Media buying agencies measurement mistakes to check first

Media buying agencies measurement mistakes are mostly diagnosable by checking scope, definition, provenance and interpretation in that order. Whether the report comes from an in-house team or from an agency you are vetting, the same checks apply. Start there before blaming campaign performance. A report can calculate perfectly and still answer the wrong question.

What to take away

How these mistakes were selected

Inclusion covers faults that can materially change a media decision and that an advertiser can check in a brief, source export or calculation. Geography is England, while cited regulator guidance has UK scope and platform documents are global.

Exclusions are invented benchmark rates, agency rankings and claims based on hands-on testing, because none was conducted for this article.

1. The outcome has more than one definition

One team counts submitted forms, another counts accepted leads and finance counts paid orders. The totals cannot answer the same commercial question. Write the event, acceptance rule, source and reporting delay before reading the trend.

Google's global conversion tracking documentation separates website, app, phone and offline conversion routes in Google Ads. If the report names Google Ads, retain the configured action rather than reducing unlike actions to "conversions".

2. The denominator changes silently

A cost rate can move because the cost scope changed, even when delivery did not. Check whether media, agency fees, production, technology and tax are included. Recalculate both periods on the same basis. A blank input means unresolved; zero says the input was observed and had no value.

Worked example: a report shows a cost per lead on media spend alone. Add agency fees, production, technology and tax, and the same leads carry a higher cost. Recalculate the earlier period on the same five inputs. If the earlier rate was never recalculated, the comparison is invalid.

3. UK evidence is presented as England data

Geographic relabelling creates precision that the source never supplied. Barb's UK television measurement method, which uses panel and census-level components, concerns the UK television ecosystem. It may inform a campaign running in England, but its published UK measure does not become England-only evidence.

The reverse error also matters. Do not describe Great Britain as the whole UK. The geography label must follow the source and the campaign scope.

Diagnosis: trace the location field to its definition. Targeting location, measured exposure and customer address are not interchangeable cuts of England.

4. A tracking gap is treated as poor response

Before changing media, inspect event firing, consent state, landing pages, imports, identifiers and reporting lag. The UK ICO advertising guidance, formally the Guidance on the use of storage and access technologies, covers ad delivery, tracking and profiling. A qualified reviewer should apply it to the actual setup. Do not estimate missing people merely to make the dashboard complete.

Diagnosis: use a controlled test event and follow it through each authorised system. Mark the affected dates and fields as incomplete until the break is understood.

5. Platform credit becomes a causal claim

"Attributed by the platform" describes a model output within that system. "Caused by the campaign" is a different claim. If no suitable comparison was designed, use descriptive language. Changing several channels, prices or landing pages together further limits what can be inferred.

6. The tidy chart hides a source break

Account migrations, renamed events, revised filters and corrected invoices can split a series. Mark the break on the reporting date, preserve the previous definition and state whether past periods were recalculated. Never join two definitions with an uninterrupted line.

Correction order and checklist

Freeze the affected decision first. Trace one reported total to its export, then verify the event definition, cost basis, period and geography. Record the fault, correction, owner and date. Release the revised view only when a second reader can reproduce the number and explain what the evidence still cannot show.

Common questions

What should you do if one team counts submitted forms and another counts accepted leads?

Pick a single governing outcome for the decision and keep earlier stages as separate funnel observations. Test a sample against the client system before accepting the label. Do not combine the totals.

How can you tell whether a cost rate change is real or just a scope change?

Confirm which inputs are included: media, agency fees, production, technology and tax. Recalculate both periods on the same basis. Treat a blank input as unresolved and zero as observed at no value.

Why is it wrong to present UK evidence as England data?

Relabelling creates false precision. Barb's UK television measure does not become England-only evidence. The label must follow the source and the campaign scope.