Operations
Part of The honest guide to marketing attribution
Turning marketing attribution numbers into a decision you can defend
How to measure marketing attribution starts with an outcome contract, path scope, identity rule, model, reconciliation, sensitivity, economics, and decision log.
What to take away
- Reconcile the eligible outcome total before distributing credit.
- Run model changes in parallel before they affect bids or budgets.
- Use a separate causal study when the decision depends on incremental effect.
How to measure marketing attribution is a governance problem as much as a calculation problem. The workflow below keeps raw outcomes, path construction, credit allocation, and business action traceable.
Use a nine-step workflow
| Step | Deliverable |
|---|---|
| 1 | Decision and claim boundary |
| 2 | Conversion and value contract |
| 3 | Touchpoint and channel taxonomy |
| 4 | Identity, consent, and match record |
| 5 | Container, lookback, and time basis |
| 6 | Reconciled eligible-outcome table |
| 7 | Versioned model output |
| 8 | Sensitivity and causal-test plan |
| 9 | Decision and mature-outcome review |
Google Ads documents its current attribution model options and explains that a conversion action's model affects reporting and some bidding uses. That is product behavior, not evidence that the assigned credit equals incremental business value. Incremental value is the question marketing mix modeling attacks from the other side, using spend and outcome totals rather than paths.
Reconcile offline outcomes carefully
Google's guide to offline conversion import discrepancies identifies date, time-zone, processing, upload, and attribution reasons why imported outcomes and external reports can differ. The guidance applies to the named workflow and does not validate an external system's records.
Offline Outcome Reconciliation Checklist
- Count eligible source outcomes by conversion time
- Count uploaded records and rejected rows
- Compare matched and unmatched outcomes
- Separate upload time from conversion and interaction time
- Reconcile count before value
- Document late records and restatement schedule
Publish a model card that names the version, training or rule basis, eligible channels, exclusions, container, lookback, refresh, change behavior, and known failure modes. Then test whether a plausible alternative model changes the intended decision. If it does, the attribution result is too model-dependent to support that action alone.
Close each reporting cycle with a three-way check: source outcomes, attributed allocations, and the business decision log. Timing differences may be legitimate, but every material difference needs an owner, an explanation, and a review date.
The reconciliation habit anchors campaign measurement, where totals are agreed before any channel is compared. Unexplained gaps stay quarantined until a second qualified reviewer accepts the resolution in writing.
Run a controlled handoff
Assign each step to a named role, and require an observable finish condition. The analyst must reproduce the result, and the decision owner must explain the action and stop rule.
Include one broken-data case and one revoked-access case, then record the repair, its time, and any vendor help so the written process reflects normal operation, not a prepared demonstration.
The GOV.UK technology selection guidance recommends choices that can change over time, preserve data control, address security risk, and include ownership cost. Those public-service rules become useful buying questions, but they are not private-sector mandates or product endorsements.
CISA's software acquisition fact sheet points buyers toward development practice, supply-chain exposure, deployment, and vulnerability management. Add those questions to any review of marketing analytics tools that will carry attribution data. The government context does not replace a buyer's own technical testing, contract review, or risk decision.
Keep the evidence record beside the decision so a reviewer can rebuild the reasoning without relying on memory. Set the next review date and name the change that would trigger an earlier check. Record the rejected options as well as the chosen path, because the constraint that ruled them out may not hold next quarter.
Common questions
Should attribution be measured daily?
Operational checks may run daily, but decision timing should respect processing delay and outcome maturity.
What if model totals do not match the source?
First align eligibility, time basis, filters, exclusions, and processing. Quarantine the result if an unexplained material gap remains.
Can attributed ROAS guide planning?
It can provide a consistent operating view when definitions are stable. Budget impact claims still need incremental economics.
