# Choose an attribution rule without turning credit into causation

Constructed teaching cases with full context. No actual people, accounts, payments, research participants or forecasts.

Define the outcome, eligible touchpoints, identity limits, lookback window and credit rule before reporting acquisition. Keep first-user, session and key-event scopes distinct. A rule can assign credit to an observed payment without proving what would have happened without the article; preserve consent gaps, refunds and customer deduplication separately.

## Write the outcome and credit contract

Decide whether the outcome is a valid enquiry, a verified paid order or another clearly defined action. Specify what a touchpoint means, which source tags count, the lookback window and the handling of repeat purchases. Do not start with pageviews and rename them customers.

Use only permitted existing links between observations. A campaign parameter labels a URL, but does not establish identity, consent, a joined path or payment. When a source is unavailable, keep attribution unavailable rather than assigning it to the channel you hope performed well.

## Keep first-user, session and key-event scopes separate

Google Analytics documents user-scoped First user dimensions, Session dimensions and event-scoped key-event attribution. These can answer different acquisition questions from the same recorded path. A returning session’s source does not change the historical first-user source into a new person.

The selected event attribution model affects applicable event-scoped reporting, not every user/session dimension. Reporting identity and available observations also matter. Keep the actual dimension, metric and model with a screenshot or export; matching a channel name does not make differently scoped rows interchangeable.

## Distinguish current tool models from illustrative rules

Google’s current attribution overview lists data-driven, paid-and-organic last click and Google-paid-channels last click. It says first-click, linear, time-decay and position-based models stopped being available in November 2023. Do not give instructions for a removed dropdown option because an older tutorial still lists it.

You can still describe an explicit first-observed-touch rule in a separate teaching ledger; that does not make it a current Google model. Google describes data-driven modeling and its experimental/counterfactual inputs, but this guide does not audit that algorithm or translate its vendor description into a validated causal estimate for your own blog.

## Resolve three constructed attribution statements

A1 — Constructed path: a recorded blog visit, later an email visit, then a direct return and one verified order. An illustrative first-observed-source rule credits the blog; an explicitly declared last-non-direct rule credits email. Decision: keep the one order and name the rule. Outcome: two credit descriptions, not two customers or proof that either source caused the order.

A2 — Constructed report mismatch: one table uses First user source with event count, another uses session source with sessions, and a third attributes a key event. Decision: preserve scope and metric instead of subtracting the totals as lost buyers. Outcome: different valid questions may produce different values without identifying a reporting defect.

A3 — Constructed refund and repeat: two payments belong to one verified customer and one order is later refunded. Decision: retain order/payment identities, customer deduplication and the chosen gross/net treatment. Outcome: not two distinct new customers and not unreconciled payment clicks counted as revenue.

## Keep consent, missing paths and causal questions visible

Observed paths are limited by consent, browser/device identity, retention, tagging and available backend evidence. A direct label is not proof that the person never encountered the blog. A missing tag is not authority to reconstruct private browsing or overwrite the payment source.

For a rule-based report, write “credited under this model,” not “would not have happened without this article.” Causal lift requires an appropriate design and assumptions beyond that credit rule. Keep unsupported competitor conversion, revenue and SEO claims out of the conclusion.

## Save a versioned attribution statement

The worksheet retains outcome source, permitted identity path, dimension scope, eligible touchpoints, rule/window/version, QA exclusion, duplicates/refunds and unresolved attribution. Preserve old model versions when comparing reports instead of silently rewriting history under a new rule.

Primary scope/model guidance was read on 1 November 2026. All paths and orders are constructed teaching examples, not evidence of actual purchases, measured lift or reader preference.

## Primary sources

- [Google Analytics: traffic-source dimension scopes](https://support.google.com/analytics/answer/11080067?hl=en)
- [Google Analytics: current and deprecated attribution models](https://support.google.com/analytics/answer/10596866?hl=en)