Claims center: closing a project built on temporary inspection

Not ready to close.

The headline 7.8% to 2.1% is not attributable to the permanent process for three reasons.

1. Two temporary resources were part of the treatment. The Belt's daily audit of 30 claims is a feedback and correction mechanism, and the second-shift reviewer is 100% inspection. Both are interventions that stop on closure. The post-closure process is therefore a different process from the one measured, and regression toward a higher error rate is the expected outcome, not a surprise.
2. The error-proofing covers a small slice of the problem. Blocking 3 of 11 format modes that represent 12% of defects leaves roughly 88% of the defect mass untouched by the fixed countermeasure.
3. The other controls decay with the population. Training alone with 40% annual turnover on the shift is a control that evaporates in under a year, and the reaction plan does not name an owner, a threshold, or a containment action, so it cannot be audited.

Evidence required before closure

• At least two to three months of the metric with the Belt's audits and the temporary reviewer removed, demonstrating the target is still met.
• An adherence audit showing process owners, not the Belt, are running the control plan.
• Updated failure mode analysis covering all 11 error modes, with a documented decision on which are error-proofed, which are detected, and which are accepted.
• A quantified benefit verified by Finance against the new baseline, not against the pilot window.

A defensible closure statement is: the process holds 2.1% without project resources. Without that evidence the gain will be claimed and then quietly lost.

Chart choice

The p-chart is the right family for proportion defective with a daily subgroup. Two qualifications: at 2.1% and n = 250, the expected count per point is about 5 defectives, so the limits are wide and detection of a one-point shift is weak; a subgroup of 500, or a chart on defects per claim if a claim can carry several defects, would be more sensitive. The control plan must also state which signal rules are in use, because six points on one side is a signal under several standard rule sets but not under a bare three-sigma-only rule.

The signal

Six consecutive points above the center line has a probability of about 1.6% under randomness, so it is a genuine signal of a shifted mean, not noise. The plausible assignable cause is already known: the temporary second-shift reviewer was withdrawn, so second-shift defects are no longer being caught before they are counted.

What the supervisors are doing wrong

Adjusting the field-mapping settings after each point is tampering, over-adjustment to common-cause noise. It does not remove the cause, it adds variability, and it destroys the chart's baseline so that limits no longer describe the process. Repeated adjustment to chase a single point is one of the classic ways a control chart is defeated.

Correct response sequence

1. Stop adjusting. Freeze settings and record any change already made, with timestamps.
2. Investigate with stratification by shift, error mode and claim type to confirm where the shift sits.
3. Treat the withdrawal of the temporary reviewer as a known process change: document it, then re-baseline the chart as a new phase once the process is stable at the new level, rather than treating every subsequent point as an isolated special cause.
4. If the new level is unacceptable, that is an Improve-phase problem to be solved with a permanent countermeasure, not a control-chart conversation.

Control plan, minimum viable version

What is controlled	How and how often	Owner	Reaction if out of standard
Error rate	p-chart, daily n = 500, rules stated explicitly	Team leader	Signal triggers stratification by shift and mode within 24 hours; no setting changes
Top 3 error modes	Weekly Pareto from the system, not by audit	Quality analyst	Any mode above its target goes to the process owner with a dated action
Field-format entry	Error-proofing on all 11 modes where technically possible, with the 8 remaining prioritized by frequency	IT and process owner	Blocked-edit failure logged and escalated
Training currency	Skills matrix by shift; new-hire qualification before independent work	Supervisor	Unqualified staff do not work un-reviewed queues
Reaction plan	Named owner, containment and escalation path, reviewed monthly	Project sponsor	Plan audit every quarter

Priorities if the budget is tight

1. Extend the error-proofing to the highest-frequency remaining modes. It is the only control that is independent of turnover and attention. Twelve percent coverage is not an improvement, it is a demonstration. Target the modes that make up the next 50% of the defect mass.
2. Fix the reaction plan. A plan reading investigate and correct is not a control. Naming an owner, a threshold and a containment action costs nothing and prevents the tampering behavior already observed.
3. Attach training to the hiring pipeline, not to a one-time event. With 40% turnover, qualification before independent work is the only durable form of training control.
4. Retire the manual audit deliberately. Keep a reduced sampling audit as a verification check on the system, with a stated purpose and sample size, and hand ownership to the process owner. Do not keep it as the thing that produces the result.

Close the project only when these elements have run for two to three months without project resources and the metric has held.