Measurement
Marketing mix modeling (MMM) services
Click-based tracking cannot see a billboard, a radio spot or the search someone made after seeing your ad on TV. MMM looks at the whole budget over time and estimates what each channel adds to sales.
What MMM answers, in plain terms
Picture two years of weekly figures: what you spent on Google, Meta, radio and billboards, and how many leads or sales came in. Some weeks spend went up and sales followed. Some weeks it went up and nothing happened. Ramadan, summer and the school holidays moved things too.
A model separates those effects. It estimates the baseline you would sell with no marketing at all, the extra each channel adds, how long that effect lasts after the spend stops, and the point where each extra unit of spend in a channel brings back less.
The output is a budget question answered with evidence: if you move money from one channel to another, what is likely to happen to sales, and how sure can we be?
When MMM is worth it, and when it is not
| Your situation | Is MMM worth it? | What to do |
|---|---|---|
| One or two online channels, modest budget | Not yet | Get attribution and CRM data clean first |
| Several online channels, steady spend all year | Sometimes | Run controlled tests; model if questions remain |
| Online plus TV, radio or billboards | Usually | Build the model on at least two years of weekly data |
| Less than a year of history | Not reliable yet | Start recording spend and results weekly, by channel |
| Large budget decisions every quarter | Yes | Model, then check the big calls with tests |
How we build a model
Gather the data
Weekly spend and activity per channel, leads and sales from the CRM, prices, promotions, holidays, and anything else that moves demand in your market.
Clean and check
Gaps, tracking changes, one-off events. Most of the time on an MMM goes here, and it decides whether the result can be trusted.
Build the model
With open-source methods such as Meta’s Robyn or Google’s Meridian, chosen for your data. Results come as ranges, never as a single exact figure.
Test it
Does it predict weeks it was not trained on? Do its findings agree with the lift tests we have run?
Plan the budget
Scenarios for next quarter: what happens to sales if each channel goes up or down.
Refresh
Every quarter, with the new weeks of data added.
Checking the model against real experiments
The strongest check on a model is a test. Switch a channel off in some regions and keep it on in others, or use the lift studies Meta and Google offer, and measure the difference in sales. If the model and the test disagree, the test wins and the model is adjusted.
Tests cost something: a few weeks of reduced spend somewhere. We plan them where the budget decision is big enough to justify the cost, and we use the results to make the next model run more reliable.
What the model gives you
Contribution by channel
The share of sales each channel drove over the period, with a range around it.
Return per channel
Revenue per unit of spend, on average and for the next unit spent.
Saturation point
Where extra spend in a channel stops paying for itself.
Carry-over
How long a channel keeps working after its spend stops, which matters for billboards and TV.
Budget scenarios
Expected sales under different splits of the same budget, ready for planning. See media planning.
Frequently asked questions
How much data does MMM need?
Ideally two years or more of weekly data per channel, with enough variation in spend to learn from. Less than a year rarely gives reliable results. If your history is short, the most useful step is to start recording it properly now.
Is MMM better than attribution?
They answer different questions. Attribution follows individual leads and suits steering campaigns week to week; our revenue attribution ties each lead to its campaign. MMM looks at totals over time and suits splitting the budget between channels, offline ones included. Use both.
Which tools do you use?
Open-source libraries such as Meta’s Robyn, Google’s Meridian or PyMC-Marketing. You are not locked into a vendor, and you can see how the model reaches its numbers.
Can a small business use MMM?
Usually not well. The method needs volume and variety in spending. With a smaller budget, clean attribution plus a few simple tests answer the same questions for less money. Our budget calculator is a better first step.
How often should the model be updated?
Every quarter, or after a big change such as a new channel, a price change or a new market. A model built on last year’s mix slowly loses touch with this year’s.
Last updated · By the Napexy team
Keep reading
Tell us where you want to grow.
A 30-minute call with a strategist. You get a plan in plain words and a fixed quote, whether or not you work with us.