MMM and Incrementality
Who needs them, who does not, and what they cost.
A $15k-a-month store once asked me for an MMM. A vendor had told them attribution was "dead" and modeling was the future. The vendor's minimum was $3,000 a month. I told them to keep the $3,000, turn off retargeting in Sweden for three weeks, and see what happened. Sales in Sweden did not move. That was their incrementality test.
The one-sentence version
MMM estimates each channel's effect from spend and sales patterns; incrementality testing measures it by switching something off. Both answer "did this actually cause sales?" Both need enough scale for the answer to be bigger than the noise.
The setup
Three tools, three questions
| Tool | Answers | Needs | Speed |
|---|---|---|---|
| Order-verified attribution | Which ad touched which order, every day | Tracking + surveys + orders | Minutes |
| Incrementality test | Did this channel cause sales at all | Enough regions or people to split, 2 to 4 weeks | Weeks |
| Marketing mix model | How much each channel contributes, including TV and retail | 12+ months of history, $40k+ a month spend | Weekly refresh |
Try it: 20-second check
Do you need MMM or incrementality yet?
Monthly ad spend
$25k
Paid channels
Offline, TV, retail or Amazon in the mix?
Fix attribution first.
At this spend, the noise in a model is bigger than the effect you are trying to measure. Order-verified attribution with surveys will answer 90% of your questions. Revisit at $40k a month.
MMM is the last tool you add. Not the first.
Why it matters
Models need ground truth
A model is only as honest as its inputs. Feed an MMM the platforms' self-reported conversions and you get a beautifully calibrated version of the same over-counting. Attribution is not replaced by modeling; it is what you calibrate the model against. Haus, Prescient and Northbeam all say some version of this in their own documentation.
Attribution: daily steering
Which creative, which campaign, which audience. Fast, granular, needs an order to exist.
Incrementality: the occasional audit
Is brand search really incremental? Is retargeting? One question, one test, a clean answer.
MMM: the annual plan
How should next year's budget split across channels, including the ones without clicks. Slow and strategic.
The Sweden store turned retargeting off for good, moved the budget to prospecting, and grew. Total cost of the test: three weeks of not paying Meta to show ads to people who were already buying.
The TrueROAS version
Attribution first, model when it earns its keep
Every plan gets order-verified attribution with surveys. Larger budgets can add True Horizon, an MMM calibrated against those verified orders, for planning and offline channels. The cheap holdout test on the incrementality page works with either.
Honest footnote
Who should ignore this page
Under $20k a month on one or two channels: fix tracking, run surveys, read the journey view. You will learn more in a week than a model would tell you in a quarter.
Questions people actually ask
Straight answers
What is MMM in one sentence?
A statistical model that looks at weekly spend and sales across channels over a long period and estimates how much each channel contributed, without tracking any individual.
What is incrementality testing in one sentence?
An experiment: turn a channel off (or up) for some people or regions, keep it steady for others, and measure the difference in sales. The difference is the incremental effect.
Why do MMM vendors want $100k a month in spend?
Because the model needs variation in spend and sales across many weeks to tell channels apart. Small budgets do not move total sales enough to be detected above normal noise.
Does TrueROAS do MMM?
Yes, as a separate product (True Horizon) for larger budgets, calibrated against the same order-verified data. Most stores should not start there, which is why it is not on every plan.
See your own numbers, verified against your orders.
Free under $10k/month in revenue. 14-day trial above that. Nothing changes in your ad accounts. You will know within a week which platform has been telling stories.
"I built this because I was making $50k-a-month decisions on numbers I knew were wrong. If TrueROAS does not show you something Meta or Google got wrong in the first two weeks, tell me and I will help you cancel."
Rasmus Arvidsson
Founder, TrueROAS. Answers support himself.
Keep going