From one retail media problem to a proceed, pause or stop decision

A retail media strategy framework should connect one observed problem to a choice that a named owner can reverse. For example, an England retailer may have an approved on-site placement but no shared definition of an eligible impression. The immediate decision is whether to run a controlled evidence test, not whether to declare a new media business successful.

Frame the job in one line

Use this blank statement:

For [retailer property] in [England operating context], [owner] must decide whether to [retain, test, change or stop] [advertising exchange] for [defined customer or advertiser task] by [review date].

IAB UK describes retail media as retailer or marketplace controlled advertising space, data assets and in-store opportunities. Use that industry definition to exclude ordinary merchandising, trade terms and unrelated digital advertising. Then name the environment: on-site, off-site or in-store.

Record what happened, how it was observed and what remains unknown. A brief rejected by an advertiser is evidence of that decision, not proof of England-wide demand. A dashboard count is a supplier-defined event until its population, exclusions and reconciliation are checked.

Build the evidence chain

Create six linked records:

1. The problem record gives the affected task, period, source and owner.
2. The baseline explains the current route, including a no-change option.
3. The option record describes the smallest material difference between routes.
4. The assumption register says what must be true and what would disprove it.
5. The gate register assigns separate advertising, consumer, privacy, accessibility, security, rights and finance decisions.
6. The test record defines fixtures, expected results, evidence and rollback.

Do not let a commercial score override a failed gate. The CAP Code rule on recognition of marketing communications requires marketing communications to be obviously identifiable. Whether the actual placement meets that standard needs a qualified advertising review of its content and context.

Define the decision measure

Write the primary measure as a specification, not a target. Include event, eligible population, numerator, denominator, period, Europe/London timezone, source query, deduplication, missing events, corrections and owner. Leave the threshold blank until the buyer has evidence for the decision.

Keep operational delivery apart from customer behaviour and commercial effect. HM Treasury's current Magenta Book evaluation guidance distinguishes observing outcomes from attributing them to an intervention. It is public-sector methodology, not an advertising rule, but its counterfactual boundary helps prevent a before-and-after change being presented as causal proof.

Test only what the decision needs

Use synthetic product, audience and event records unless specific authorised data is necessary. Include approved and withdrawn creative, clear and unclear ad labelling, tracking refusal, duplicate and late events, an inaccessible path, revoked access, export and rollback. The NCSC's guidance for secure online services treats access, logging, monitoring, incidents and supplier responsibility as connected operating questions.

For each case, capture expected result, observed result, evidence ID, tester and verdict. No live customer targeting, purchase or payment is needed to test the framework.

Make the final choice explicit

The owner chooses proceed, revise, pause or stop. Proceed means the bounded test passed and every specialist gate has recorded approval. Revise requires a stated defect, correction owner and retest. Pause means necessary evidence is absent. Stop applies when the proposition depends on unsupported claims, impermissible data, an inaccessible journey or no workable recovery route.

This draft gives no compliance finding. Before publication, named UK specialists must review the actual exchange, and the source and test records must be re-opened on the decision date.