Attribution allocates credit according to a rule or model. Incrementality asks what would have happened without the activity. These are different questions.
A customer may click a branded search advertisement immediately before purchasing, even though they already intended to buy. The click can receive credit without proving that the advertisement created the sale.
Define the economic outcome
Choose an outcome linked to the decision: incremental qualified opportunities, sales, contribution or another defensible business result.
Return on advertising spend usually compares attributed revenue with advertising spend. It does not automatically account for product margin, returns, service cost or sales that would have occurred anyway.
Use an agreed economic definition and time horizon. A campaign can look profitable on booked revenue while consuming cash and producing poor contribution after delivery.
Use attribution for the right purpose
Attribution can help describe journeys, identify reporting changes and support tactical investigation. Its output depends on observed interactions, identity matching and the chosen rules.
Avoid presenting one model’s allocation as a causal fact. Several platforms may claim credit for the same purchase, and unobserved interactions may matter.
A hypothetical company totals channel-reported conversions and finds more credited revenue than the business actually generated. Reconciliation should come before optimization.
Use experiments where practical
Randomized holdouts can estimate the difference between exposed and comparable unexposed groups. Define the population, outcome and evaluation period in advance.
Geographic experiments can be useful where individual randomization is impractical, but require careful design around comparability, spillovers and statistical power.
A test estimates an effect under its conditions. Do not assume that doubling spend will double the effect. Audience saturation, capacity and competitive response can change the result.
Use models with explicit assumptions
Marketing mix modeling relates aggregated outcomes to marketing activity and other influences. Google’s Meridian documentation frames this as a causal-inference problem with assumptions; a sophisticated model does not remove the need to examine confounding and model design.
Where available, experiments can inform or calibrate model estimates. Meridian’s GeoX documentation describes this connection between geo experiments and mix modeling.
The management question is whether the data, variation and expertise support the decision. A smaller business may learn more from a well-designed focused test than an unstable large model.
Triangulate before reallocating
Compare commercial outcomes, experimental evidence, model estimates and customer information. Investigate disagreement instead of selecting the most favorable number.
Show uncertainty ranges and practical limitations. A result can be directionally useful without supporting a precise budget recommendation.
Allocate the next unit of spend based on expected marginal value, constraints and risk. Historical average returns can conceal diminishing returns.
For the next budget review, select one major channel and state what is known about its incremental effect, what is merely attributed and what remains uncertain. Design the next learning activity around the most consequential uncertainty.
Marketing accountability improves when leaders ask a clearer question and accept an honest answer about what the evidence can establish.