In a quarterly business review, the CFO of a regional retail group asked a straightforward question: "We spent eight million dirhams on outdoor and broadcast last quarter. What did it actually drive?" The marketing team had dashboards showing website sessions, social engagement, and last-click conversions from digital. None of it touched the outdoor or TV spend. The room went quiet. The honest answer was that nobody knew, not because the channels were not working, but because nobody had built the measurement to find out.
That question does not come from a difficult CFO. It comes from a reasonable one. And the uncomfortable truth is that for most marketing teams, including sophisticated ones with good analytics infrastructure, the honest answer is that they do not have one. Not because the spend is not working. Because the measurement was never built to see it.
Marketing Mix Modelling is the technique that answers that question. It has been around for decades. For most of the last ten years, it looked unnecessary. The conditions that made it look unnecessary have changed significantly and probably permanently. This post explains what changed, what MMM actually does, and whether it belongs on your measurement agenda.
The Measurement Confidence Gap
For roughly a decade, digital marketing had something that seemed like a complete measurement solution. Pixels tracked users across the web. Cookies followed them from ad impression to purchase. Platform APIs reported conversion events in near real-time. It was not perfect, but it was specific enough to justify budget decisions and report ROI with apparent confidence.
In that environment, MMM looked slow, expensive, and unnecessary. Why model the relationship between TV spend and sales when you could just track every click? MMM became something large FMCG brands did once a year for above-the-line planning and not something a digital-first team needed to think about.
From my own work with clients across the region, the attribution problem shows up most visibly not as a dramatic collapse but as a quiet accumulation of doubt. Campaigns report strong platform ROAS figures (Meta and Google both claiming credit, often for the same customer) while the business results tell a murkier story. For one client in the consumer services space, the core challenge was not offline measurement at all; it was that digital channels could not be cleanly separated from one another. Every platform claimed contribution. The aggregated claimed attribution consistently exceeded what the actual revenue figures could support. Nobody was lying, each platform was measuring what it could see, but the overall picture was incoherent.
That experience pointed to something broader. If we cannot cleanly attribute within digital, we certainly cannot attribute across the full media mix. The signal loss problem is not just about TV and outdoor sitting outside the measurement framework. It begins earlier, inside the digital ecosystem itself, where platform-reported numbers and actual business outcomes have quietly diverged without anyone declaring it a crisis.
Each of those events removed a layer of visibility. Cookies that persisted for months now expire in days. Mobile advertising IDs require explicit opt-in that most users decline. Walled gardens share less data than they used to, and what they share is modelled rather than measured. The attribution picture most digital teams look at today is incomplete in ways that compound quietly, until someone asks the TV question.
MMM never went away. The conditions that made digital attribution seem sufficient have gone away. That is a different thing entirely.
What MMM Can Answer That Attribution Cannot
The most common misunderstanding about MMM is that it competes with digital attribution. It does not. They answer different questions at different speeds. Digital attribution tells you which ad drove which conversion within a trackable digital journey. MMM reads the aggregate relationship between all marketing spend, including TV, outdoor, radio, and print and business outcomes over time.
That distinction matters enormously in the GCC context.
The shift toward digital advertising in the GCC is real and well-documented: digital now accounts for around 68 percent of total ad spend across UAE and Saudi Arabia. But the remaining share is not trivial, and it is concentrated in channels that are structurally difficult to measure. Outdoor advertising across the GCC is approaching a billion dollars in annual spend, with Saudi Arabia and the UAE as the dominant markets, and is forecast to nearly double by 2030. Television remains the second-largest conventional channel in the region by spend and OOH, according to Ipsos, is still not adequately measured across MENA.
In my own work with an automotive client in the region, the majority of the media budget sat in outdoor and broadcast, not as a legacy decision, but because the category demanded it. High-consideration purchases with long research cycles and a physical showroom endpoint are not won through performance marketing alone. The digital channels were measurable, vocal about their contribution, and reported regularly. The outdoor and TV spend sat largely unexamined. That asymmetry, the significant investment in channels that resist easy attribution, is not unique to automotive, but it is particularly visible there, and particularly acute in a region where the measurement infrastructure for offline media is still catching up to the scale of investment in it.
The practical consequence of signal loss is not that digital attribution is now useless. It is that it is now directional rather than definitive for strategic decisions. For tactical optimisation within a well-tracked channel, it is still useful. For the question of whether to shift budget from digital to TV, it no longer provides reliable answers. That is where MMM picks up.
Digital attribution tells you which ad won the race. MMM tells you whether you entered the right race in the first place.
How MMM Works
MMM is a statistical technique that models the relationship between marketing inputs and business outcomes using aggregate, time-series data. It does not track individuals. It does not require cookies, pixels, or user identifiers. It reads patterns in data that already exists in most organisations: weekly spend by channel, weekly or monthly revenue, pricing information, promotional activity, and external variables like seasonality.
At the end of a well-run MMM project, the outputs typically look like this: a decomposition showing what percentage of revenue each channel contributed over the measurement period; a marginal return curve for each channel identifying the point at which additional spend starts to yield diminishing returns; and a scenario planning tool that lets you model what would happen to projected revenue if you shifted budget from one channel to another. Those three outputs answer the CFO’s question directly: what the TV and outdoor spend drove, whether it was efficiently deployed, and what the right mix looks like going forward.
Three concepts are worth understanding before you commission an MMM project or evaluate one someone else has built:
Why Right Now: Open Source Changed The Economics
Until recently, MMM was effectively gated behind specialist econometrics consultancies with proprietary methodologies and fees to match. A full engagement could run to six figures and take several months. That pricing excluded most mid-market businesses and most in-house teams from the conversation entirely.
Two open-source releases have changed that significantly: Meta's Robyn in 2021 and Google's Meridian in 2024. Both make the underlying statistical methodology transparent and freely available to any team with the analytical capability to run them.
The release of these frameworks does not mean MMM is now easy. Data preparation (ensuring spend data, sales data, and external variables are clean, consistently defined, and correctly aligned temporally) is still where most project effort sits. And interpreting the outputs requires analytical judgement that comes from experience. What the frameworks have done is remove the black-box methodology and the associated pricing. A data team with BigQuery and Python capability can now run a credible MMM without engaging a consultancy.
Is MMM Right For Your Organisation?
MMM is not appropriate for every organisation. The conditions for a useful project are specific, and commissioning one without meeting them produces unreliable outputs and erodes confidence in the technique. Before considering it, apply this check honestly.
| MMM makes sense when… | MMM is not the right tool when… |
| ✓ You have 12+ months of spend and sales data | ✗ You have less than 6 months of consistent data |
| ✓ You run both online and offline channels | ✗ You only run digital campaigns with good tracking |
| ✓ You need to justify or reallocate a large media budget | ✗ You need to optimise individual ads in real time |
| ✓ Signal loss is degrading your digital attribution quality | ✗ Your digital attribution is still accurate and trusted |
| ✓ You have Python or R capability in-house or with a partner | ✗ You need a result in days rather than weeks or months |
The data requirement deserves particular emphasis. MMM is a time-series technique that learns from observing how changes in marketing inputs correlate with changes in outcomes over time. That requires sufficient history and sufficient variation. If your spend patterns have been consistent for twelve months with no significant changes to the channel mix, the model has little variation to learn from. If you have less than six months of data, it does not have enough signal to separate channel effects reliably from seasonal noise.
MMM is also not a complete measurement solution in itself. It cannot help optimise individual ad creatives or audience targeting decisions; those remain in the domain of platform analytics and experimentation. It does not provide in-flight guidance; the model is backward-looking and requires historical data, not live campaign signals. And it can be disrupted by major structural changes in the business: a significant product launch, a market entry, or a sharp competitive shift can confuse the time-series patterns the model depends on. Used correctly, MMM tells you how to allocate a budget strategically. It does not tell you how to execute a campaign once that budget is set.
Three Things I Would Tell Any Team Before They Commission An MMM Project
Not a step-by-step guide: the frameworks have documentation for that. These are the three things that tend to determine whether an MMM project delivers something useful or produces outputs that sit in a slide deck and are never acted on.
The data audit is the project.
Most MMM work in terms of effort and elapsed time, happens before the model is even opened. Collecting spend data by channel at weekly granularity, aligning it to sales data in the same period, and adding external variables like promotions and seasonality is where teams underestimate. A model fed misaligned or inconsistent data will produce plausible-looking but unreliable outputs. Before you touch Robyn or Meridian, spend two to four weeks on the data.
Define one question before you define a model.
MMM can answer several different questions ranging like channel contribution for the first time, budget reallocation, scenario planning, leadership justification of above-the-line spend and so on. Each requires a different model setup. The teams that get the most from MMM are the ones who go in with a single, specific question. The ones that struggle commission a model and then try to work out what to do with the outputs.
Validate before you act and be honest about the uncertainty.
MMM outputs are estimates with ranges, not precise measurements. Before making significant budget decisions, test the model's predictions against a holdout period it was not trained on. If the fit is reasonable, you have a basis for directional decisions. If not, investigate the data before proceeding. And when you present findings to leadership, present the uncertainty honestly. A model that overstates precision loses credibility the first time a prediction misses.
I have not run a full MMM engagement directly, my measurement work has sat closer to digital attribution, channel mix planning, and the moments where clients started asking questions that attribution alone could not answer. MMM is a capability I am actively building toward, and this post is my attempt to make the case clearly: why MMM has re-emerged, what it can and cannot do, and when a GCC team should seriously consider it. If a client asked me today, I would bring in a specialist partner for the modelling work while contributing the commercial framing, the media input structure, and the interpretation of outputs for business decisions, which is where the practitioner value sits regardless of which tool runs the regression.
The measurement gap created by signal loss is real, it is growing, and it is not going to reverse. The organisations that recognise this early and build MMM into their measurement infrastructure, alongside digital attribution, not instead of it, will be making budget decisions with evidence that competitors without it simply cannot produce. The technique has existed for decades. The conditions that make it necessary have arrived. That is the case for looking at it now, not later.