Strategy
Part of Marketing attribution strategy: matching claims to evidence
A blank attribution planning template with buyer-set stop rules
Use a blank attribution planning template for an England decision, with evidence definitions, method boundaries, owners and buyer-set stop rules.
Complete this template for one allocation or measurement decision. Blank fields are deliberate. The buyer must approve thresholds, time limits and acceptable uncertainty after reviewing its own risks. Nothing here certifies a model, campaign or data practice.
What to take away
- Write the decision before choosing the metric, and let the buyer approve thresholds, time limits and uncertainty.
- Attach source-system definitions because page events, accepted leads, orders and settled payments are different outcomes.
- Record system, field, owner, extraction time, query version, transformation, matching rule, missingness and retention for every input.
- Leave stop values blank until approved, and name who can pause collection or restore the previous configuration.
- Keep a revision log row for each material change and state whether earlier results remain comparable.
Decision record
Field / Buyer entry
- Decision ID and version
- [ ]
- Decision to be made
- [ ]
- Authorised owner
- [ ]
- Options, including no change
- [ ]
- Evidence cut-off and review date
- [ ]
- Action if evidence is inconclusive
- [ ]
Write the decision before the metric. The Government Analysis Function's evaluation-planning guidance sets out questions, outcomes, measures, data, evaluation design and use for government work. This commercial template borrows that ordering but does not claim public-sector approval.
Decision record fields
- Decision ID and version
- Decision to be made
- Authorised owner
- Options, including no change
- Evidence cut-off and review date
- Action if evidence is inconclusive
Outcome and population
Definition / Buyer entry
- Outcome event and business meaning
- [ ]
- Eligible population and denominator
- [ ]
- England geography rule
- [ ]
- Reporting period and Europe/London convention
- [ ]
- Included and excluded channels
- [ ]
- Refund, cancellation or duplicate treatment
- [ ]
- Data latency and correction window
- [ ]
Attach the source-system definition. A page event, accepted lead, order, settled payment and net contribution are different outcomes. Do not change the label while leaving the calculation unchanged.
Outcome and population definitions
- Outcome event and business meaning
- Eligible population and denominator
- England geography rule
- Reporting period and Europe/London convention
- Included and excluded channels
- Refund, cancellation or duplicate treatment
- Data latency and correction window
Add a short rationale beside every exclusion. A channel may sit outside scope because its cost record cannot be reconciled for the chosen period, not because it seems unimportant. Name the person responsible for closing that gap and the point at which the plan must be reconsidered. This prevents a temporary data problem from becoming a permanent analytical blind spot.
Evidence and method
For every input, record system, table or report, field, owner, extraction time, query/version, transformation, matching rule, missingness and retention. Then select the evidence job:
Select the evidence job
What evidence job fits the decision?
Platform credit allocation, reconciliation, MMM, experiment or incrementality
Do not proceed until the evidence job is named
- platform credit allocation within a named product and configuration;
- reconciliation of records across systems;
- aggregate marketing mix modelling with stated variables and assumptions;
- controlled or quasi-experimental comparison; or
- causal incrementality estimate supported by a credible counterfactual.
HM Treasury's Magenta Book is government evaluation guidance. Its emphasis on planning evaluation early and selecting proportionate methods is useful here, while its institutional scope must remain explicit.
Assumptions and uncertainty
Record expected confounders, selection effects, spillovers, seasonality and missing channels. Record changes in offer or price, model dependence, transfer limits. For each, name the test, sensitivity or disclosure used.
The Analysis Function's quality and uncertainty guidance says potential bias and distortive effects should be reported alongside official statistics. Apply that disclosure habit without treating marketing output as official statistics.
Independent approval gates
| Gate | Reviewer | Evidence | Decision and expiry |
|---|---|---|---|
| Privacy and PECR | [ ] | [ ] | [ ] |
| Security and access | [ ] | [ ] | [ ] |
| Accessibility | [ ] | [ ] | [ ] |
| Advertising claims | [ ] | [ ] | [ ] |
| Finance and tax | [ ] | [ ] | [ ] |
| Contract and exit | [ ] | [ ] | [ ] |
If cookies, pixels or comparable device technologies are involved, map the exact purpose and configuration against the ICO's current storage and access guidance. Do not use a consent flag as a substitute for the review.
Decision and stop conditions
Leave the following values blank until approved: minimum source completeness [ ]; maximum reconciliation difference [ ]; acceptable uncertainty [ ]; maximum decision exposure [ ]; and test stopping boundary [ ]. Name who can pause collection, reverse a change and restore the previous configuration.
Finish with the decision, dissent, unresolved evidence, monitoring owner and next review trigger. Archive the signed version with data and model references so a later reader can reproduce what was known at the time.
Revision log
Keep one row for each material change: version [ ]; date [ ]; field changed [ ]; reason [ ]; evidence owner [ ]; approver [ ]; effect on comparability [ ]. A changed outcome, population, attribution window or model normally creates a new analysis version. The log should state whether earlier results remain comparable rather than assuming continuity.
Before you act
- Write the decision before the metric.
- Attach the source-system definition.
- Add a rationale beside every exclusion.
- Select the evidence job and record inputs.
- Name who can pause collection.
- Log each material change with comparability.
Common questions
What should be recorded for every input?
Record system, table or report, field, owner, extraction time, query or version, transformation, matching rule, missingness and retention. Then select the evidence job, such as platform credit allocation, reconciliation, marketing mix modelling, a controlled comparison or a causal incrementality estimate with a credible counterfactual.
Which approval gates does the template list?
The template lists privacy and PECR, security and access, accessibility, advertising claims, finance and tax, and contract and exit. Each gate needs a reviewer, evidence, decision and expiry. If cookies, pixels or comparable device technologies are involved, map the exact purpose and configuration against the ICO's current storage and access guidance.
What goes in the revision log?
Keep one row for each material change: version, date, field changed, reason, evidence owner, approver and effect on comparability. A changed outcome, population, attribution window or model normally creates a new analysis version. The log should state whether earlier results remain comparable rather than assuming continuity.
