Name the event before calculating its duration.

A single “response time” can mean an automatic acknowledgment, a staff reply or a department accepting responsibility. Those are different events. Agree on the start and end of each measure, the clock used and the data source.

For a nonemergency service request, useful recorded events may include received, triaged, ownership accepted, work completed, resident updated and closure verified. Not every service needs every stage, but a stage should have a stable meaning. Reopening should remain visible rather than erasing the earlier event history.

The workflow guide explains the ownership and closure decisions behind these events. This article concerns measurement definitions rather than the workbook’s operating instructions.

Use a small metric dictionary.

MeasureDefinition to agreeInterpretation limit
Time to accepted ownershipAccepted-owner event time minus received time.An acknowledgment is not owner acceptance.
Time to checked completionVerified completion time minus received time for the defined cohort.Closed duplicates or referrals are different outcomes.
Open-case ageReview time minus received time for cases still open.Young cases and old cases need separate visibility.
Reopened casesCount or rate under an explicit reopening window and denominator.A short observation window can miss later returns.
Completion outcome mixCounts by completed, referred, duplicate, withdrawn and other defined outcomes.A closed status does not establish that the requested work was done.

On a small screen, scroll the table sideways to see every column.

For durations, state whether you use elapsed or business time. A business-time calculation needs a documented calendar and rules for pauses, weekends and holidays. Do not label an elapsed-time figure as business hours.

Compare the same group of requests.

A received cohort groups requests by when they arrived. A completion cohort groups cases by when they finished. Each can be useful, but they answer different questions. A week with many old cases completed may have long completion durations even if the team reduced a backlog.

Fictional calculation: suppose a received cohort contains 20 requests and 12 have verified completion at the review date. That is 60% verified completion to date for that cohort, not a final service success rate. The other eight remain visible, along with the observation date. These are illustrative numbers, not Civicfield client results.

Where a digital transaction has a clear start and finish, GOV.UK’s completion-rate guidance (opens in a new tab) helps explain why transaction definitions matter. Completing an online submission is different from completing the operational work requested in it.

Look at unfinished work and the spread of durations.

A completed-case average omits cases that have not completed. Show open counts and age groups alongside completed work. Choose age bands that make sense for the service and display them with their actual thresholds.

A median and an upper percentile can describe different parts of a duration distribution, but include the case count and calculation method. Small samples or changing case categories can make comparisons unstable. Do not combine routine information requests and complex fieldwork into a single performance verdict without examining their differences.

A service team reviews unfinished request folders at a meeting table, an open-work stack and a separate completed tray remain visible.
Keeping unfinished service work visible.

Check the event data before judging a team.

Review missing timestamps, negative intervals, repeated IDs, reopened cases and statuses that do not agree with the event history. Record how you treat duplicates, referrals and requests outside the agency’s responsibility. Keep category changes visible when they affect comparisons.

Connected tools may record an event at different times. Use the integration planning checklist to agree on event meaning, authoritative fields and delayed-message handling before treating dashboard figures as comparable.

Close photograph of an analyst reconciling request history: paper event notes, one case folder and a laptop keyboard, careful realistic desk work without legible data.
Checking request event history.

Turn a finding into a service question.

If accepted ownership is slow, inspect routing and responsibility. If ownership is prompt but open ages grow, look at workload, dependencies and the next-action plan. If cases reopen frequently, inspect the completion criteria and the resident update rather than rewarding a higher closed count alone.

Publish only the level of detail approved for the audience and avoid exposing personal request records. Use a documented baseline and comparable observation period before claiming improvement. The measures describe the recorded process; they do not establish that software caused a result.

Use these definitions in a bounded workflow pilot, or review workflow design and implementation when the next step requires a working tool.

Validate the exported history before calculating performance.

Ask whether a row represents a current request, a status event or an attachment. Counting each event as a new request inflates intake. Using the latest status as though it existed throughout the period rewrites history. Use the export field map to distinguish the files.

Reconcile unique Request IDs with the source, validate event-parent links and examine timezone handling. Keep missing timestamps as missing. If an event was recorded late, preserve the distinction between when it occurred and when it entered the system if those values are available.

For a reopened case, retain its earlier closure and the new open period. State how the report treats repeated cycles, duplicates, referrals and cases still open at the observation date. A snapshot-only export may support a current backlog count but cannot establish earlier ownership duration or historical completion rates.

Publish a data-quality note with the denominator, observation period, exclusions and unresolved gaps. If event history is unavailable, limit the question to what the data can answer rather than inventing events to complete a chart.