Multi-touch attribution against Incrementality testing, dimension by dimension.
| Dimension | Multi-touch attribution | Incrementality testing |
|---|---|---|
| Question answered | Which touchpoints were present? | What would have happened anyway? |
| Type of claim | Correlational | Causal |
| Effect of consent loss and cross-device | Severe — missing touchpoints get redistributed to observable ones | Minimal — works on aggregates |
| Cost | Tooling and configuration time | Real revenue forgone during the holdout |
| Speed | Continuous, always on | Per test, typically weeks |
| Main failure mode | Reading the model as causal and reallocating budget on it | Running a test too small to detect the effect that would change the decision |
Which one your situation calls for.
Choose Multi-touch attribution when
- You need to route leads, score them, or diagnose where journeys break operationally.
- You want directional visibility across many campaigns continuously.
- The decisions at stake are tactical rather than budget-level.
The wrong reason: Trusting a data-driven attribution model because it sounds rigorous. It still only sees the touchpoints that survived consent, device switching and offline conversation.
Choose Incrementality testing when
- You are deciding whether to keep funding a channel at all.
- Brand search or retargeting is taking credit you suspect it has not earned.
- Platform-reported returns look implausibly good and nobody can explain why.
The wrong reason: Running one inconclusive test and treating it as proof the channel does nothing. An underpowered test produces a wide confidence interval, which is an absence of evidence rather than evidence of absence.
Follow-ups
What gets asked next.
Terms used above
Multi-touch attribution (MTA)
Multi-touch attribution distributes credit for a conversion across the marketing touchpoints that preceded it, using a rule (first touch, last touch, linear, time decay) or a data-driven model. It describes observed sequence; it does not establish that any touchpoint caused the outcome.
DefinitionIncrementality testing
Incrementality testing measures what a marketing activity actually caused, by comparing a group exposed to it against a comparable group that was not. Geographic holdouts, audience splits and time-based tests are the common designs, and the output is a causal estimate rather than an attributed share.
DefinitionMarketing mix modelling (MMM)
Marketing mix modelling uses aggregate historical data — spend, sales, seasonality, price, external factors — to estimate each channel’s contribution statistically. Because it works on aggregates rather than individual journeys, it is unaffected by cookie loss and consent restrictions.
DefinitionGA4 key events
Key events are the GA4 events a property has marked as significant — the renamed successor to "conversions" in that interface. They drive reporting and can be imported into advertising platforms, and their counts may include modelled as well as observed data.
Definition