Audience research
Part of Planning retail media audience research with six populations kept apart
Retail Media Target Audience Data: A Source and Frame Audit
Audit retail media target audience data with named source records, frame counts, UK data-permission checks and a review date to retire stale specifications.
A retail media target audience specification should let a reviewer trace each inclusion rule back to the data supporting it. This audit guide covers retail media target audience data: how to name source records, inspect relevant fields and permissions, reconcile frame counts and set a review date.
Examples include retailer loyalty records, campaign logs, advertiser CRM systems, survey panels and clean-room outputs. These sources do not all establish the same thing: a purchase record is not an ad-exposure record, and an available contact is not proof of permission to use it. The aim is a reproducible source-to-frame record, not a count presented as the size of a market.
Keep a versioned specification
Give each version a stable identifier, owner, status, effective date, last review date and expiry date. Store the exact audience rule with that version. If the rule changes, record the change and issue a new version rather than silently widening or narrowing the old one.
Record the operation, retail-media setting and criteria that define the audience. For each criterion, identify the field or evidence that supports it; label unsupported criteria as unsupported rather than filling the gap with a convenient proxy. Keep the specification distinct from the number of records currently available to apply it: source coverage may be incomplete even when the rule is clear.
The wider strategic rationale can sit in the retail-media strategy framework; this record is for the auditable source-to-frame specification.
Name sources and inspect their fields
List each source separately, with its owner or custodian, extraction date, version, fields used, known omissions, selection filters and intended use. Include the source's role in the specification and any joins or transformations applied.
Named businesses to investigate for relevant records, as in an evidence ledger for competitors, might include Amazon Ads, Tesco Media, Nectar360, Boots Media and Asda Media. Their names alone do not establish that a particular record, field or permission is available: confirm the source, scope and terms for the specific use under review.
| Source record | Fields to check | What those fields can support, and what they cannot establish alone |
|---|---|---|
| Retailer loyalty or transaction record | Transaction date, product or category identifier, account or loyalty identifier | A recorded purchase in scope; not, by itself, exposure to an advert or a causal effect. |
| Campaign or platform log | Campaign and placement identifiers, event type, event timestamp | A recorded delivery, impression or click, depending on the event definition; not necessarily a purchase. |
| Advertiser CRM record | Account identifier, role field, decision or activity date | A recorded account or role; a job title alone may not establish that the person handled a particular decision. |
| Panel record | Recruitment or response date, stated eligibility fields, source and method | What the panel record documents; self-report and panel coverage have limitations that should be recorded. |
| Clean-room output | Input sources, match or aggregation method, output fields and access controls | What the documented output represents; do not assume identifiers being hidden makes an output anonymous. |
For every field used, note whether it is directly recorded, derived, self-reported or inferred. Record missing values, conflicting values and any proxy. For example, a transaction date and category may support a purchase criterion, while a campaign event and placement identifier may support an exposure criterion. Do not treat either as evidence for the other.
Check permissions and data-protection records
Name the responsible organisations and establish their roles for the particular processing: who decides its purposes and means, who processes data on another's instructions, and who shares or receives it.
Record the relevant source terms, privacy information, data-sharing terms, permitted purpose, retention arrangements and any restrictions on onward use. A supplier's ability to provide a file does not itself demonstrate that the proposed use is permitted.
For UK audience data, identify the applicable UK GDPR and Data Protection Act 2018 requirements. Record the lawful basis relied on for personal-data processing and the evidence supporting it; consent and legitimate interests are distinct bases, not interchangeable labels. The UK regulations evidence register holds the wider legal record.
Check PECR where electronic marketing or access to information on a user's device is involved. Assess whether a data protection impact assessment is required, including where the proposed processing is likely to result in high risk.
The Information Commissioner's Office (ICO) is the UK data protection authority. The ICO register of fee payers is not a register of audience lists or evidence that a particular list may be used for a particular campaign. Record the actual source terms and permission evidence relevant to the proposed use, and leave the permission status unresolved if the evidence does not answer it.
Reconcile the frame from source to retained records
Keep counts at each processing boundary and state what is being counted. Record rows separately from deduplicated people or organisations; they are not interchangeable. Use the same count basis when comparing versions, and describe any change in the deduplication or exclusion rules.
| Processing stage | Audit record |
|---|---|
| Source extract | Source and version, extraction date, fields received and rows received. |
| Cleaning | Duplicates or unusable records removed, the rules applied and the remaining count. |
| Rule application | Records passing, failing or not assessable against each criterion, with the reason. |
| Retained frame | Final count on the stated basis, unresolved gaps and any known coverage limits. |
If the counts do not reconcile, preserve the discrepancy as unresolved and investigate it. Do not conceal it by combining unlike categories or presenting the final count as the target audience's total size. A changed count may reflect a new extraction, altered coverage, missing fields or a changed rule; identify the explanation only where the audit record supports it.
Replay the rule with synthetic records
Before relying on a version, test whether another reviewer can apply its criteria consistently. Use artificial records, not real people's data. Include straightforward cases and edge cases, such as a missing date, conflicting role fields or an event that falls outside the stated period.
Recruitable frame checklist
- Name source list and owner
- Record extraction date and fields
- Note omissions and contact permission
- Count frame, eligible, invitations, responses
- Count completions and analysed records separately
- Log exclusions and non-response risk
- Test eligibility rule on synthetic records
Record the specification version, fields supplied, expected outcome, reviewer outcome and any disagreement. If the test exposes ambiguity, clarify the rule, issue a new version and rerun the affected cases. Retain the earlier version and test outcome so a reviewer can see what changed. This tests whether the specification is repeatable; it does not validate the quality or completeness of the underlying source.
Set review and expiry controls
Name an owner and put an expiry date on each version. There is no universal review interval established here: choose and record a date appropriate to the decision and the stability of the sources.
Bring the review forward if a source changes, a field disappears, the setting or buyer authority changes, the workflow shifts, or the purpose changes.
At review, record whether the source, terms, fields and criteria still support the specification. Mark an expired version as no longer current; reuse it only after the owner has completed and recorded that review. A release note should identify the version, source versions and extraction dates, processing changes, unresolved gaps, count basis, review owner and expiry date. State plainly what the evidence supports and what remains unknown.
For the separate task of planning audience research across distinct populations, see the retail-media audience research guide.
