Key takeaways
- Journey evidence is stronger than bounce rate alone.
- Cart, purchase, refund, and fulfilment are different outcomes.
- Attribution and fraud classification must remain separate.
- Risk scores are not revenue probabilities.
Map the journey without collapsing it
Link the paid interaction to the measured session, visitor boundary, product views, search, cart changes, checkout attempts, purchase occurrence, payment result, cancellation, refund, and fulfilment where legitimately available.
Each event has different reliability and business meaning. A fast visit may be low intent, a returning buyer, a tracking failure, or automation; context determines its usefulness.
Evaluate acquisition and outcome signals separately
Compare campaign, query, device, location, network, velocity, behaviour, identity continuity, product mix, and later outcomes. Maintain canonical purchase occurrences so duplicated browser, server, and provider events do not inflate conversion counts.
- Reconcile event sources instead of deleting duplicates.
- Preserve provider and internal attribution side by side.
- Exclude unavailable outcomes from denominators.
- Report currency and refund timing explicitly.
Use evidence-safe impact language
Report measured suspicious sessions, affected campaign spend, and verified actions as distinct metrics. Do not multiply a risk score by revenue or claim savings before the calculation, eligibility, provider credit, and uncertainty are defined.
Limitations
What this guide does not claim
Ecommerce journeys vary by platform, consent state, payment provider, attribution window, and refund cycle. Not all downstream evidence will be available or appropriate to collect.
Evidence
Primary sources
- About invalid trafficGoogle Ads Help
- Clicks and sessions discrepancy troubleshootingGoogle Ads Help
- Consent mode overviewGoogle for Developers
Read how we source, review, update, and correct content in our editorial standards.
