MMM vs Attribution vs Incrementality: Which Actually Tells You It Worked?
Three methods, three different questions. Attribution divides credit, marketing mix modelling estimates contribution, and incrementality experiments isolate cause. Knowing which one you can honestly run — given your data and your spend — matters more than picking the “best” one in the abstract.
By William Walczak — CEO, Hiilite Creative Group Inc.The question hiding inside “did it work?”
Most marketing reports answer a question nobody actually asked. A dashboard says a channel “drove” 40 conversions — but drove compared to what? Would those 40 have happened anyway? The phrase “did it work?” is really shorthand for “how many of these results would not have happened if I’d turned this off?” That’s a causal question, and only some methods are built to answer it.
Attribution, marketing mix modelling (MMM), and incrementality experiments all get pointed at the same spend, but they sit on a spectrum from descriptive (what the data shows) to causal (what your money actually changed). Picking the wrong one isn’t a rounding error — it can flip a budget decision in the wrong direction. The good news: each method has a clear job, and you can match the job to where your business honestly sits today.
Attribution: it divides credit, it doesn’t prove cause
Attribution — last-click, first-click, or multi-touch (MTA) — takes the conversions you got and splits the credit across the touchpoints along the path. It’s the most accessible method because it’s baked into ad platforms and analytics tools, and it’s genuinely useful for understanding the journey: which channels people encounter, in what order, before they buy.
The catch is the assumption underneath it. Attribution looks at the path a converter took and assigns credit as if every touch caused the outcome. But a customer who was going to buy anyway might still click your branded search ad on the way in — and last-click hands that sale to search, even though search changed nothing. Attribution describes correlation along a path; it quietly treats that path as causation. Treat it as a map of behaviour, not a verdict on what your spend produced.
- Good for: understanding customer journeys, spotting which channels assist, day-to-day pacing
- Weak for: proving a channel caused incremental sales
- Watch out for: branded search and retargeting inflating their own credit
MMM: top-down contribution, good for the big budget picture
Marketing mix modelling takes a step back. Instead of tracking individual paths, MMM uses your historical data — spend by channel, sales, plus factors like seasonality, promotions, and price — and builds a statistical model that estimates how much each channel contributed to the total. Because it’s aggregate and privacy-friendly (no cookies, no individual tracking), MMM has come back into fashion as third-party tracking erodes.
MMM’s strength is breadth: it can include channels you can’t click-track at all, like out-of-home or TV, and it’s well suited to high-level budget allocation — roughly how to split spend across channels. Its limit is honesty about cause. MMM finds correlations in historical patterns; if two channels always move together, or you’ve never varied spend much, the model can’t cleanly separate their effects. It estimates contribution; it doesn’t run a controlled test. It’s a strong planning instrument, not a causal proof.
Incrementality experiments: the causal gold standard (but narrower)
Incrementality experiments answer the causal question directly by creating a comparison that doesn’t otherwise exist: a group that didn’t get the marketing. In a holdout test you withhold ads from a random slice of your audience; in a geo test you turn a channel off in some regions and leave it on in others. Because the groups are comparable and the split is (ideally) random, the difference in outcomes is the lift your spend actually caused — the counterfactual made real.
This is the closest you can get to proof, which is why it’s the top rung of our Measurement Ladder. The trade-off is scope and cost: a single experiment answers one question about one channel or campaign over one window, it requires enough volume to detect a difference, and you give up some conversions in the holdout. You can’t experiment on everything at once — so you run experiments on the decisions that matter most, and use cheaper methods to triage the rest. For the deeper mechanics, see what incrementality really means.
Match the method to your data maturity
The honest answer to “which method?” isn’t a winner — it’s a sequence. They stack. Attribution gives you a cheap, always-on read on behaviour. MMM gives you a top-down allocation view that survives privacy changes. Incrementality settles the high-stakes arguments with an actual test. The skill is knowing which one you can credibly run right now, given your data, your volume, and the size of the decision.
That’s exactly what the Measurement Ladder is for: it sequences these methods by rigour and by what your business is ready for, so you don’t over-engineer a $500 test or under-prove a $50,000 one. Start where your data lets you stand, and climb as the stakes rise.
- Just starting / low volume: lean on attribution for direction, stay sceptical of the credit
- Steady spend, multiple channels: add MMM for budget allocation across the mix
- A big, expensive, recurring decision: run an incrementality experiment to settle it
- Always: let the size of the decision set the rigour you pay for
A practical way to decide
Before you reach for a method, ask three questions. First, how big is the decision — is this a five-figure recurring commitment or a small test? Second, what data can you honestly stand on — do you have the volume and variation for a real experiment, or just platform-reported clicks? Third, what’s the cost of being wrong? The bigger the decision and the higher the cost of error, the further up the ladder it’s worth climbing.
This is the discipline behind how we measure growth: don’t default to the method that’s easiest, and don’t demand a randomised experiment for every decision. Match the method to the question and the money. We sell outcomes, not effort — and an outcome you can’t distinguish from “would have happened anyway” isn’t one.
Keep reading
Questions, answered
Is MMM better than attribution?
They answer different questions, so “better” depends on the decision. Attribution maps the customer journey and credits touchpoints along the path — useful for day-to-day pacing but it assumes the path caused the sale. MMM takes a top-down view of each channel’s contribution from historical spend and sales, which is stronger for budget allocation and survives privacy changes. Neither proves cause cleanly; for that you need an incrementality experiment.
If incrementality is the gold standard, why not just use it for everything?
Because it’s narrow and costly. Each experiment answers one question about one channel over one window, needs enough volume to detect a difference, and sacrifices some conversions to the holdout group. That makes it the right tool for big, recurring, high-stakes decisions — not for measuring every campaign. Use attribution and MMM to triage, and spend your experiment budget where being wrong is expensive.
Which method should a small business start with?
Start with what your data can honestly support. If volume is low, use attribution for directional reads while staying sceptical of the credit it assigns. As spend steadies across several channels, add MMM for allocation. Save incrementality experiments for the decisions large enough to justify the cost. The Measurement Ladder sequences this so you climb only as the stakes rise.
More from this series
Seven plain-English pieces on measuring marketing and turning tests into durable capability:
Want to know whether your marketing is the reason?
One conversation to map what to measure and how we’d prove it.