like-for-like basis.
• Identifies diminishing returns and saturation points so you know where to stop investing.
• Quantifies cross-channel halo and synergy effects.
• Estimates the impact of price, distribution, weather, seasonality and macroeconomic factors.
• Gives you scenario plans for future budget allocation.
• Justifies investment into harder-to-track channels (TV, OOH, podcast, sponsorship, influencer brand work).

What MMM is not for:

• Optimising creative or audience inside a single platform. That is the job of attribution and platform-level testing.
• Real-time campaign tweaks. MMM is refreshed quarterly or annually.
• Forecasting next week's revenue.
• Telling you which keyword to bid on.

The mistake founders make is to expect MMM to do everything. It does not replace attribution. It sits above it. MMM tells you how much should sit in each channel. Attribution then tells you how to spend that money inside each channel. The two get sharper together.

MMM and econometrics: same thing

Econometrics is the academic discipline of applying statistical methods to economic data. Marketing Mix Modelling is the commercial application of econometrics to marketing. Same maths, same methods, same outputs. Some agencies use the word "econometrics" because it sounds more rigorous and lets them charge a premium. Others prefer "MMM" because clients understand the term. If a vendor tells you econometrics is fundamentally different from MMM, they are either confused or selling something. Industry bodies like the Marketing Accountability Standards Board use the two interchangeably.

The data prep problem

Here is the open part of the industry that vendors don't always talk about. The modelling itself takes about 20% of the time. The other 80% is data preparation. Pulling spend data from every platform, normalising it to a consistent grain, aligning currencies and time zones, handling missing weeks, building derived variables, validating against finance, reconciling to ground truth.

This is why traditional MMM costs £150,000 to £300,000 and runs 6 to 12 months. Most of that budget and time goes into wrangling spreadsheets. The maths, by comparison, is almost a footnote.

The data prep work is the part of the workflow that has changed most in the last two years. Pattern recognition, automated mapping and AI-assisted validation can compress that 80% down to a small fraction of what it used to take. The maths itself has not changed; the speed at which you can get clean data into it has. This is the technology behind Hanya's 12-week build, and the economic reason traditional MMM is being unbundled.

Where MMM falls down

It is not magic. The honest limitations:

• It needs at least two years of weekly data with reasonable variation in spend. New channels with three months of history cannot be modelled credibly.
• It cannot measure what it cannot see. Unmeasured factors like PR coverage, viral moments and cultural events show up in the baseline and inflate it.
• It assumes the future will roughly resemble the recent past. Step-changes in market conditions degrade accuracy until the model is refreshed.
• It is overkill below a million pounds a year of marketing spend. Below that, the cost of the build outweighs the value of the optimisation.
• It depends on internal appetite to act on the findings. Bad implementations stall when leadership is not aligned on using the results, no matter how good the model is.

What MMM unlocks for a growth-stage founder

The reason MMM is worth the investment is what it lets you do next.

The first lever is reallocation. Most brands have at least one channel they should be pulling money out of and at least one they should be putting more into. Doing only that, with no extra budget, typically lifts ROI by around 20%. That number is the basis of the 20:12 Guarantee that underpins every Hanya engagement.

The second lever is scaling. Once you can see the saturation curves, you know how much further you can push each channel before returns dilute. Most growth-stage brands are materially underinvested in their best channels. They could grow spend 20 to 30% without ROI compression and don't realise it.

The third lever, and the one that gets genuinely interesting, is unlocking channels you previously couldn't justify. Brand-building, top-funnel video, podcast, sponsorship. The channels that work but cannot be cleanly attributed. MMM gives you the evidence base to invest in them, which opens the next stage of growth that performance-only stacks tend to leave on the table.

Stack the three. A brand spending £3m a year at 3:1 ROI typically grows to around £3.75m of spend at 4:1, returning roughly £15.5m instead of £9m. That is £6.5m of incremental revenue, on a budget that has only grown by a quarter, in the same business with the same product. The growth was sitting in the allocation question all along. MMM made it visible.