- Does attributed revenue prove that advertising caused additional sales?
- How can a credible counterfactual and explicit evidence limits improve the budget decision?
Start with the decision, not the attribution chart
A Saudi business reviewing acquisition budgets may see branded search receive most conversion credit while video appears weak. That is a reporting observation, not proof that moving every available riyal into search will create more customers. The immediate decision should be narrower: which budget change is justified by the evidence, and what uncertainty must be resolved first?
Marketing attribution distributes credit across observed interactions. Incrementality asks how many outcomes happened because of an intervention, compared with what would have happened without it. These questions are related but not interchangeable. Google explains how Analytics attribution assigns conversion credit; those assignments should not be interpreted as a causal experiment.
Keep three evidence layers separate
Use an operational scorecard with three distinct sections:
- Observed outcomes: qualified opportunities, completed orders, revenue, and the date each outcome became available.
- Attributed contribution: the channels receiving credit under a named model, reporting window, and conversion definition.
- Incremental evidence: results from a deliberately designed comparison, including uncertainty and reasons the comparison may be biased.
A channel can capture existing demand and still be commercially useful. Another can introduce prospects who later convert through branded search. Neither explanation is established simply because a dashboard shows a particular path. Unobserved interactions, identity gaps, consent choices, and incomplete sales records constrain interpretation.
Before debating models, align the business outcome with the CRM outcome workflow. A submitted form and a sales-accepted opportunity answer different budget questions.
Design a test around a credible counterfactual
Where feasible, randomly assign eligible units to a treatment and a holdout, specifying the outcome and analysis before launch. Define whether the unit is a person, account, or geographic area, and avoid mixing units when interpreting results. Geographic tests need enough comparable areas; a single city against another can confuse advertising effects with local differences.
Check whether audiences can cross between groups, sales teams apply different follow-up, or another campaign changes during the test. Promotions, availability, seasonality, and procurement timing can all distort the comparison. A simple before-and-after increase does not isolate advertising from these influences.
Choose a duration that accommodates conversion delay and a sample capable of supporting the decision. If outcomes are sparse, uncertainty may remain too wide for a confident budget recommendation. Report that limitation rather than turning an inconclusive result into a claim of no effect.
Complete an unfilled causal-question brief
Record the decision, primary business outcome, eligible population, proposed comparison, observation window and analyst responsible for the design. Leave the expected difference and observed result unfilled. An attribution report can describe assigned credit, but it cannot supply the counterfactual required to estimate an additional effect.
Before commissioning an incrementality study, document how exposure could differ between comparison groups and which operating conditions must remain comparable. Check overlapping accounts, sales handling and outcome maturity with the relevant owners. Seek appropriate analytical review rather than treating a convenient before-and-after comparison as a controlled experiment.
Specify how uncertainty and implementation limits will be reported before results arrive. Keep daily attributed reporting separate from the causal conclusion, and document which budget decisions each can support. If the available design cannot answer the question credibly, narrow the decision or gather better evidence. Do not fill the gap with an invented holdout result or a precise unsupported revenue claim.
Approve a decision with explicit limits
Before changing budgets, confirm the outcome definition, assignment method, contamination checks, sales lag, uncertainty, and maximum acceptable downside. Document what would trigger expansion, repetition, or cancellation.
Common mistakes include treating model agreement as causal proof, judging long-cycle sales too early, and presenting a point estimate without its uncertainty. Another is assuming a successful test applies unchanged to every audience, spend level, or season.
For a decision-ready measurement plan, contact Kitely to connect attribution reporting, CRM outcomes, and a realistic incrementality test.
