A CAPA system that closes everything and fixes nothing

The numbers describe a process that is producing paperwork, not correction.

• 41 of 78 overdue at 90 days with a mean closure of 128 days means the target is routinely missed; either the target is unrealistic or the process lacks capacity and escalation. An overdue CAPA with no escalation is a control that does not operate.
• 62% 'operator error - retrained' is the strongest signal. A root cause that appears in two thirds of investigations and produces the same action each time is a template, not an analysis. It also distributes blame downward and hides the system causes the standard expects to be found - procedure design, labelling, workstation layout, look-alike packaging, workload.
• The four recurrences at LP-2 are the proof. Recurrence after corrective action means the action did not address the cause (8.5.2 requires action proportionate to the effects of the nonconformity and review of the effectiveness of the action taken). Retraining as the sole action for a picking error at a station handling look-alike labels is a predictable failure.
• Effectiveness checks completed by the owner on the day of closure are not effectiveness checks. There must be an interval, a defined measure, a threshold and an independent evaluator.
• The mega-CAPA destroys traceability: twelve unrelated causes, one action plan, one closure decision. It optimises the backlog metric at the cost of the only thing the metric was proxying for.
• Objectives without targets, owners or measurement methods fail 5.4.1 and remove the basis for judging whether improvement happened.

A defensible diagnosis: the process is compliant in form and ineffective in substance, and the backlog is a symptom of investigating too many things too shallowly. Fewer, deeper investigations with real containment and verification would reduce the count.

Containment first. Verify whether any mislabelled product left the building, using traceability and batch records for both references; if so, assess regulatory reporting and customer notification. Containment also covers the station: segregating the two label stocks physically and adding a verification step at packing.

Root cause analysis. The failure recurs with identical circumstances, so the analysis should examine the process, not the person:

• Observe the station at shift change and at peak rate; measure light levels, label stock proximity, and how the pick instruction is presented.
• Review the work instruction: does it require a scan or a second-person check, or does it rely on visual discrimination of two similar labels?
• Interview the operators who made the errors and those who did not - the difference is often in the conditions, not the individual.
• Use a structured method (5-why plus a cause-and-effect diagram, or fault tree) and record why each candidate cause was accepted or rejected.

Corrective action. Choose actions that remove the possibility of error rather than the memory of it: barcode scanning with a system interlock that prevents the line running with the wrong label roll; physical separation of label stock with different-coloured backing; a line-clearance check at changeover recorded on the batch record; and revised instructions only as a supporting measure.

Effectiveness check. Define before implementation: metric = mislabelled units detected at or after packing for LP-2; baseline = four events in twelve months; target = zero events over six months, with a leading indicator of scan-bypass exceptions per 1,000 packs; data source = batch records and complaint log; owner = packing supervisor; evaluator = quality, independent of the action owner. If the interlock is being overridden, the check fails even if no event occurred. Record the review date and the conclusion explicitly.

Make the data usable (8.4). Aggregate nonconformities, complaints, CAPAs and audit findings by failure mode, product family, line and supplier, not just by count. A Pareto of failure modes over 24 months usually shows that a small number of causes drive most of the volume - at Polaris the LP-2 label issue is obviously one. Analysis of data is a requirement, and a chart in a management review pack is not analysis unless it changes a decision.

Set objectives that can be judged (5.4.1). Replace 'reduce defects and improve customer satisfaction' with something like: 'reduce packing-related nonconformities by 50% against the prior 12-month baseline by Q4; owner: operations manager; measured monthly from the nonconformance log'. Each objective needs a target, a metric, an owner and a review frequency. Objectives should be consistent with the quality policy and deployed to relevant functions.

Triage the CAPA backlog honestly. Sort the 78 open CAPAs by risk and reclassify those that are really corrections (fix the batch) rather than corrective actions (fix the cause). Close corrections quickly with records; keep genuine CAPAs but reduce their number and increase their depth. Set an escalation rule for overdue CAPAs that reaches top management, so overdue becomes visible rather than tolerated.

Management review (5.6). The inputs required by the standard include feedback, complaints, reporting to regulators, audits, process and product conformity, corrective and preventive actions, follow-up from previous reviews, changes and recommendations. Polaris reviews three of these. The review must produce decisions and actions - resource allocations, objective changes, CAPA priorities - with owners and dates, minuted and followed up at the next review. A review with no outputs is not a management review; it is a briefing, and an auditor will treat the absence of outputs as evidence that top management is not engaged with the QMS.