Measurement
Part of Lead network measurement that preserves the state ledger
Lead network key metrics: define each rate, show data loss and tie it to action
Lead network key metrics for England: counts from submitted to credited, each rate with its numerator and denominator, data loss, PECR consent and TPS checks.
Lead network key metrics: define each rate, show data loss and tie it to action
Use the route below as a decision tree: start with the question that matches the reporting problem, then follow the stated action. Keep state counts, their definitions and any data loss visible before treating a rate as a performance signal.
If you need to map where records stop
Count records separately at every service state, rather than rolling the path into one conversion percentage. An England operator needs counts for submitted, quarantined, validated, matched, disclosed, accepted, rejected, credited, contacted, fulfilled, complained, suppressed and exited records. Each count answers a different question about where records stop moving.
Keep rates secondary to those state totals. For each percentage, specify its numerator, denominator and arrival window before interpretation; the service standards article connects such counts with contract clocks and tolerances. If submitted, matched and disclosed records are blended, the point of stoppage becomes difficult to identify.
Rule of thumb: if the question is where work stopped, show the state counts first and use a rate only to compare two named states over a stated window; do not substitute one blended conversion percentage for the service path.
If a rate is due to go into a report
Give every measure a written contract before comparing results. Record purpose, eligible population, numerator, denominator, unit, dates, time zone, inclusions, exclusions, source event, query version, owner, threshold, response and review trigger; without these fields, analysts can produce different figures for the same month.
Choose criteria that are clear and fit the problem. The OSR model guidance is worth reading first. Give each measure a named owner and review trigger as well, so analysts use the same definition and know who examines a result when its agreed review point arrives.
Rule of thumb: if two people could produce different figures for the same month, do not publish the percentage until its population, time basis, query version and accountable owner are recorded together with its numerator, denominator and review trigger.
If stage rates need to be compared
Use each stage's own denominator and label invented figures as illustration. The table keeps quarantine, disclosure, credit, acceptance, rejection, contact and fulfilment rates distinct, so a decline at one point cannot be disguised by combining unlike transitions. These figures represent one illustrative month.
| Measure | Numerator | Denominator | Illustrative result |
|---|---|---|---|
| Quarantine share | 150 quarantined | 2,000 submitted | 7.5% |
| Disclosure rate | 1,260 disclosed | 1,400 matched | 90% |
| Credit rate | 80 credited | 1,000 accepted | 8% |
| Acceptance rate | 1,000 accepted | 1,260 disclosed | 79.4% |
| Rejection rate | 200 rejected | 1,260 disclosed | 15.9% |
| Contact rate | 700 contacted | 1,000 accepted | 70% |
| Fulfilment rate | 350 completed services | 700 contacted | 50% |
Report each rate beside raw counts and state that retries were deduplicated before counting submitted event IDs. A credit is a commercial remedy, not evidence that a record was invalid. Keep accepted, rejected, disputed and credited states separate, even though only the credit rate appears in this illustration.
Rule of thumb: a percentage is interpretable only when its own numerator and denominator are visible, retries have a stated treatment, and a commercial credit remains separate from the question of whether the underlying record was invalid or not in the reporting record.
If intake volume or evidence raises concern
Count distinct submitted event IDs received during the period, after stating how retries are deduplicated, and show quarantined items beside submissions. Calculate quarantine share as quarantined records over eligible submissions, using the same arrival window for both counts; otherwise the comparison is not aligned.
Use this as source control. Missing notice evidence, unexplained fields and unknown provenance should stay visible, not be quietly removed. The quality checklist describes the provenance gates behind these counts, while measurement mistakes explains what to do when a figure changes.
Rule of thumb: when submission and quarantine counts do not share a defined arrival window, pause the comparison and inspect the source evidence before treating the quarantine share as a change in performance or adjusting the intake route based on that percentage alone.
If a proposed match does not become a disclosure
Count actual transfers as disclosures, separately from proposed matches. Their gap can point to availability, safety or operational holds; combining the counts conceals where movement stopped and makes a later fix harder to target. Check recipient eligibility, service availability and transfer holds before treating the gap as a demand problem.
Keep buyer decisions separate from the proposed-match step. Reconcile the state counts before applying attribution methods; otherwise the reporting cannot show whether records stopped before transfer or at a later decision.
Rule of thumb: treat proposed matches, actual disclosures and buyer decisions as separate steps; check the reason for a break between states, and use attribution only after the underlying ledger has reconciled for the reporting period in question and counts are complete.
If contact is being treated as fulfilment or permission
A dial, sent email, answered conversation, appointment and completed service are different events. Define the evidence that qualifies as contact before reporting it. For a fulfilment rate, put completed services in the numerator and use eligible contacted or accepted records as the denominator.
Contact for a requested service does not grant permission for later direct marketing. Make consent for electronic mail specific to the kind of electronic mail intended, and check TPS and CTPS registers before a live marketing call under PECR and UK GDPR. Review UK regulations separately, including the suppression population.
Check each record's subscriber type before building a campaign metric: sole traders and some partnerships count as individual subscribers, while corporate subscribers are treated differently for consent and soft opt-in. Establish that classification before deciding how the campaign measure is constructed for the intended route.
Rule of thumb: keep contact and fulfilment separate, treat requested-service contact and later promotion as different purposes, and build any campaign measure only after each record's subscriber type and applicable suppression status are clear for the intended route and electronic-mail plan.
If a rate comes with incomplete event data
Display data loss beside each rate: missing transitions, duplicate IDs, late records, rejected joins, freshness time and reconciliation difference. Count eligible events lost only when the loss population has been defined, and never rewrite prior results silently after a query update; otherwise the rate lacks an interpretable record base.
Read these counts with the 90% disclosure rate: its value depends on the records behind it. Complaint reporting should show stage, source and recipient category using the minimum necessary information. Suppression reporting should verify propagation without displaying objectors. Put these panels on the reporting dashboard.
Rule of thumb: a rate should travel with its raw counts, freshness and reconciliation information, and any defined loss; if the loss population is unknown, say so rather than implying the denominator is complete for that reporting window and query version.
If a measure crosses its agreed threshold
A metric without an agreed response is decoration. If quarantine share rises beyond the source owner's tolerance, pause only the affected intake and inspect provenance. If proposed matches stop becoming disclosures, check recipient eligibility, service availability and transfer holds before assigning the change to demand.
A rise in buyer rejections should trigger a review of rule versions and reason codes, not an automatic quality claim. A suppression failure should stop the relevant marketing route while the requested-service pathway is assessed separately. Keep these responses distinct from any decision about restarting a route.
Name who may restart each route and what evidence they need. Apply that control separately to affected intake, marketing and requested-service activity; a threshold should cause the agreed review, not an unrecorded reset. If no response can be named, drop the measure instead of leaving it as decoration.
Rule of thumb: every threshold needs a named response, restart owner and evidence requirement; pause only the affected route where specified, and remove a measure whose signal has no agreed action or authorised decision-maker before relying on it operationally for that signal.
If validation timing is the decision
Use completed validation decisions as the numerator and eligible quarantined records whose approved clocks have ended as the denominator. Include both pass and fail decisions; exclude only documented open clocks. State the rule version, pauses and late events so reviewers can trace the measure.
Rule of thumb: count only eligible quarantined records whose approved clocks have ended, include pass and fail decisions, and disclose open-clock exclusions, pauses, late events and rule versions so the result can be reviewed against the same definition by another reviewer later.
