A distributor reports ten pump returns this month, and purchasing wants to conclude that a production batch has deteriorated. First establish what “ten” means. Is it ten requests, ten serial numbers or several repeat repairs? Is the denominator shipments, sold units or units with a comparable period in service?
Without consistent definitions, the count measures service workload more readily than product performance. A useful dashboard separates those questions instead of forcing them into one apparent failure percentage.
Keep four quantities separate
Request counts describe the number of incoming service events. Unique affected units show how many pumps entered the complaint process. Confirmed product faults describe the outcome of completed investigations. Repeat repairs reveal unresolved problems or weaknesses in repair confirmation.
Transport damage, unsuitable application, conditions outside the agreed boundary and faults not yet reproduced also need explicit states. Do not force every return into one fault category. A single ticket can cover several pumps, while one pump can generate multiple tickets, so the records need an equipment identity as well as a ticket identity.
Where serial information is incomplete, improve the association rule and report uncertain identities separately. Deduplication should follow a documented definition, not a desire to improve the percentage.
One assumed dataset produces several legitimate measures
Suppose a cohort shipped 240 pumps, of which 200 have been sold. By the stated observation cutoff, ten requests concern eight unique pumps, with five pumps confirmed to have product faults. These are illustrative figures, not company data.
Using sold units as the denominator gives a request-count ratio of 5%, an affected-unit proportion of 4%, and a confirmed-fault report proportion of 2.5%. All can be useful if labelled correctly. None automatically represents the probability of failure over the product’s whole life.
Unresolved investigations may change classification. Some problems may never be reported. The forty unsold pumps have not experienced equivalent user exposure. Switching the denominator to 240 makes a ratio smaller without improving the pumps.
| Measure | Numerator | Main question |
|---|---|---|
| Request workload | Requests or tickets | What service capacity is needed? |
| Unique-unit claim proportion | Pumps with at least one claim | How many sold units entered the process? |
| Confirmed-fault report proportion | Pumps with confirmed product faults | What has the evidence established? |
| Repeat-repair proportion | Units returning again under the agreed definition | Was repair or closure effective? |
Align the observation window before comparing cohorts
A batch sold two weeks ago cannot fairly be compared with another that has completed a wet season. Group units by sale or first-service month, then compare at the same service age. Ninety days is one illustrative checkpoint, not a universal reporting requirement.
Duty intensity, water quality and operating hours can differ. If those data are unavailable, state the limitation. A simple claim proportion cannot be converted into failures per thousand operating hours without exposure information. Small changes in a small sample also do not establish that a redesign has succeeded.
Formal life or reliability estimates require suitable statistical methods, including treatment of units that have not yet failed and records with incomplete observation. Keep that analysis distinct from a routine service dashboard.
Use the dashboard to select the next investigation
Each row should retain cohort, configuration revision, denominator source, cutoff date and investigation status. Begin with a manageable set of actionable cause codes rather than dozens of overlapping labels. Preserve dates of classification changes so later revisions do not silently rewrite an earlier report.
A concentration around one cable revision should trigger a review of that revision’s material and termination evidence. Repeated repairs may instead indicate an inadequate verification procedure. A proportion identifies where to look; it does not establish the root cause. Protect customer and personal information when sharing service findings.
Display numerator and denominator alongside each percentage. One unit can move a small cohort noticeably, and reporting habits may differ between channels. Show pending investigations and their age so delayed closure does not artificially suppress confirmed-fault reporting. The metric should support corrective action rather than reward a change in record keeping.
Frequently asked questions
Is shipment quantity an invalid denominator?
No. It can serve a defined supply-chain measure. Label it clearly and do not mix it with sold-unit or in-service measures.
Does no complaint mean the pump is healthy?
Not necessarily. It may be unused, unreported or resold. Reporting status and physical condition are different.
Can fewer claims immediately prove that an improvement worked?
Check sales volume, observation age, application mix and classification changes first, then follow comparable cohorts.