UGC marketplaces measurement: keeping six evidence classes apart

UGC marketplaces measurement starts before publication. A marketplace and its buyer need to connect the approved brief, accepted asset version, rights term and release decision to later records without quietly changing the population or definition. A platform total on its own cannot show that the right asset ran, that an interaction came from an eligible exposure or that the asset caused a commercial outcome.

This guide continues the fictional England commission F-UGC-01: one 20-second vertical video for one English brand unit and one buyer-controlled social account. No creator, product claim, fee, audience, publication or result exists. All values and thresholds stay blank. The fixture is useful because it exposes the evidence contract before numbers create false confidence.

What to take away

Keep six evidence classes apart

Use exactly these six labels for evidence values:

Do not blend the labels in a dashboard total. The 2026 Magenta Book distinguishes monitoring, process evaluation, impact evaluation and value-for-money work in its government context. Its structure is transferable to a buyer's evidence register, but it does not validate a private campaign.

Lock the asset identity first

Give F-UGC-01 one commission identifier. Link it to the approved brief version, claim register, creator authority, rights schedule, accepted file identifier, disclosure decision, accessibility review, publisher approval and account. Keep raw material, rejected edits and the release candidate distinguishable.

The publication record should store the exact asset, channel, account, placement, time and timezone. A repost, paid placement, revised caption or edited cut is a different release event. Create a relationship rather than overwriting the original row.

Rights boundaries also affect eligibility. The IPO's commissioned copyright guidance explains why commissioning does not automatically transfer copyright. A metric cannot treat a reuse as eligible until the actual permissions cover that version, channel, territory and period.

Write the metric card before querying data

Every measure needs these fields:

ONS describes statistical quality as fitness for purpose, including relevance, accuracy, timeliness, clarity, coherence and comparability. F-UGC-01 is not an official statistic. Those dimensions still provide useful questions for a private register, provided the article does not claim ONS endorsement.

Measure delivery before attention

The first measures should show whether the commission operated as authorised. Examples include admitted brief completeness, accepted asset version, complete specialist gates, authorised publication, correction containment and exit evidence. These are not reach or effectiveness measures.

Define a release-integrity result as the number of eligible releases whose published version matches the accepted asset and approval record, divided by all eligible F-UGC-01 releases in the period. State how deleted, corrected, duplicated and test releases are treated. The value remains blank because nothing was published.

Advertising evidence remains a gate. CAP Section 3 addresses documentary substantiation for objective claims before distribution, while Section 2 covers recognition of marketing communications. A high interaction count cannot compensate for an unsupported claim or an unrecognisable advert.

Preserve platform definitions

If the channel reports impressions, views, interactions, clicks or attributed events, copy the exact label and definition from the dated technical record. Store source geography, account type, placement, reporting window, deduplication method and known exclusions. Do not rename a platform field to a generic term and combine it with another system.

Separate a release from an exposure and an exposure from a person. A repeated display may not identify a unique individual. An interaction can be automated, duplicated, late or outside the buyer's eligible population. Without an independently supported identity rule, leave person-level interpretation out.

The dashboard should show missingness and reconciliation. A source outage, delayed export or changed query version is part of the result, not a footnote added later.

Reconcile buyer outcomes separately

The buyer's internal record may contain an enquiry, order, cancellation, return, payment or retained contribution. Define each event in the buyer's system and period. Do not let a platform-reported event replace the commercial ledger.

Join records only through an authorised, documented method. State unmatched platform events, unmatched buyer events and duplicated credit. For personal information, the privacy owner records purpose, lawful authority, access and retention.

ICO guidance on data protection by design and default says organisations should consider privacy from design through decommissioning and limit use to what is necessary for the purpose. It does not select a basis for this fictional measurement plan.

PECR analysis, audience permissions and advertising substantiation stay distinct. A technically possible join may still be out of scope for the approved purpose.

Attribute credit without calling it impact

A deterministic rule can assign an outcome to the last eligible contact, first contact or another declared event. A platform may apply its own window or model. These methods answer where a rule placed credit. They do not show what would have happened without F-UGC-01.

Report the rule, eligible events, lookback period, source version and duplicate handling beside the attributed value. If several systems claim the same outcome, show overlap rather than adding all credited totals.

Association can support investigation. A change after publication may coincide with product availability, price, seasonality, paid media or another campaign. Record those competing explanations instead of presenting sequence as proof.

Reserve causal language for causal evidence

The government Quality in Policy Impact Evaluation guidance says monitoring can identify change but does not establish whether the intervention caused it. It explains the role of a counterfactual, selection bias and design quality for policy evaluation. A private buyer should adapt those concepts proportionately with a qualified evaluator.

Define the intervention, eligible units, outcome, comparison, allocation or identification strategy, analysis plan and exclusions before looking at results where feasible. Record contamination, missing data and concurrent activity. If a credible comparison is not possible, publish an observed or attributed result with its proper label. Do not upgrade it to incrementality.

One asset may provide too little information for a stable causal estimate. The right answer can be "not measured" rather than a precise-looking effect.

Make corrections part of the dataset

A corrected caption, withdrawn asset, refunded order or changed rights status can alter eligibility. Never silently edit the old value. Store the original, correction reason, approving owner, effective time and revised output version.

CMA guidance for brands using content creators discusses monitoring and correcting improperly labelled endorsements. The measurement register should therefore connect a defect to containment and the corrected publication. The guidance does not provide a response target or campaign benchmark.

The Office for Statistics Regulation's guide on transparent revisions and corrections applies to official statistics. Its emphasis on explaining changes is a helpful voluntary standard for buyer reports, not a legal status claimed for them.

Keep cost and benefit definitions visible

Link committed, incurred, invoiced, paid and forecast costs to the commission record. Keep creator consideration, supplier charges, production, rights, media, buyer labour, specialist review, correction and exit separate. Finance decides the currency, VAT and contribution treatment for the real entities.

Gross revenue, platform-attributed sales and retained contribution are different values. ROI requires reviewed incremental retained contribution and whole-life cost with matched periods. If causal contribution is unknown, ROI remains blank. The Green Book's 2026 appraisal framework is public-sector guidance; its separation of objectives, options, costs, benefits, risk and uncertainty is an analytical prompt, not a private return calculation.

Build the dashboard around decisions

Start with evidence-state and exception panels: missing asset links, unresolved gate decisions, source failures, late records, duplicate identity, rights expiry, correction queue and access-removal status. Put activity charts after those panels. A dashboard is an interface to the evidence register, not evidence by itself.

Every refresh stores the cut-off time, source snapshots, query version, definition version, correction policy and operator. Show unavailable fields as missing, never as zero. Give readers access to metric cards and a route to report an error.

Security protects the reporting chain. NCSC social media publishing guidance covers account logging, protected credentials and recovery. Measurement access should not create a back door to publishing. Separate read, export, approval and posting permissions, and test revocation.

Reject borrowed benchmarks that do not match

No direct authoritative England benchmark for a single UGC marketplace asset was identified in the official sources reviewed for this guide. A UK advertising statistic, supplier case study or global platform average has a different population and cannot fill the gap.

Admit an external comparator only if its publisher, geography, population, selection method, creator and asset definition, platform, placement, denominator, period, paid or organic status, missingness, privacy coverage, corrections and uncertainty are visible. Record differences and reject the comparison when they would change the decision. An external rate is not a target.

Current decision

F-UGC-01 remains HOLD. There is no approved asset, publication record, platform export, buyer outcome, cost ledger or causal design. The guide therefore reports definitions and evidence requirements only.

Before use, assign named statistics, evaluation, finance, advertising, consumer, IP, privacy and PECR, accessibility and security reviewers. Recheck source status and every channel definition on publication day. The framework can make uncertainty visible; it cannot guarantee data completeness, legal compliance or commercial effect.

Before you act

Common questions

How should a buyer treat a repost or edited cut?

A repost, paid placement, revised caption or edited cut is a different release event. Create a relationship rather than overwriting the original row, so the published version stays distinguishable from the accepted asset and approval record.

Can a deterministic attribution rule show that the asset caused an outcome?

No. A deterministic rule assigns an outcome to the last eligible contact, first contact or another declared event. It answers where a rule placed credit. It does not show what would have happened without F-UGC-01, so it is not impact.

What should happen when no credible comparison exists?

Publish an observed or attributed result with its proper label and do not upgrade it to incrementality. One asset may provide too little information for a stable causal estimate, so the right answer can be not measured.