Community management key metrics: honest denominators

Community management key metrics should answer decisions about the member service. Each measure needs a clear definition, denominator, collection limits and owner. Do not compare a rate borrowed from another platform unless its population, action, period and exclusions match; otherwise it may describe different things.

What to take away

Begin with the member outcome

Name the task the community is meant to improve. For peer support, measure whether an eligible question receives an appropriate answer and whether the person can proceed. For introductions, examine whether a relevant connection occurs and leads to the intended activity.

Put denominators into the calculation

Set eligibility against written service-scope rules before counting. For example, define eligible needs as in-scope questions received during the reporting period; exclude spam, tests and out-of-scope requests under documented rules, and report exclusions separately.

State the cohort, period, unit counted and exclusions for each measure. These definitions are examples, not benchmarks; use the same rules when reviewing results over time.

Define success before looking at results. A reply is not automatically an answer, and an administrator marking a thread solved may not reflect the member's view. Review samples for quality.

The GOV.UK Service Manual’s Designing with data: an introduction is intended for public services. Its distinction between measurements and meaningful service metrics is useful for a commercial community, but its government indicators do not automatically apply.

Measure access and reach

Track invitations, joining attempts, successful access, abandoned registration, account recovery and use of alternative support. Segment only where there is a valid purpose and adequate data.

Membership totals exclude eligible people who never reached the service. Interviews and usability work can uncover barriers hidden by platform analytics. Keep England-only findings separate from wider UK or international audiences.

Use activity to understand capacity

Registrations, active members, posts, replies, searches, events and reports help an operator forecast work. Each needs a time period and rule. "Active" might mean login, view or contribution, which can produce materially different counts.

Do not use activity as impact. A rise in questions may reflect growth, confusion or missing documentation. Pair the count with journey evidence.

Monitor workload and service

Measure outstanding needs, acknowledgement, resolution, expert referrals, moderation time, appeals and administrator tasks. Use percentiles or ranges where an average hides difficult cases.

Record work completed outside the platform. Email escalation and product-team answers still consume capacity even when the dashboard cannot see them.

Review quality and safety

Sample answers for relevance, accuracy, qualification and correction. Review moderation decisions for applicable rule, context, communication and appeal result. Count incidents by consequence and route, not merely by labels chosen at intake.

A low complaint rate may mean few problems or a difficult reporting process. Test the route and ask unsuccessful members about their experience.

Include cost and sustainability

Relate complete cost to a defined service unit. For cost per appropriately resolved eligible need, divide the total people, technology, specialist review, change and exit costs for the same period by the number of eligible needs appropriately resolved under the agreed quality criteria. State the period and cohort. A lower cost per post can reward unnecessary volume.

Collect proportionately

The Information Commissioner's Office explains the data minimisation principle. Do not collect profile attributes or track behaviour merely because segmentation is possible. Establish purpose, access, retention and explanation.

Review variation without overclaiming

Segment results only where the group, sample size and purpose support interpretation. A difference between two small groups can arise from chance, recruitment or unequal opportunity to participate. Show counts and uncertainty, then use research to investigate the experience.

Look for distribution as well as average. The median time to answer can improve while a few high-risk questions wait much longer. Report the tail that matters to the service promise.

Compare periods only after checking member mix, tracking, staffing and campaigns. Annotate launches, outages and rule changes. If the service changed materially, treat the new data as a different operating condition rather than forcing continuity.

Assign a review action to each key measure. A threshold might prompt a sample, capacity check or service redesign, not an automatic judgement about staff performance.

Maintain a metric dictionary with name, question, numerator, denominator, exclusions, source, frequency, owner and decision. Retire measures that nobody uses. A small set with quality review is more useful than a crowded scorecard that makes weak proxies look authoritative.

Before you act

Common questions

Why is a universal engagement rate not useful?

Engagement can mean a login, view or contribution, so a rate may use unlike numerators and populations. Define it explicitly; for example, members making at least one qualifying contribution in a reporting period ÷ all eligible members in that period. Do not compare rates with different actions, populations, periods or exclusions.

How should you measure quality and safety?

Check a selection of replies and moderation calls for accuracy, context and appeal outcomes. Group incidents by their impact and how they arose, not by intake labels.

What should a metric dictionary include?

Record the measure’s name, the question it answers, its calculation (including numerator, denominator and exclusions), source, frequency, owner and the decision it triggers. Define the cohort and period so another person can reproduce it. Drop measures nobody uses.