Measurement When Attribution Breaks

Attribution was already imperfect before AI search, marketplaces and fragmented journeys made it worse. The answer is not to give up on measurement. It is to measure in layers instead of asking one model to explain everything.

Why this matters now

Last click and blended ROAS are too narrow for a world where demand is created, researched, compared and converted across several surfaces. They overreward channels close to the transaction and understate the work that shaped the customer's decision.

Retailers need to understand product visibility, feed eligibility, signal quality, contribution, customer value, channel role and ownership. That is a broader measurement job than a platform dashboard can handle.

What is actually changing

The measurement unit is shifting from channel performance to commercial system performance. Does the product appear where it should? Is the feed eligible? Are events clean? Are conversion values useful? Are customers worth acquiring? Does the channel have the right role?

POAS and contribution matter because revenue can be misleading. LTV cohorts matter because first purchase efficiency can hide poor customer quality. Incrementality matters because retargeting and branded demand can flatter themselves.

What is often misunderstood

The misunderstanding is that better attribution software will fix the problem. It may help, but it cannot correct weak product data, poor event quality, missing margin data or unclear channel roles.

Another misunderstanding is that measurement must be perfect before decisions can improve. Often the first step is simply separating revenue from contribution and new customer quality.

What retailers should review

• Which product groups are eligible and visible in key platforms?
• Are conversion events clean, deduplicated and value-rich?
• Can revenue be connected to margin and returns?
• Do acquisition cohorts differ by channel and first product?
• Are upper-funnel indicators reviewed without pretending they are last-click sales?

What good looks like

Good measurement has layers. Platform reporting shows activity. Feed and eligibility reporting shows whether products can appear. Contribution reporting shows whether sales are worth having. Cohort reporting shows customer quality. Operating reporting shows who needs to act.

No one number is asked to carry the whole business.

What not to overdo

Do not build a complex attribution model while basic conversion actions are wrong. Do not import margin values before finance and marketing agree definitions. Do not let measurement become an excuse for delayed action.

Fix signal quality first.

Practical next step

Build a Commerce Foundations Scorecard for one category. Include feed eligibility, PDP readiness, event quality, ROAS, POAS, stock depth, margin, returns, new customer quality and the owner of each next action.