Attribution and incrementality answer different questions

A channel receives credit for a sale. Does that mean the sale would not have happened without it? That question sits at the heart of attribution versus incrementality. The two approaches examine different things, and confusing them can make a campaign appear more conclusive than the evidence allows.

Attribution assigns credit across observed interactions according to a model. Incrementality asks about the additional outcome caused by an activity compared with what would otherwise have happened. A reporting model and a causal question are not interchangeable.

Use attribution to understand the recorded journey

Attribution can help organise how recorded touchpoints relate to conversions. Google Analytics, for example, provides a report for comparing how attribution models value channels. Differences between models illustrate how the allocation of credit depends on the approach used.

The view is also limited by what can be observed and connected. Missing interactions, measurement settings and the buying process affect the picture. Treat attribution as an interpretation of available data rather than a complete recording of every influence on the customer.

Ask the counterfactual question carefully

Incrementality is concerned with what would have happened without the activity. That comparison is difficult because the same person cannot simultaneously experience both conditions in the same way. A well-designed experiment can help estimate the difference for a defined situation.

The design matters. Selection effects, overlapping exposure and changing market conditions can undermine an apparent comparison. Specialist experimental support may be needed when the decision is consequential. A before-and-after chart alone is not a dependable substitute for a credible counterfactual.

Match the method to the decision

Attribution can be useful for routine journey analysis and identifying questions. Incrementality testing can help evaluate whether an investment creates additional demand under the tested conditions. Neither removes the need to consider margin, capacity and customer experience.

For an illustrative retargeting campaign, attributed sales may include people who were already likely to buy. That does not prove the campaign has no value. It establishes why a different form of evidence is needed to estimate its additional contribution.

Report the limits plainly

State what was measured, what comparison was made and how broadly the finding can reasonably be applied. AI can help summarise the analysis, but causal claims need human scrutiny. Words such as caused and drove should not appear merely because two measures moved together.

When reviewing a result, ask whether the team is discussing recorded credit or additional effect. Naming the question correctly prevents a useful metric from being asked to prove something it was not designed to establish.