UGC marketplaces key metrics: asset evidence measures

Draw the boundary around one asset

Use F-UGC-01 as an asset-level decision unit, rather than a marketplace business KPI. The worked example begins with a commissioned asset and leaves every measure blank until its definition, source and eligibility tests are set. No result overrides a failed legal, rights, safety gate, privacy, accessibility or security.

Apply the same frame whether commissioning through TikTok Creator Marketplace, a UGC agency or a direct creator brief. The marketplace can supply creators and content, while buyer-owned measures test evidence after delivery. Five checks suffice for this fictional asset if each has a definition, source, eligibility rule and invalidation trigger; UK GDPR guidance stays within the privacy gate.

Prove the release file is reviewable

Release evidence completeness answers whether a candidate can proceed to publishing review. For the worked example, count candidates submitted in the 30 days to 30 June 2026 in Europe/London, then divide those with linked PASS decisions for every required gate by all eligible candidates.

The numerator requires dated PASS decisions for brief authority, claims, recognition, rights, safeguarding, accessibility, privacy and PECR, security, and finance. Exclude withdrawn drafts and synthetic checks, while keeping FAIL and UNRESOLVED results visible rather than removing them from the record.

Use the Airtable Commission register as source. The worked query filters candidate_id, decision_date and gate_status to release candidate F-UGC-01 and dates 1 June to 30 June 2026. The publishing controller owns the measure; unlinked documents and late decisions remain visible as uncertainty, not silent passes.

CAP treats advertising recognition and substantiation as distinct gates; see advertising rules for the full advert check. The worked threshold is 95% completeness; below it, pause release review and resolve unlinked decisions. Record the date window and owner beside the result so late evidence remains visible.

Match the live item to approval

Published-version match answers whether the live item is the approved item. Count authorised releases whose identifiers and rendered elements agree with approval, divided by all eligible releases in the reporting window. Set treatment for reposts, edits, paid placements, duplicates, deletions and inaccessible checks before running the comparison.

Join the Meta Business Suite publishing log to the Google Drive render archive and approval log. Query release_id, approved_hash and live_hash, returning rows where mismatch is true. Retain the cut-off, source version and missing captures. The buyer account owner sets the worked threshold at 100% match and pauses release for correction before further promotion.

Admit only valid platform observations

Eligible platform observation rate answers whether an export supports descriptive reporting. Its population is expected release-period rows under a dated platform definition, and its numerator is admitted observations carrying release identity, timestamp and definition. The denominator is eligible records, not people; tests and unauthorised reposts stay out.

Use the TikTok Business Center analytics export, and query release_id, metric_label and observed_at for impressions dated 1 June through 30 June 2026. Keep duplicates, delayed rows and unavailable records visible. ONS quality guidance treats fitness for purpose, comparability and timeliness as quality questions; this private measure is not official statistics.

Keep platform labels intact: an impression, view or interaction is not a person and should not be presented as a comparable cross-platform unit. The dated platform definition controls eligibility; exclude tests and unauthorised reposts, and reject records missing release identity, timestamp or definition.

Reconcile buyer outcomes under one rule

Reconciled attributable outcome rate answers what share of buyer-recorded activity receives credit under one declared rule. Use last non-direct click, version 2.1, for the worked example; the population is paid outcomes for product F-UGC-01 in England, recorded during the 30 days to 30 June 2026.

Divide outcomes credited to F-UGC-01 by all eligible paid outcomes. Use the buyer order export and query order_id and attributed_asset_id for records created between 1 June and 30 June 2026. State the attribution window, unmatched records, cancellations, returns, duplicated platform credit and missing data; attribution methods develops the rule-selection question.

Call the output attributed, not causal. The 2026 Magenta Book distinguishes monitoring and impact evaluation in government, so a qualified evaluator must adapt that boundary to this setting. Attribution alone does not establish that the asset produced the buyer outcome for the specific decision.

Time defects to evidenced containment

Correction containment time answers whether an owner can limit a verified or suspected publication defect. Start at the first qualifying alert and stop only when containment is evidenced, not when someone acknowledges it. Define support hours, timezone, pauses, duplicate alerts and unresolved cases before timing begins.

Use the Jira issue tracker with the Meta Business Suite publishing log; the worked query filters UGC publication_defect issues created from 1 June 2026 in ascending order. The correction lead owns the measure. Example performance is containment within 4 working hours for 90% of qualifying alerts; a verified defect open beyond 8 working hours pauses publishing.

CMA brand guidance supports monitoring and correction of improper labelling but states no universal response time. Treat that green marketing context as support for monitoring, not as a source for the buyer-selected threshold. A threshold cannot replace a failed legal or rights gate.

Record each invalidation trigger before accepting a result:

Do not preserve a number simply by adding a footnote. Quarantine the affected period, preserve the prior version, investigate the evidence chain, and ask the named decision owner whether the measure should be restated or withdrawn. The issue is whether the result remains valid, not whether its value looks favourable.

Separate marketplace operations from asset proof

Marketplace operating figures answer different questions from asset evidence, so keep them in a separate ledger with a written definition, source and owner before showing a value. They describe commissioning workflow or rights administration, but do not replace any of the five asset-level decision measures above.

The commissioning measures have no published England benchmark, so use earlier commissions from the same buyer rather than a market norm, as benchmark research explains; keep these operating measures separate from the five asset evidence decisions and label local thresholds as worked examples.