Multilingual social marketing metrics for release control

Before any campaign result means something, the work must be approved and then released as approved. Three control metrics check that, and each has an owner and a decision attached. They sit upstream of the audience task and conversion measures that come later.

What to take away

Approval completeness decides whether an asset may enter scheduling.
Release fidelity compares each live item with its approved asset and triggers pause or correction.
Data-quality exception rate shows whether a published measure is reliable enough to use.
None of the three scores campaign success, translation accuracy or comprehension.
Reach and reactions stay diagnostics, not headline measures.

Approval completeness: may this asset be scheduled?

The measure answers one question: does every eligible asset carry all required language, subject and accessibility decisions, recorded against one version? The eligible population is target-language assets submitted in the reporting period.

The numerator is assets with every applicable approval and the denominator is all eligible assets. Exclude cancelled items and show them separately. The source is a versioned approval register and query, owned by the quality lead.

Approval completeness check

All required language decisions recorded
All required subject decisions recorded
All required accessibility decisions recorded
Every applicable approval present
Cancelled items excluded and shown separately

The decision rule is plain: hold any unapproved asset, and set no universal percentage threshold. The AQuA Book separates verification and validation, which means checking that analysis satisfies its specification and checking that it is fit for the purpose it is used for. It supplies no marketing target.

Release fidelity: does the live item match the approved one?

Release fidelity finds configuration errors quickly. The event is a live item compared with its approved asset, across releases during the stated period and buyer timezone.

The numerator is items matching copy, language, disclosure and destination, with account and audience settings matched too. The denominator is eligible releases, and platform-cancelled items are excluded before release. Sources are the publishing log, the platform identifier and the check version.

Release fidelity match check

Match

Copy: Approved copy
Language: Approved language
Disclosure: Approved disclosure
Destination: Approved destination
Account settings: Match account
Audience settings: Match audience

Mismatch

Copy: Pause or correct
Language: Pause or correct
Disclosure: Pause or correct
Destination: Pause or correct
Account settings: Pause or correct
Audience settings: Pause or correct

Any mismatch triggers pause or correction under the buyer's rule. For illustration, if an approved asset states a price of 49 and the live post shows 39, the publishing controller pauses the post and corrects it before relaunch.

A buyer's rule here means a documented decision rule owned by the buying organisation, for example a named marketing manager approves corrections above a set value. A platform label stays in its original column unless its event and identity rules are documented.

Data-quality exception rate: can the number be trusted?

This measure decides whether a published figure is reliable enough to use. The event is an eligible record failing a pre-specified validity, completeness, duplication or reconciliation check.

The numerator counts records with at least one unresolved exception and the denominator is all records eligible for the metric in the same period. Show exception classes as counts as well. Record the source extract, validation-query version, timezone, exclusions and late-data rule.

The analyst owns the investigation. The decision owner withholds or qualifies the affected metric under a buyer-defined rule. The rate does not score campaign performance. It exposes whether the evidence pipeline behaves as specified, and where bias direction can be reasoned it should be reported, or marked unknown when it cannot.

Read the three together

Measure	Numerator	Denominator	Rate	Owner	If it fails
Approval completeness	108	120	90%	Quality lead	Hold unapproved assets
Release fidelity	195	200	97.5%	Publishing controller	Pause or correct
Data-quality exception rate	Records with an unresolved exception	All eligible records	Set per study	Analyst, then decision owner	Withhold or qualify the metric

The numbers are illustrative for a single campaign. A high approval rate with low fidelity means assets were cleared but not released as cleared, so the fault sits in publishing.

Where control ends and audience measures begin

Once approval and fidelity hold, audience task completion becomes meaningful. Its denominator is eligible attempts, not impressions. Record period, timezone, language-asset version and source systems, then the joining rule, duplicates, failures and missing data.

Audience task completion flow

Define eligible task and population
Count eligible completed events
Count eligible attempts, not impressions
Record period, timezone, language version
Record joining rule, duplicates, failures
Record missing data and owner

Do not label people by inferred language, culture or ethnicity. Where research uses a sample, the ONS explains that the gap between a statistic and the population value comes from sampling error and non-sampling errors.

Attributed conversions remain observational unless a causal design supplies a counterfactual, as the guide to attribution methods sets out. The wider reporting frame is in measuring without confusing attribution with cause.

Checklist before the next release

Approval register holds language, subject and accessibility decisions per version.
Cancelled items are excluded from the denominator and listed separately.
Each live item is compared with its approved asset before it counts as matched.
A written buyer's rule says who pauses or corrects a mismatch.
Exception checks, extract and validation-query version are recorded.

Common questions

What should happen when a live item does not match its approved asset?

Any mismatch triggers pause or correction under the buyer's rule. The publishing controller owns the measure, and platform-cancelled items are excluded before release.

Does a high approval rate show the campaign works?

No. This operational measure does not show translation accuracy, comprehension or campaign success. It shows only that required decisions were recorded before scheduling.

How does the exception rate differ from campaign scoring?

It does not score campaign performance. It exposes whether the evidence pipeline behaves as specified, and the decision owner withholds or qualifies any affected metric.