The Claims Unit Kaizen: A Third Less Cycle Time, Twice the Rework
A regional insurer's auto claims unit handles 3,200 claims a month. After a three-week kaizen that standardized intake, the unit reports average cycle time down from 9.4 days to 6.1 and on-time completion at 94%. Three things complicate the claim. The cycle-time metric now starts when a claim reaches an adjuster, not at first notice of loss. A 'pending documentation' queue, holding 21% of open claims, is excluded from the calculation. Reopened claims have risen from 6% to 14%, and adjusters say the new intake checklist can be cleared without the police report, so incomplete files are pushed forward. Headcount is unchanged. The unit manager is preparing to present these results to the executive committee and to replicate the intake design in commercial lines next month.
1. Is the improvement real? What evidence would settle it?
Read the worked answer
Not yet demonstrated. Three measurement decisions and one quality signal point the other way.
- Boundary shift. Moving the clock start from first notice of loss to adjuster assignment removes the intake interval - precisely where the checklist now defers work. Recompute end-to-end from first notice of loss.
- Excluded queue. A 21% pending-documentation population outside the calculation is not a rounding issue; it is roughly a fifth of the work. Cycle time must include it, and the queue's age profile should be published.
- Rework. Reopened claims rising from 6% to 14% means the unit is closing files and reopening them. That work consumes adjuster capacity and appears in the customer's experience, not in the reported average.
Evidence that would settle it: end-to-end cycle time and 90th-percentile cycle time from first notice of loss, including pending documentation and rework loops, measured on the same cohort basis as the pre-kaizen baseline. Also compare work-in-process and throughput: with headcount and intake volume unchanged, a one-third reduction in cycle time implies either a large WIP reduction or work moving rather than disappearing. If WIP is flat, the reported gain is a definitional artifact.
Best reading: adjuster-side touch time may genuinely have improved, while customer-experienced cycle time and quality are unproven or worse.
2. The unit manager wants to replicate the intake design in commercial lines next month. What do you advise?
Read the worked answer
Do not replicate yet. The design has a known defect and no validated measure.
Scale follows proof, not enthusiasm. Two conditions are unmet: the intake checklist can be completed without the police report, which pushes defect forward into adjuster work and drives the rework increase; and the measurement set was changed at the same time as the process, so there is no trustworthy baseline.
Sequence:
- Fix the defect. Make the checklist a completeness gate with an explicit exception path for reports not yet available, so a file cannot advance as 'complete' while missing a document it will certainly need.
- Define measures before the next pilot. End-to-end cycle time, rework rate, first-contact resolution, and pending-documentation age, with definitions locked and change-controlled.
- Pilot in one commercial team with a comparison group of similar claim types, reviewed at 8-12 weeks.
Also brief the executive committee on the corrected read before any replication request reaches them. If the unit manager presents the 9.4-to-6.1 story unqualified and it is later contradicted, the credibility loss will make the next genuine improvement harder to fund - the political cost of overselling is part of the operational risk.
3. How do you keep the unit from optimizing the reported metric at the expense of the customer?
Read the worked answer
Pair every efficiency measure with a balancing measure and control the definitions centrally.
Specifics:
- Paired metrics. Cycle time is reported alongside rework rate, repeat-contact rate, and complaints; on-time completion is reported alongside the age of the excluded queue. A gain on one with deterioration on its pair requires explanation, not celebration.
- One definition owner. The measurement definitions - start point, stop point, exclusions, percentile basis - sit with a data governance function, with changes versioned and approved. Most metric gaming is definition drift, not deliberate fraud.
- Visibility of the excluded population. Publish the size and age profile of any queue outside the headline metric. Queues that are invisible in reporting become the place work is parked.
- Audit the tail. Sample 30 files a month and hand-time them end to end; compare to the system number. Divergence is the early warning.
- Review the incentive. If the unit manager's scorecard rewards average cycle time only, the behavior being observed is rational. Change the scorecard before blaming the manager.
None of this works if executives reward the single favorable number. The balancing measures must be presented at the same level and with the same prominence, or they will be treated as excuses.