Choose the question before the model
A model is a rule for allocating conversion credit. It does not define company revenue or prove that an interaction caused a purchase. A finance reconciliation, an acquisition review and a creative decision therefore need different evidence. Write down the decision, owner and commercial measure before selecting a reporting view.
For example, a team reviewing how customers discover a new collection may examine early interactions. A team reviewing the final steps before purchase may inspect later interactions. Neither view, on its own, settles which channel deserves more budget.
Understand what each rule emphasises
First-click attribution gives credit to the earliest eligible interaction; last click gives it to the final eligible interaction under the report’s rules. Linear attribution divides credit equally. Time decay gives more weight to recent interactions. These are useful concepts to compare, but not every platform offers every model.
Data-driven attribution uses a model based on the data available to the platform. It is not a complete record of all customer influences. Check the supported model, eligibility rules, available history and reporting scope in your actual account. Our interactive example illustrates arithmetic allocation; it does not simulate a platform’s data-driven model.
Write a reporting agreement
Agree the purchase event, revenue definition, lookback window, eligible interaction types, timezone and reporting period. Record how returns, consent gaps and cross-device journeys affect interpretation. Compare reports using the same definitions wherever possible, and label differences that cannot be reconciled.
Choose one consistent view for recurring comparisons. Keep a secondary view for investigating differences rather than switching models until performance looks favourable. Annotate any configuration change so the team can separate a reporting change from a business change.
Use experiments for the budget question
When the question is whether activity generates additional demand, attribution alone is insufficient. Consider an appropriate controlled experiment with a defined primary metric, assignment method and evaluation period. Some businesses or channels will not support a reliable test at the desired scale; make that limitation visible.
A useful outcome is a short decision record: the model used, what it can explain, what remains uncertain and which evidence would change the decision. CeeCee can help bring that context into the discussion, while the responsible team reviews the evidence and owns the budget decision.
Source: Google Ads: Conversion Lift ↗
A collection launch review
A shopper first encounters a paid social launch, later returns through organic search and finally purchases after an email. First click, last click and linear attribution allocate the same order differently. The team keeps its agreed reporting view, inspects the full observed journey and proposes a separate experiment if the question is whether the launch created additional purchases. The different allocations are not three separate orders.
Turn the guide into a useful review
- Name the decision and the metric it affects.
- Confirm the models and interactions supported by your tool.
- Document windows, revenue and identity limitations.
- Separate credit allocation from evidence of incremental impact.
Reference material
Platform guidance checked 3 October 2026. Examples and working checklists are Faccelerate editorial illustrations.
Bring the question.
Keep the context.
Map the relevant sources and specialist workflow for your team.
Explore your stack
