Measurement
Part of Fitness creator measurement in England: defining health metrics from the CMO activity guideline
A fitness creator dashboard with delivery, audience, commercial and health panels kept apart
Build a fitness creator dashboard that keeps delivery, audience, commercial and health evidence apart while exposing definitions and uncertainty.
A fitness creators reporting dashboard should make weak evidence visible, not turn it into a confident headline. Build four panels for operations, audience behaviour, attributed commercial activity and health evaluation. Do not total them into a single score.
This is a desk-designed specification for an England-based publisher, researched on 5 September 2026. It reports no tested implementation or performance. Platform fields are global product definitions, while privacy controls derive from current UK guidance. Advertising controls also apply UK-wide. It is written to support 2027 planning.
What to take away
- Keep operations, audience, commercial and health panels separate and never total them into one score.
- Give every tile a dictionary row with population, numerator, denominator, owner and update trigger.
- Label commercial results campaign-attributed, not proof that content caused a purchase.
- Link each tile back to its raw export, transformation log, dictionary version and approval.
- Hold the dashboard when a denominator cannot be reproduced or a causal conclusion lacks an evaluation.
- Write the decision first and state what evidence would change it before choosing a metric.
- Attribution models assign credit under their own rules and do not show incremental revenue.
- Health claims need an evaluation gate and actual substantiation, not views or testimonials.
1. Give every tile a decision and definition
Before connecting data, create a dictionary row with population, numerator, denominator, unit, period, timezone, inclusions, exclusions, source, uncertainty, owner and update trigger. Add the decision supported and current version. A tile with no denominator or owner stays hidden.
Write the decision first: continue a format, repair a release control, investigate an audience drop, reconcile a sponsor report or commission a health evaluation. Then state what evidence would change that decision, so a convenient platform number cannot stand in for the question that matters.
The 2026 Magenta Book separates process, impact and value-for-money evaluation and says design should be proportionate to the decision. It governs central government evaluation, not private publishers, but the same distinction applies to creator work. Routine monitoring can show that something moved; it cannot identify why.
The "publication control rate" covers all releases scheduled Monday to Sunday. Its numerator counts eligible releases whose published version has all required approvals; its denominator includes every scheduled release, including late and withdrawn items, but omits approved pre-schedule tests.
It is a percentage on Europe/London time, sourced from release records and owned by operations, and a missing approval triggers an immediate hold. Mis-tagged risk and wrong-version matching are stated uncertainties.
For this control, an illustrative threshold is 95 per cent completion and zero missed clinical reviews. It is a local control measure, not an England benchmark.
Use a versioned metric contract template, for example a Google Sheets tab named Metric contract.
2. Separate the four panels
Give every report one of four labels: content operations, audience behaviour, commercial outcomes or health outcomes. The labels stop a smooth but invalid progression from impression to view, from view to purchase and from purchase to health benefit.
A video can be published on time, watched for longer than usual and followed by orders, yet none of that proves it improved anyone's health.
The operations panel shows control completion, late releases, corrections and open incidents. The audience panel may show platform impressions, views and watch time, only with the exact product definition and filter.
YouTube says its registered impressions are thumbnail displays on YouTube and reports external traffic separately in the Content analytics documentation. Keep surface, format and extraction date beside the value.
The commercial panel reconciles tagged sessions, eligible orders, refunds and contribution. Google's URL-builder record says campaign parameters identify referring campaigns in acquisition reports. Label results "campaign-attributed" under the named rule. They are not proof that the content caused a purchase.
The health panel contains protocol status, recruitment, follow-up, missingness, adverse-event handling and approved outputs. It should not contain views presented as wellbeing impact. The UKHSA Evaluation Strategy separates monitoring from process, impact and economic evaluation. That strategy applies to UKHSA, but the distinction prevents an editorial dashboard from overstating evidence.
3. Preserve the route back to evidence
Each tile links internally to its raw export or reproducible query, transformation log, dictionary version and approval. Record account, property, report, field, filters, extraction timestamp and responsible person. Use a visible data freshness time rather than an unexplained green status.
Show missing values and late feeds. Do not replace them with zero. Break the series when a platform definition, consent design, campaign taxonomy or attribution model changes. A footnote should name the affected dates and whether earlier values were restated.
4. Reduce privacy risk
Prefer aggregated counts and role-based access. The ICO's data minimisation principle says personal data should be adequate, relevant and limited to what is needed. Its current tracking-technology guidance covers cookies, pixels, link decoration, scripts and tags. A privacy reviewer must settle purpose, lawful basis, any consent, retention and vendor access before collection.
Do not show participant-level health data in a general business view. Restrict an approved research dataset to the evaluation team and publish only disclosure-controlled outputs. Small cells, free-text comments and location combinations can expose people even when names are removed.
5. Add interpretation and release gates
Every reporting period gets a data-owner sign-off, an analyst check and, where relevant, health, privacy and advertising review. The note beneath each panel states what the figures do not show. CAP says testimonials do not substantiate objective claims, so favourable comments never move into the health-outcome panel.
A registered health professional, statistician or evaluator, privacy adviser and advertising reviewer must approve the parts within their competence before publication. For a small England-based publisher, use a named operations lead, a retained statistician or evaluator, an external privacy consultant and a freelance advertising reviewer. Record each sign-off and the basis for competence.
Hold the dashboard if a denominator cannot be reproduced, a filter changed without a break note, identifiable health data appears outside its approved area or a causal conclusion lacks an evaluation. The honest display is "not established", not a polished substitute metric.
Before you act
- Create a dictionary row before connecting any data.
- Hide any tile with no denominator or owner.
- Record surface, format and extraction date beside each value.
- Show missing values and late feeds instead of zero.
- Settle purpose, lawful basis, consent and retention with a privacy reviewer.
- Get data-owner sign-off and analyst check each reporting period.
Common questions
What should a tile show before any data is connected?
Each tile needs a dictionary row covering population, numerator, denominator, unit, period, timezone, inclusions, exclusions, source, uncertainty, owner and update trigger. It also records the decision supported and current version. A tile with no denominator or owner stays hidden.
Why should commercial results be labelled campaign-attributed?
Google's URL-builder record says campaign parameters identify referring campaigns in acquisition reports. The article says results should be labelled campaign-attributed under the named rule. They are not proof that the content caused a purchase, so the label keeps the claim within the evidence.
When should the dashboard be held rather than published?
Hold it if a denominator cannot be reproduced, a filter changed without a break note, identifiable health data appears outside its approved area, or a causal conclusion lacks an evaluation. The honest display is not established, not a polished substitute metric.
