Marketing attribution vs incrementality: making better budget decisions
An attributed conversion and an incremental conversion answer different questions. This explainer shows how that distinction affects campaign reporting, experimentation and decisions about the next unit of marketing spend.

Know which question each metric answers
A channel report can help a performance team understand recorded conversions within that platform’s measurement rules. It is not automatically a complete estimate of the sales that would have been lost without the spend.
Consider a buyer exposed to several channels before purchasing. Different reporting systems can associate that purchase with their own interactions. Before presenting a combined return figure, agree how conversions, revenue, time windows and attribution rules are reconciled.
Use modelling as evidence with assumptions
Google’s Meridian is an open-source marketing mix modelling framework designed to estimate channel contribution, response curves and budget scenarios. Its documentation explicitly connects causal estimates to modelling assumptions and describes uncertainty in the outputs. See Google’s introduction to Meridian.
A model output is not a permission slip to move budget without review. Ask what data was available, which assumptions matter and how the recommendation changes under plausible alternatives. Experiments, where feasible and properly designed, can provide another source of evidence.
Create one allocation record
- Record the outcome being optimised and the budget period.
- Show the source data, key assumptions and uncertainty ranges.
- Document channel floors, ceilings and operational constraints.
- Identify who approved the allocation and which version is active.
- Track actual spend and results against the approved version.
Separate recommendations from spend authority
Our suggested operating model gives analysts space to test scenarios while keeping live-spend authority explicit. An assistant can explain a change or draft a plan without being authorised to execute it.
The execution layer should work from the approved allocation, with spending caps, monitoring and a way to stop automated changes. A recommendation that looks attractive in a model may be impractical because creative assets, channel capacity or contractual commitments constrain delivery.
Bring finance into the review early
An illustrative monthly review can begin with three views: recorded platform performance, modelled incremental contribution and actual spend against plan. Differences between them should be explained rather than blended into a single unqualified figure.
Agree the commercial outcome with finance before selecting a model. Revenue, contribution margin and customer acquisition are different objectives. A plan that increases attributed revenue may not be the plan that improves the business result being targeted.
The useful output is a traceable decision: what changed, why it changed, what evidence supports it and when the team will review the result.
Common questions
Is attribution useless?
No. Attribution can support channel operations and analysis. The limitation is treating assigned credit as proof of the causal effect of the spend.
Does marketing mix modelling guarantee a better budget?
No. Its usefulness depends on data quality, assumptions, validation and the constraints under which recommendations are applied. Review uncertainty and implementation limits.
Sources & further reading
Put the insight into practice
Explore the related Catalyst47 services and accelerators.