Retail media measurement for 2027 planning, with every metric given a full definition record

Retail media measurement needs a defined event and population before a result can support a decision. For an England retailer or marketplace, a request to select advertising, a rendered placement and a sale credited to a campaign answer different questions. Reporting should preserve those differences instead of presenting every stage as evidence of effectiveness.

This 2027 planning guide uses public sources inspected on 6 September 2026. It contains no observed 2027 results, campaign dataset or performance benchmark. Its reporting records are proposed house methods for retailer-controlled advertising inventory and associated data operations, with a source refresh and qualified review required before publication.

Begin with the question the report must answer

Write the decision beside the report specification. An operator investigating delivery needs different evidence from an advertiser asking how sales were credited. Someone estimating causal effects needs an evaluation design as well as reliable event records.

HM Treasury's Magenta Book analytical annex, updated on 15 May 2026, describes experimental and quasi-experimental impact methods. It provides UK government methodological context, not a mandatory measurement procedure for private retailers or evidence of a campaign's success.

Use that distinction when commissioning analysis. A dashboard can organise observations without identifying what would have happened in the absence of advertising. Do not ask a report designer to conceal an unanswered causal question behind a more persuasive chart title.

Keep the service boundary visible. Identify the retailer property, placement type, advertiser and catalogue scope. General website traffic, unrelated channels and total retail sales should enter the report only where their defined role in the analysis is explained.

Give every metric a complete definition record

The following proposed record applies to every metric discussed in this guide, including simple counts. A count has no denominator unless it is used to form a ratio; record that as not applicable rather than leaving the field ambiguous.

Definition field	Information to preserve
Event and unit	What occurrence is counted or estimated
Denominator	Exact eligible base for a ratio; otherwise not applicable
Cohort	Included property, format, campaign and population
Period and time zone	Boundaries for events and reporting
Attribution window	Eligible lookback and credit rule, or not applicable
Exclusions	Filtering, missing records and coverage restrictions
Corrections	Treatment of late data, returns, duplicates and revisions
Source	Dataset, extraction time and specification version
Uncertainty	Missingness, sampling or modelling limits and unresolved definitions

Attach an owner to the record so questions reach someone able to explain it. These fields are a proposed editorial control, not a claim that every supplier dashboard exposes them. If a required field cannot be established, label the affected measure undefined or unavailable.

Do not assume a default time zone or attribution window. Record the actual choice before comparing exports. A reporting period and a lookback window describe different boundaries, so neither should be used as shorthand for the other.

Preserve the definition with each published version. If an analyst changes the eligible population, the new result is not directly comparable with the earlier series until that change has been examined and disclosed.

Keep technical delivery stages distinct

An ad request is a call for an advertising decision. A served or returned response records what the system sent back under its documented response definition. Neither event, on its own, establishes what appeared on the shopper's device.

A rendered placement concerns content beginning to appear in the page or app under the adopted measurement specification. A viewable impression satisfies the applicable viewability criteria. Attention concerns a shopper's engagement with the advertising and needs its own disclosed observational or research method.

The IAB/MRC January 2024 retail-media guidelines provide an international measurement reference for delivery and related metrics. Treat it as industry methodology; confirm which specification the actual implementation follows rather than assuming every field named impression means the same thing.

For the proposed delivery record, counts of requests, returned responses and rendered events each retain their own event source and cohort. Their denominator and attribution window are not applicable when shown as counts. Any transition ratio requires an explicitly matched eligible base, with duplicate requests and unmatched records addressed.

Keep viewability measurement coverage separate from the viewability result. Where a placement was not measurable, do not silently count it as either visible or invisible. Attention remains UNKNOWN unless suitable evidence exists; a successful rendering observation cannot fill that gap.

Define interaction and sales measures before calculating them

For a click count, specify the qualifying interaction and filtering rule. If constructing a click-through ratio, define the eligible impression denominator and ensure its cohort and period correspond to the numerator. An undefined or empty base does not support a numerical rate.

For sales, distinguish a recorded transaction from a transaction receiving campaign credit. State whether the unit is an order, item or monetary value. Record how cancellations, returns and late corrections affect the measure before comparing results across retailers.

The IAB Europe May 2026 commerce-media standards address gross and net sales, attribution and incrementality. They are European industry guidance rather than an England performance dataset. Their definitions do not supply a measured result for the publisher.

An attributed-sale count has no denominator when reported alone, but it still needs its attribution window, eligible interaction types and credit allocation rule. An attributed conversion ratio must specify whether its base concerns people, interactions or another defined unit. Do not switch between those units within a comparison.

Where monetary measures are later introduced, retain the source currency and use GBP for an England comparison only with a documented conversion basis if needed. Specify whether tax, returns and fees enter the chosen definition. This guide inserts no monetary amount or assumed conversion factor.

Identify modelled estimates where they enter the report

A modelled estimate may represent unobserved activity, a projection or an inferred effect. Record which of those jobs it performs. Keep its input data and assumptions separate from the observed event stream so a reader can identify where inference begins.

For each estimate, use the same definition record as for observed metrics, adding model version and validation limits. Specify the target cohort, output unit and period. Record a denominator where the output is a rate and explain any uncertainty interval rather than inventing one.

Do not combine measured and extrapolated sales without exposing the components. If a report cannot separate them, disclose that limitation near the total. A modelled value does not become an observation because it is displayed beside observed counts.

Causal incrementality requires an explicit outcome and a defensible comparison with the absence of the intervention. State the evaluation population, assignment or comparison method, outcome period and analysis unit. A credited-sale total is not a substitute for that design.

Make permission and coverage part of interpretation

The ICO's online advertising guidance explains consent requirements for advertising storage and access technologies, including associated measurement. It does not approve any particular retailer data flow.

Ask a qualified privacy reviewer to assess the actual collection and processing arrangement. For reporting purposes, document what population the available evidence covers. Do not assert that missing observations represent the same behaviour as observed ones without a supported analytical basis.

Keep unavailable data marked as unavailable. A decision not to collect a particular event is not evidence that the event never occurred. Neither should the need for a more complete dashboard be treated as permission to collect additional shopper information.

Design a report that exposes its limits

Use separate areas for delivery observations, attributed outcomes and analytical estimates. These are proposed presentation choices, not required dashboard categories. Let the reader inspect the applicable definition record from each displayed measure.

Place the report period, time zone and extraction version where they remain visible on an export. Avoid showing a comparison arrow if the previous period used a different scope or definition. Explain what changed before suggesting that the underlying performance improved.

Give unavailable measures a reason. UNKNOWN benchmark, undefined denominator and unmeasured attention are different states, each requiring different work. Replacing them with nil values would create an unsupported numerical claim.

Control corrections and claims made from the report

Keep the original export and the corrected version linked by a change record. Identify the affected metric, reason for revision and whether earlier external statements need review. This is a proposed reporting control; no supplier correction schedule is assumed.

When a result becomes an objective claim in marketing, CAP's substantiation rules require appropriate documentary evidence. That advertising-code requirement should not be misdescribed as a universal rule governing every internal dashboard.

Check the wording of any outward claim against the evidence category. A statement about delivery should not become a statement about shopper attention, and an attributed outcome should not be promoted as incremental revenue without the relevant support.

Decide what evidence is still needed

Select the next real report and complete its metric records before populating charts. Request missing source definitions and document any comparison that remains invalid. A benchmark for another geography, format or attribution convention should remain outside the comparison unless compatibility is established.

This planning edition provides no benchmark verdict or performance assurance. Publication requires real authorship, independent checking, qualified privacy and measurement review, confirmed conflicts and refreshed sources. Operational decisions need the actual dataset and its limitations, not an invented example presented as a result.