Measurement
Build a community dashboard that names its decisions and shows its uncertainty
Build a community dashboard that connects member outcomes, service quality, workload, risk and cost to named decisions without hiding uncertainty.
A dashboard should help an owner decide what to continue, investigate, staff or stop. It is not a display of every number the platform exports. Build it from decision questions and keep definitions beside the result.
What to take away
- Build the dashboard from decision questions, not from every number the platform exports.
- Name the audience and the decisions each person needs to make before choosing metrics.
- Attach a metric dictionary with numerator, denominator, period, exclusions, source, owner and interpretation limit.
- Preserve data lineage and keep uncertainty visible rather than converting weak proxies into facts.
Name the audience and decisions
List the people who will use the report and what each may decide. The community operator needs outstanding work and capacity. The accountable owner needs outcome, risk and cost. Specialists may need restricted case evidence that should not appear in a general executive view.
Set frequency from the decision. Urgent safety work belongs in an operational alert, while an outcome trend may require a monthly or quarterly sample. A real-time chart does not improve a choice that should not be made hourly.
Create four layers
Member outcome: appropriately resolved needs, relevant introductions, successful learning or another defined result.
Four dashboard layers
Layer
- Member outcome
- Resolved needs, introductions, learning
- Service quality
- Access, acknowledgement, resolution, effort
- Operation and risk
- Workload, reports, appeals, incidents
- Cost and sustainability
- People time, supplier spend, cost per unit
What it covers
- Member outcome
- Service quality
- Operation and risk
- Cost and sustainability
Service quality: access, time to acknowledgement and resolution, corrections, unanswered cases and member effort.
Operation and risk: workload, reports, appeals, incidents, supplier issues and staffing gaps.
Cost and sustainability: people time, supplier spend, exceptional work and cost per meaningful service unit.
The four layers cover the main decisions an owner must make: whether needs are met, how well the service performs, what risks are live, and what the work costs.
One first screen could use four tiles:
| Tile | Layer | Question it answers |
|---|---|---|
| Resolved member needs | Outcome | Are needs being met? |
| Median time to first response | Quality | How quickly do members get help? |
| Open safeguarding incidents | Risk | What urgent harm is live? |
| Cost per resolved case | Cost | What does each result cost? |
Define every calculation
Attach a metric dictionary with numerator, denominator, period, exclusions, source, owner and interpretation limit. Show both the rate and its underlying count. A 100 per cent result from one case should not look more certain than 90 per cent from hundreds.
Metric dictionary fields
- Numerator
- Denominator
- Period
- Exclusions
- Source
- Owner
- Interpretation limit
One worked row for a metric dictionary:
Field / Example entry
- Numerator
- Cases resolved within the published service target
- Denominator
- All cases closed in the period
- Period
- Calendar month
- Exclusions
- Test accounts, duplicate cases and staff-generated cases
- Source
- Case management system closed-case table
- Owner
- Community operations lead
- Limit
- Does not show whether the member was satisfied
The GOV.UK Service Manual's setting performance metrics guidance covers UK public services. Its emphasis on using data to improve a service is relevant, but a commercial community must select measures for its own member task and governance.
Annotate changes to definitions, tracking, rules and membership. Do not draw a smooth trend through two methods that count different events.
Add qualitative evidence
Include a short, permissioned account of unsuccessful and successful journeys. Quotes need context and should not be selected only for emotional force. Link findings to the research record rather than placing identifiable detail in the dashboard.
Qualitative evidence rules
- Permissioned journey accounts
- Quotes with context
- Link to research record
- Sample answer quality
- Sample moderation quality
Use sampling to assess answer and moderation quality. Automated counts cannot decide whether specialist advice was correct or an appeal was fair.
Preserve data lineage
For every tile, link the source query, transformation and quality check. Record which system owns the event and how duplicates, staff activity, deleted accounts and late records are treated. A corrected calculation should carry an effective date and explanation.
Reconcile a sample back to the underlying member journey. Counts can drift after an integration update even when the chart still loads normally. Keep automated failure alerts separate from evidence that the metric remains conceptually valid.
Use consistent reporting cut-offs. Mixing real-time activity with a monthly cost total makes relationships appear current when periods differ. Display the last complete date and expected lag.
Allow authorised readers to reach supporting detail without copying sensitive case material into the main dashboard. Aggregation should protect people and preserve important exceptions rather than smoothing them away.
Protect access and privacy
Give viewers the minimum detail needed for their decision. Small segments can identify people even without names. The Information Commissioner's Office's accountability framework can help examine governance, access and reporting practice; it does not approve a particular dashboard.
Set retention and export controls for case-level data. Avoid sending static copies broadly when a controlled view will do.
Test the dashboard
Ask each intended reader to interpret a change and state the action they would take. If identical evidence leads to incompatible conclusions, improve the definition, note or decision rule. Check that adverse results remain as visible as positive ones.
Maintain an owner and change log. Remove charts that never affect a decision. A useful dashboard reduces ambiguity while keeping uncertainty visible; it does not convert weak proxies into facts through colour and polish.
Before you act
- List the users and the decisions they will make.
- Set reporting frequency from the decision, not from the data feed.
- Define every calculation and show the underlying count.
- Annotate any change to definitions, tracking, rules or membership.
- Give viewers the minimum detail needed for their decision.
- Ask each reader to interpret a change and state an action.
Common questions
How should a dashboard handle activity metrics such as posts and visits?
Activity figures are useful for context, but they do not show whether needs were met or risks were handled. A dashboard should lead with member outcome, service quality, operation and risk, and cost and sustainability. Keep posts, visits and new accounts visible, but not as the first screen.
Why should a rate always be shown with its underlying count?
A single case can produce a misleading 100 per cent. Pair every rate with its count so small samples are not read as firm results.
What should happen when a calculation is corrected?
Give it an effective date and a short note explaining what changed. Do not join methods that count different events into one trend.
