Quick definition: Attribution assigns credit for a conversion, revenue event, or other outcome to one or more sources or interactions according to a defined measurement rule.
What is attribution?
Attribution organizes a customer journey into a reportable story. A purchase might be associated with an organic search, display impression, email click, product notification, referral, or direct visit. Because several touchpoints can precede the same outcome, an attribution rule decides which touchpoint receives credit and how much.
This is useful for operational questions: Which channels appear most often before signups? Which campaign should a team investigate? How do reported revenue totals change after a landing-page launch? Attribution makes these descriptions consistent. It is not, by itself, evidence that a credited touchpoint caused the outcome. People who see an ad or open an email may already be more likely to purchase, and a journey record may miss other influences entirely.
That distinction matters when an organization reallocates budget or claims impact. Descriptive credit answers, “How did our chosen convention allocate observed outcomes?” Causal incrementality answers, “How many additional outcomes occurred because of this intervention compared with what would otherwise have happened?” The questions complement one another but require different evidence.
Definitions and boundaries
A touchpoint is an observed interaction or exposure, such as an ad impression, campaign click, message delivery, site visit, or in-product prompt. A conversion is the defined outcome of interest, such as a purchase, qualified lead, activation, or subscription. A conversion window is the period after a touchpoint during which the outcome can receive credit.
Attribution maps conversions to touchpoints under a rule. Contribution is the resulting assigned share. Incrementality is the causal change in outcomes produced by an intervention. A platform can report high attributed revenue for an audience that would have converted anyway; conversely, an intervention can create incremental demand that a last-touch report credits to another channel.
| Question | Appropriate evidence | What it does not establish alone |
|---|---|---|
| Which recorded channel received credit? | Consistent attribution rule and event data | Causal impact |
| How many journeys included email? | Identity and event coverage assessment | Whether email changed outcomes |
| Did campaign spend create extra conversions? | Randomized holdout or credible causal design | Exact credit across every touchpoint |
Attribution is also distinct from accounting reconciliation. Media platforms, analytics tools, and finance systems may use different identities, time zones, revenue recognition rules, and windows. Their totals may legitimately differ. Define the intended use before demanding that every dashboard match.
Concrete scenario: paid search, email, and direct conversion
A retailer records the following journey: on Monday, a shopper clicks a paid-search ad; on Wednesday, they open a promotional email; on Friday, they enter the site directly and buy a $100 item. Last-click attribution may award $100 to direct. First-click awards $100 to paid search. A linear rule allocates roughly $33.33 to each of the three recorded interactions. All are internally valid descriptions if their definitions and windows are stated.
None tells the retailer whether paid search or email created the sale. The shopper may have been intending to buy before the ad, or the email may have reminded them at a decisive moment. Direct traffic may also be a label for visits where the referrer was unavailable, not an absence of marketing. A randomized holdout that withholds the email from a comparable eligible group can estimate the email’s incremental effect on total purchases, including purchases later labeled “direct.”
The scenario also exposes a measurement boundary. If the person read a review on another device, received an offline recommendation, or declined tracking, the recorded path is incomplete. Attribution should describe what was observed and avoid implying complete knowledge of the decision journey.
A decision framework for attribution
- Start with the decision. Use descriptive attribution for reporting, investigation, and workflow; use causal evidence when deciding whether an intervention should be expanded or funded.
- Define the outcome. Specify the conversion event, revenue treatment, cancellation handling, currency, time zone, and whether one person can create multiple conversions.
- Define touchpoints and eligibility. State whether impressions, clicks, deliveries, visits, and in-product events qualify. Avoid silently mixing them.
- Set and disclose windows. Choose click-through and view-through windows that are defensible for the product’s purchase cycle. Apply them consistently.
- Choose a credit rule. Select a rule that matches the descriptive question, not the channel that a team hopes will look best.
- Measure coverage and duplication. Assess cross-device identity, consented measurement, offline conversion imports, bot filtering, and whether the same conversion arrives from several systems.
- Triangulate material claims. Pair reports with holdouts, lift tests, or other causal designs before making large budget decisions.
For a path with touchpoints t1 through tn and conversion value V, an attribution scheme assigns weights wi such that wi ≥ 0 and Σwi = 1. The assigned credit is V × wi. This formula explains the key limitation: weights distribute an observed value; they do not identify the counterfactual value if a touchpoint had not occurred.
Attribution and experimentation
Randomized experiments create a comparison group that approximates the outcomes that would have occurred without the intervention. An email holdout, geo experiment, or audience-level suppression can estimate incremental conversions by comparing treated and control groups under a pre-specified design. The result is strongest when assignment, exposure, outcome measurement, and analysis rules are stable across groups.
Attribution remains helpful in an experiment. It can show whether the treatment changed observed journeys, identify tracking anomalies, and explain why an effect differs by segment. But do not use post-treatment attributed revenue as the sole proof of lift. Treatment can alter the path labels themselves: an ad may bring someone back to the site, causing a last-click system to assign it credit even when total conversions do not increase.
When reading an experiment, report both the causal outcome and the attribution convention. The site’s A/B testing guide provides context on randomized comparison, while primary and guardrail metrics explains why a single attractive reporting number should not decide a rollout.
Pitfalls and limitations
- Calling credit “impact.” Attribution weights are not a counterfactual estimate.
- Changing definitions mid-comparison. Different windows, identity logic, or conversion definitions can manufacture apparent performance changes.
- Double-counting across platforms. A sale imported into several systems can be credited more than once when reports are added together.
- Ignoring missing journeys. Consent choices, browser restrictions, blocked tags, offline interactions, and device changes affect who and what is observed.
- Using a model as an incentive mechanism. If channel owners are judged solely on a credit rule, they may optimize for touchpoints that capture credit rather than create value.
FAQ
Is attribution the same as marketing incrementality?
No. Attribution allocates observed outcomes under a rule; incrementality estimates causal additional outcomes using a comparison design.
Which attribution rule is best?
No rule is universally best. Choose one that fits the reporting question, document it, and use causal tests for high-stakes impact claims.
Should direct traffic receive credit?
It can under the selected convention. “Direct” often includes unknown or unavailable referrers, so interpret it cautiously.
How long should an attribution window be?
Use the product’s purchase cycle and evidence about plausible response timing. Report the window and test whether conclusions are sensitive to reasonable alternatives.
Can privacy restrictions affect attribution?
Yes. Consent, retention, identity limits, and minimized collection can reduce path coverage. State the observed population and measurement gaps.
Summary
Attribution is a transparent convention for distributing credit across recorded touchpoints. It is valuable for describing journeys and operating reports, but it does not prove that credited activity caused a conversion. Define outcomes, touchpoints, windows, identity rules, and coverage; use a stable model; and validate material budget decisions with randomized incrementality tests or another credible causal design.