Influencer agency measurement mistakes to check in reports

When shortlisting influencer agencies, use these measurement mistakes to check whether campaign reports distinguish populations, attribution and uncertainty.

Mistake: Calling attributed credit causal impact

Calling an attributed outcome causal impact is the most costly error: a code, link or attribution model allocates credit under a rule, but cannot show what would have happened without the work. UK government QPIE guidance says outcome monitoring alone cannot establish intervention causation.

Fix: Use causal wording only when a qualified reviewer accepts the counterfactual design. Otherwise label the result credited or attributed, and make no claim about the no-work outcome. Request design review before release; neither a code or link nor outcome monitoring replaces acceptance of that design.

The evidence check was dated 6 September 2026 and considered UK regulator, government and official-statistics guidance alongside Google and TikTok attribution records. It covered reporting or inference failures relevant to agency commissions, excluded supplier case studies, market averages and unsupported advice, involved no supplier funding or testing, and left the other mistakes unranked.

Mistake: Treating more tracking as stronger proof

More tracking can appear to strengthen evidence, but it does not establish that a result is reliable. The ICO final 2026 announcement on storage and access technologies covers pixels and similar tools under PECR and, where relevant, UK GDPR; the data flow still needs qualified review before claims are accepted.

Fix: Before presenting campaign return, reconcile full economic cost across creator consideration, agency fees, paid media, production, internal labour and remediation. Have the configured report and data protection controls reviewed by statistical, privacy and PECR reviewers, alongside advertising, consumer, finance and accessibility reviewers. Apply the release checklist before issue.

Place a release hold if the denominator cannot be rebuilt, a source definition is missing, credit is duplicated or causal wording exceeds the design. This material is not professional advice or a compliance guarantee. The reporting owner should apply the hold before issuing the configured report.

Mistake: Using one base for different creator stages

A denominator spanning eligible, contacted, responding, contracted, approved and published creators combines different funnel stages. It therefore describes no single stage, and a response or delivery rate built on it may mislead anyone making a campaign performance judgement.

Fix: Keep a separate count for each stage and retain evidence for every transition. The guidance on monetised social media posts stresses briefing and checking content, so reach cannot replace evidence of contracted control. Compare rates only when the stage and included population match.

Mistake: Recording an unavailable value as zero

An unavailable value is not a zero: unknown, suppressed, late and not applicable are different states. A missing export cannot establish zero engagement for the relevant population and period, because the absence of a file says nothing about the result it would have contained.

Fix: ONS discussion of survey uncertainty includes sampling and non-sampling sources, while campaign reports also need a separate missingness assessment and correction log. Preserve each state, record corrections, and hold release if the evidence available for the reporting period cannot rebuild its denominator.

Mistake: Combining modelled and observed conversions

Combining modelled records with observed totals presents inferred values as though they were directly recorded. Google's first-party explanation says conversion modelling predicts unobserved conversions in certain circumstances, so these categories do not carry the same evidential status within a single report total.

Fix: Show observed, estimated and modelled values separately in every platform total. A supplier field is not an independently verified sale. Keep each category visible regardless of platform so report readers can tell which records were observed directly and which are inferred or estimated before interpreting the figure.

Mistake: Adding platform credits as unique sales

Adding credits reported by separate platforms can count one commercial event more than once. The sum then reflects platform reporting rules rather than a deduplicated count of internal commercial events, so it cannot be presented as one distinct-event total for the whole campaign.

Fix: Google's conversion window controls which later events enter its reports, while TikTok attribution settings offer click-through and view-through choices. Where permitted, reconcile records against internal IDs, then show duplicate and unmatched records alongside the attributed count rather than concealing either category.

Mistake: Joining fields with similar names

Similar-looking platform fields may have different eligibility rules or rely on estimates. YouTube's audience documentation describes unique viewers as an estimate distinct from views; merging those measures combines unlike units and obscures what each reported figure means to a report reader.

Fix: Archive the field definition, account, period and export version for each source. Merge values only where the units match; keep them separate when eligibility or estimation differs. That leaves a reviewer able to trace the source version and population behind each reported value before interpreting it.

Mistake: Reporting a percentage without its specification

A percentage without the decision, event and eligible population has no stable interpretation. Even a plausible-looking rate may answer the wrong question when its numerator or base does not fit the decision, or when its source and reporting period are unclear.

Fix: Start from the decision question when documenting a channel choice. Record the numerator, denominator or unit, period, timezone, source version, exclusions, late data, owner and blank threshold. ONS treats statistical quality as fitness for purpose, including coherence and comparability; agency data are not official statistics. Apply fixed metric definitions and align them with the reporting dashboard.

Mistake: Using a UK recognition result as an England agency benchmark

Applying a broad UK figure to an England agency benchmark confuses geography with the question measured. The UK influencer-recognition research concerns people's recognition of advertising, not England agency response or return, so it cannot substitute for a like-for-like England campaign result.

Fix: Before comparing, match geography, population, creator type, platform, format, denominator and period; if they differ, state that no usable benchmark is available. Benchmark research should test those fit conditions rather than treating a broad national recognition finding as an England agency equivalent for the campaign decision.