Attribution Is Credit. Incrementality Is Cause.

Attribution decides which marketing interaction receives credit. Incrementality estimates how much of the result would not have happened without the marketing. Those are different questions, and neither should be asked to answer the other.

The distinction matters whenever a report moves from “this channel was present before the conversion” to “this channel caused the conversion.” A customer can see an ad, search the company name later, return directly, and then buy. Every recorded interaction may be real. The decision about credit still depends on a model, while the decision about cause requires a comparison with what would have happened otherwise.

This is not an argument against attribution. It is a way to use attribution for the decisions it can support and to recognize when a stronger form of evidence is required.

Begin with the two questions

Attribution: Which recorded interaction should receive credit for a conversion?
Incrementality: How many additional outcomes occurred because the marketing ran?

Attribution helps explain the observed path. Incrementality asks about an unobserved alternative: what would the same market have done without that advertising, or with a meaningfully different level of it?

If the business needs to decide which keyword, ad, or landing page deserves attention, attribution and campaign records may be enough. If it needs to decide whether the advertising created additional demand, reallocated existing demand, or claimed credit for customers who would have arrived anyway, attribution alone is not enough.

Attribution is a rule for distributing credit

A last-click model gives credit to the final eligible interaction. Other models distribute credit across several interactions or estimate how particular touchpoints contributed to the recorded path. Each model can produce a different answer from the same customer journey because each applies a different rule.

That does not make the report meaningless. Attribution can reveal which campaigns appear in converting paths, how performance changes under a consistent model, and where campaign management should investigate next. It becomes misleading only when assigned credit is presented as independent proof of causal lift.

Incrementality estimates the result that would not otherwise exist

The business cannot observe the same customer both exposed and unexposed at the same moment. Causal measurement therefore creates a comparison. A well-designed experiment might hold advertising back from a comparable control group, expose a treatment group, and estimate the difference in outcomes between them.

Google describes Conversion Lift in these terms: randomized treatment and control groups are used to estimate incremental conversions attributable to an advertising campaign. Google also distinguishes incrementality, attribution, and marketing-mix modeling as separate measurement approaches rather than interchangeable reports. See Google’s explanations of Conversion Lift and modern measurement methods.

An incrementality result is still an estimate within a defined population, period, campaign, and test design. It does not automatically transfer to another market, season, offer, or budget. Some businesses also lack the conversion volume, geographic structure, platform eligibility, or clean outcome data needed for a useful controlled test.

Do not run a causal test on a broken evidence path

Use the lightest evidence that can answer the decision, in this order:

Define the outcome. Decide whether the business is evaluating a form submission, qualified opportunity, booked job, completed project, collected revenue, or profit.
Validate the event. Confirm that the conversion fires at the intended moment and does not inflate repeat actions, test submissions, spam, or unrelated calls.
Reconcile the records. Connect campaign events to unique customers and downstream outcomes before comparing platform credit with business value.
Use attribution for path questions. Keep the model and comparison period visible when deciding which observed interactions deserve attention.
Use causal measurement for lift questions. When the business decision justifies it and the data can support it, design a holdout or other credible counterfactual comparison.
Apply business economics. Incremental conversions are not automatically incremental profit. Include fulfillment, margin, cancellation, and collection where the decision requires them.

A more advanced measurement method cannot repair an undefined conversion or a customer record that loses the outcome. Measurement maturity begins with a dependable chain of evidence, not with the most complex model available.

Match the method to the decision

Improve ads, keywords, audiences, or landing pages: Use validated campaign events, attribution, customer quality, and controlled campaign changes.
Understand the customer path: Compare attribution models and customer records without treating either model as causal proof.
Decide whether the campaign created additional outcomes: Use an appropriately powered incrementality experiment when the platform, market, and volume allow it.
Evaluate the total marketing portfolio: Consider experiments, business records, and marketing-mix analysis together; no single dashboard answers every level.
Decide whether the result is economically worthwhile: Connect incremental outcomes to margin, capacity, fulfillment, and collected revenue.

Before requesting another report, write down the action the business would take if the answer changed. If no decision would change, the next measurement may be interesting without being useful.

What this does not prove

A high attributed return does not prove that every credited sale was caused by the campaign.
A low attributed return does not prove the campaign had no influence outside the selected model or window.
A positive lift estimate does not explain which creative, message, or customer experience produced the effect.
An inconclusive experiment does not prove zero effect; the test may lack sufficient signal or an appropriate design.
A disagreement between attribution and business records does not, by itself, prove dishonesty or poor campaign management.
Incrementality does not repair weak lead handling, poor fit, limited capacity, pricing problems, or disconnected internal records.

When conversion volume is low, the first useful move may be simpler: validate the conversion definition, inspect individual customer paths, and make smaller decisions with explicit uncertainty. False precision is not an improvement over an honest evidence limit.

Ask what kind of claim the evidence can support

A useful paid marketing review labels each statement correctly: observed activity, assigned credit, reconciled business outcome, experimental estimate, or interpretation. That vocabulary makes it easier to see which conclusions are dependable and which still require another test.

If the immediate problem is campaign management, explore paid marketing. If customer identity, status, or revenue disappears between internal tools, the evidence problem may belong in business systems consulting. The services are separate; the shared requirement is a question that can be traced to a business result.