Reporting errors that quietly undermine corporate communications measurement

This note lists seven corporate communications measurement mistakes that can change a decision or overstate evidence.

Method: desk review on 6 September 2026. Scope: an illustrative fictional organisation in England; cited UK government or regulator material keeps its stated scope. Ordering: workflow sequence, not severity. Conflicts: none declared.

What to take away

1. Starting with a number instead of a question

A dashboard tile is not a decision. Record the authorised question, eligible audience and next action before choosing a measure. GOV.UK's guidance on setting service metrics starts with service purpose and hypotheses. It applies to government services, so it is a useful prompt rather than an England corporate benchmark.

2. Treating output as outcome

Sending approved copy, receiving a supplier status and observing audience action are different events. The UK government's communication functional standard requires government campaigns to monitor outputs, outtakes and outcomes. Preserve the distinction instead of turning delivery into understanding.

3. Leaving the denominator unnamed

A rate fails if the numerator covers delivered messages while the denominator covers all stored contacts, or if employee and customer populations are mixed. The ONS description of statistical quality includes coherence and comparability. Although written for official statistics, it supports the practical check that like must be compared with like.

For example, a report divides link clicks by messages delivered, then compares that rate with the proportion of the full contact database who responded. The first denominator is delivered messages; the second is stored contacts. The two are not comparable.

4. Hiding missing and revised records

Late imports, rejected rows and changed status definitions can alter a series. Publish source coverage, missingness and revision notes beside the measure. The UK Statistics Authority's Trustworthiness, Quality and Value framework stresses transparency and suitable methods. Voluntary use does not confer official status.

5. Calling chronology causation

A favourable event after a message may reflect other operational, economic or stakeholder changes. HM Treasury's QPIE guidance explains that monitoring outcomes does not identify whether an intervention caused the change.

A defensible comparison would set out a pre-specified counterfactual, such as a matched control group, a comparable period or an interrupted time series, and state its assumptions and limits. Without a defensible comparison, label the evidence descriptive or attributed by rule.

6. Collecting identity data because it is available

Personal-level reporting can create unnecessary risk when an aggregated exception count would answer the question. The ICO's data-protection principles include purpose limitation, data minimisation, accuracy and storage limitation; the page also notes a DUAA-related review. A qualified privacy reviewer must assess the real processing.

7. Publishing a visual that cannot be interrogated

A chart may conceal definitions from screen-reader users or anyone needing the underlying table. The Government Analysis Function's 2026 dashboard accessibility guidance recommends testing and machine-readable supporting data. Its government-analysis remit should be stated.

For Cedar Vale Engineering Ltd, the fictional company in this example, any one of these failures moves the affected measure to unresolved. The owner then corrects the source, definition or presentation and issues a new version. A deadline, attractive result or senior preference cannot convert an evidential defect into a pass.

The repair log should name the affected report, original measure definition, evidence exposing the fault, correction owner and date. It should also identify downstream decisions that used the old version.

Where a correction cannot be propagated, mark the affected conclusion as withdrawn, not leave readers to compare two unexplained figures. Re-admit the measure only after an independent reviewer can reproduce it from the retained source and current query.

Before you act

Common questions

What should be recorded before choosing a measure?

The authorised question, the eligible audience and the next action. Without those three items, a dashboard tile cannot guide a decision.

Why does an unnamed denominator undermine a rate?

Because numerator and denominator may cover different populations or stages. Check that both cover the same eligible group before using the rate.

How should a measure be re-admitted after correction?

Only when an independent reviewer can reproduce it from the retained source and current query. The repair log should also list downstream decisions that used the old version.