See what each channel really adds, where extra spend stops paying, and how to split your budget.
Move £5,000 a month from Meta prospecting to YouTube, then confirm it with a lift test.
Brightpace runs fixed-price marketing mix modelling projects with Google’s open-source Meridian model. The model shows what each channel really adds and where extra spend stops paying, ends with a budget plan, and is handed over to you with the code.
The question platforms can’t answer.
Platform reports and attribution follow individual clicks and views. They miss what they can’t track, credit sales that would have happened anyway, and each one marks its own homework. None of them can tell you whether the next £10,000 should go to Meta, Google, YouTube or somewhere else.
A marketing mix model answers that from your own data: weekly sales and spend for every channel, alongside the things that move sales without any advertising, such as seasonality, pricing and underlying demand. No individual tracking is needed, so it isn’t affected by cookie consent or ad blockers. We bring your awareness campaigns into the model too (brand video, upper-funnel social, TV, out-of-home and sponsorships) alongside your performance channels, because their effect shows up later in searches and sales that click tracking never credits. Our founder has built marketing mix models in-house, using Google’s Meridian, for multi-channel businesses with highly seasonal demand. Wondering whether you’re ready? Read is marketing mix modelling worth it for your budget?
Before crediting any channel, the model accounts for seasonal patterns, holidays, price changes and underlying demand, using Google search interest as a guide. Otherwise advertising gets credit for sales the calendar would have brought anyway.
Brand search mostly captures demand that other channels created, so we model it separately from the rest of Google. Lumping them together makes Google look like it created sales it only collected.
Channels are split only where the data can support it. Splitting one platform into many small pieces gives each piece too little signal to estimate reliably.
Recent weeks are held back to test the model’s predictions, every result is sense-checked, and lift test results are used to calibrate it where they exist.
The model shows where to look and a lift test proves it. For geo tests, Google’s free GeoX tool plans the test and feeds the result back into the model.
Built with Google’s Meridian. No licence fees, and the model, code and assumptions are handed over.
We check you have enough history, channels and spend variation for a model to be worth building. If not, we’ll say so and suggest lift tests instead.
Data cleaned and joined, the model built and then tested against weeks it hasn’t seen, with lift test results used to calibrate it where you have them.
Each channel’s contribution and return, its diminishing-returns curve, a budget plan and the next lift tests to run. Try the maths behind a lift test with our incrementality calculator.
Whether you have the history, channels and spend variation a model needs, before anything is built.
Weekly sales and spend for every channel, plus seasonality, pricing and demand signals.
Tested on held-back weeks and calibrated with lift tests where you have them.
Channel returns, diminishing-returns curves and a reallocation plan, explained without jargon.
Where the model is least certain, turned into a test plan.
Open-source and yours to keep, with the assumptions documented.
Contribution and return for every channel, measured against the sales you’d have had anyway.
Diminishing-returns curves for each channel, so you can see which ones still have room to grow.
A reallocation plan for the same total budget, and what changes if the budget goes up or down.
The channels where the model is least certain, which become the next lift tests.
Quoted up front once we’ve seen your data and channels: data work, model, validation and a read-out with budget recommendations. Refreshes every three to six months are quoted separately.
Not sure you’re ready? Start with a short data review. We’ll tell you whether a model is worth building yet, or whether lift tests and tracking fixes should come first.
MMM is the top layer of our measurement work. The foundations are accurate tracking (including server-side tracking) and honest reporting on blended numbers such as MER; see MER vs ROAS vs POAS for why they matter.
Anything else, ask the person who would run your account.
Ask a question →Marketing mix modelling (MMM) is a statistical model of your sales over time. It estimates how much each marketing channel contributed, after allowing for things like seasonality, pricing and underlying demand, without tracking any individual person. You use it to decide how to split budget across channels.
Attribution follows individual clicks and views to give credit for each sale, so it misses anything it can’t track and tends to over-credit the last touch. MMM works from total sales and total spend over time, so it captures the effect of channels attribution struggles with, such as video, TV and upper-funnel social. It answers a different question: how should the budget be split.
Yes, and it’s one of the main reasons to use it. Brand video, upper-funnel social, TV, radio, out-of-home, sponsorships and PR can all go into the model, as long as there’s a record of when they ran and how much was spent (or impressions or ratings for TV). Their effect usually shows up later, in brand searches and sales across other channels, which click-based attribution never credits. Where a result matters for budget, we confirm it with a lift or geo test.
At least a year of weekly sales and spend data, and ideally two, so the model can separate seasonality from the effect of advertising. It also helps if spend has varied over time: a model can’t learn much from a channel whose budget never changes.
Usually only once you run several channels with meaningful spend. If most of your budget is in one or two platforms, lift tests on those platforms will tell you more, sooner and for less. We’ll tell you which fits after a short data review.
It isn’t a crystal ball: it estimates what happened over the period it was built on, so big changes in your market or offer mean it needs refreshing. It guides budget between channels, while day-to-day bidding and creative still happen inside each platform. And it needs good data: gaps, renamed campaigns and missing spend weaken it, so cleaning the data is part of the project.
Google’s Meridian, an open-source model. Geo tests, which run ads in some areas and not others, are a good way to check the model, and Google’s free GeoX tool can plan them and feed the results straight back into Meridian. You don’t pay for software licences, and the model, code and assumptions are yours to keep.
We hold back the most recent weeks of data and check that the model predicts them well, sense-check every channel’s result against what we know about how it works, and calibrate the model with lift test results where you have them. A model that can’t pass those checks doesn’t get used for decisions.
Usually a few weeks from receiving clean data to the first read-out, depending on how many channels you run and how tidy the data is. Pulling and cleaning the data is often the longest part.
It’s a fixed-price project, quoted up front once we’ve seen your data and channels. The price covers the data work, the model, the validation and a read-out with budget recommendations. Refreshing the model each quarter or half-year is quoted separately.
Every three to six months for most businesses, and after any big change in budget, channels, pricing or your market. Each refresh adds the latest data and any lift test results, so the model keeps pace with how your marketing is actually performing. Refresh early if its predictions for recent weeks start missing actual sales, or a lift test result disagrees with it: both are signs it has drifted.
Yes. MMM works from your sales and spend data, so it gives an independent view of every channel, whoever runs them.
Tell us which channels you run and how much history you have, and we’ll tell you whether a model is worth building.
