If you spend real money on marketing, you have heard the lines. Attribution is broken. Incrementality is the gold standard. MMM is having a comeback. All roughly true, and none of it tells you which method to reach for when there is a decision in front of you. This guide compares the three on what they prove, where they break, and how they fit together.
In short
Attribution tells you what happened inside the channels it can see. Incrementality testing proves whether a specific spend caused extra sales. Marketing mix modelling shows where the next pound should go across everything you spend on, including channels with no click. Growing brands use all three, and usually start with MMM, because that is where the unspent growth tends to sit.
The short version
Before the detail, here is each method in one line:
- Attribution counts which tracked touchpoints came before a sale. It is best for fast, in-channel steering.
- Incrementality testing holds a comparable audience or region out of exposure and measures the gap. It is the cleanest read on cause and effect.
- Marketing mix modelling explains your revenue using two years of your own weekly data. It is the only one of the three that can put a value on channels with no clickable signal.
The comparison at a glance
| Criterion | Attribution | Marketing mix modelling | Incrementality testing |
|---|---|---|---|
| The question it answers | Which touchpoint preceded this conversion, and what to steer this week. | Where should the next pound go across every channel, and how big can each one get. | Did this specific spend cause extra sales that would not have happened anyway. |
| What it proves | Correlation. It counts co-occurrence, not cause. | Contribution. It estimates each channel's effect on results, above a baseline. | Causation. A controlled holdout isolates true lift. |
| Time horizon | Hours to days. In flight. | Quarters to years. Strategic. | The length of the test, usually a few weeks. |
| Granularity | Keyword, creative, audience. | Channel and driver level, including non-digital. | The specific tactic or channel under test. |
| Data needed | Pixels, tags, platform tracking. | Two years of weekly spend and revenue. No pixels. | A clean test and control group, plus the discipline to hold out. |
| Speed and cost | Built in, close to free, always on. | Weeks with modern tooling and a project cost. Traditionally months and six figures. | Cheap to run, but you forgo revenue in the holdout group. |
| Where it is strong | Fast tactical optimisation within a channel. | Whole-mix allocation, scaling, and valuing brand and offline. | Settling a high-stakes argument with a clean causal read. |
| Where it breaks | No click, no view. Misses offline, brand, and demand already coming. Around 60% of iOS users opt out of tracking. | Useless if spend never varies, heavy below roughly £1m of media, and not for creative or keyword calls. | Only answers the one thing you tested. Hard to run across everything at once. |
No single row makes the decision. The fit depends on the question you are trying to answer, which is the subject of the rest of this guide.
What each one is
Attribution
Attribution counts the tracked touchpoints that came before a conversion and assigns the credit to them, usually inside a short window. It is the cornerstone of thousands of healthy, growing businesses, and it is genuinely good at one thing: telling you what to do next inside a channel this week. Which keyword cluster, which audience, which creative variant.
Its limits are well known. Last-click credits a single ad when the average buyer saw many across several platforms. Around 60% of iOS users opt out of the tracking that multi-touch attribution depends on. It cannot see channels with no click, such as TV, out of home, sponsorship or word of mouth. It counts demand that was already coming, so only a minority of attributed conversions are incremental at all. And each platform marks its own homework. For the long version of this argument, see Beyond attribution.
Marketing mix modelling (MMM)
Marketing mix modelling, often shortened to MMM and sometimes spelled marketing mix modeling, is a statistical model that explains your business results using your own historical data. It takes around two years of weekly spend by channel, your weekly revenue, and other drivers such as price, promotions, distribution and seasonality, then works out how much each one contributed above a baseline. It does not use pixels, cookies or tracking, so the privacy changes that weakened attribution do not touch it.
The reallocation is the part worth paying for. MMM shows where you are under-invested, where a channel is saturating, and what the next pound returns across the whole mix. That is a growth lever: most brands can move money to higher-returning channels and scale into them with confidence. Open-source frameworks like Meta's Robyn and Google's Meridian, alongside bodies such as the Marketing Accountability Standards Board and Nielsen, have made rigorous MMM far more accessible than it was a decade ago. For a fuller primer, see What is Marketing Mix Modelling?
Incrementality testing
Incrementality testing is an experiment. You hold a comparable group out of exposure, by region, by audience or through ghost ads, then compare the exposed group with the held-out one. The gap is the lift that the advertising actually caused, rather than the sales that would have happened anyway. Of the three methods, it is the only one that proves causation directly.
Running the test is cheap. The real cost is the revenue you deliberately forgo in the holdout group, and the fact that a test only answers the one question it was built around. You cannot economically run a separate experiment for every channel and tactic at once. Incrementality is at its best as a calibrator: a clean way to settle the highest-stakes arguments and to sense-check what attribution and MMM are telling you.
Which should you use when
Match the method to the decision, not the other way around:
-
Which keyword, audience or creative should I change this week?Attribution. It is fast, in-channel, and already running.
-
Did that campaign or channel actually cause extra sales, not just coincide with them?Incrementality testing. A holdout is the only clean way to prove cause.
-
Where should the next pound go across everything I spend on, including channels with no click?Marketing mix modelling. Nothing else values the whole mix at once.
-
How much bigger can a channel get before it stops paying back?Marketing mix modelling. Saturation and the marginal return are core MMM outputs.
-
My tracked ROAS looks great but blended growth does not. What is real?MMM to see the whole picture, then an incrementality test on the channels in question to confirm.
-
All of the above, because we are growing fast.All three, in sequence. MMM sets the strategic allocation, incrementality validates the big calls, attribution steers the week.
When MMM is the wrong answer
It is worth being straight about this, because the wrong tool at the wrong time wastes money and trust:
- You are below roughly £1m of media spend. The modelling overhead is hard to justify until there is enough at stake in the allocation decision.
- Your spend never changes. If you put the same money into the same channels every week, the model has nothing to learn from.
- You need creative or keyword answers. MMM works at the channel and driver level, not the asset level. That is attribution's job.
- You are B2B with long, multi-stakeholder deals and low volume. Isolating marketing's effect is far harder when a handful of deals take months to close.
- Nobody will act on the findings. Measurement is only worth the decisions taken from it. Without the appetite to move budget, even a perfect model changes nothing.
None of this is a reason to avoid MMM forever. It is a reason to start it when the allocation question is big enough to move the business.
How the three work together
The strongest measurement setups do not pick a winner. Attribution handles the weekly, in-channel decisions. Incrementality testing validates the calls that carry real risk. Marketing mix modelling answers the strategic question of how much should sit in which channel in the first place, which is the question that unlocks the next stage of growth.
There is a useful loop between them. Cleaner attribution discipline feeds cleaner inputs into the MMM. A well-designed incrementality test can calibrate the model and raise confidence in its read. The methods sharpen each other, which is why mature brands end up running all three rather than arguing about which one is correct.
Common questions
Is incrementality testing the same as A/B testing?
They are related but not the same. An A/B test compares two variants of something, a subject line or a landing page, to see which performs better. An incrementality test asks a bigger question: would these sales have happened anyway, without this advertising at all. It answers that by holding a comparable audience or region out of exposure, then measuring the gap. A/B testing optimises a thing you are already doing. Incrementality testing checks whether the thing was worth doing.
Can attribution and MMM disagree, and which should I trust?
Yes, and they often do. Attribution tends to flatter the channels it can see, because it counts demand that was already coming and ignores channels with no click. MMM works from your own business results and credits incremental revenue across everything, including offline and brand effects. When they disagree, treat attribution as the tactical read inside a channel and MMM as the strategic read across the whole mix. If the gap is large, an incrementality test is the tie-breaker.
Do I really need all three?
Not on day one. Most growing brands start with MMM, because the biggest money is in the allocation question attribution cannot answer: where the next pound should go across every channel. Attribution is already running in most teams and is fine for weekly steering. Incrementality testing comes in to validate the highest-stakes calls. Running all three for the sake of it misses the point. Match the method to the decision in front of you.