A P&C insurer had its claims intake process live on Pointee for eight weeks. Cases were completing, so nothing looked broken — but the median case took 2.9 days against a 1-day target and the credit bill was running 38% over the go-live forecast. Three one-click changes later, the process was hitting its target, costing $104,000 a year less to run, and the evidence behind it sold six more processes.

The situation

A six-state intake: receive, classify, triage, review, assign, close. Cases completed, but the median was 2.9 days against a 1-day target, 11% looped back through review, and the credit bill ran 38% over the go-live forecast. Opening cases one at a time explained nothing.

What the team did with Pointee

Wk 8 Read the map. Performance split time per case into machine, human wait and idle: review owned 2.0 of the 2.9 days, sitting idle on a 48-hour timer.
Wk 8 Follow the credits. Classification burned 25 of the 41 credits per case on a frontier model, and one exception arm had not carried a single case in six weeks.
Wk 9 Ask, then click. Pointee proposed three changes as one-click actions: timer 48 h → 4 h, classification down to the standard tier, retire the dead arm. Approved and published.

The results

ResultsBeforeAfterChange
Median time per case2.9 days0.9 days−69%
Idle time per case2 days0.2 days−90%
Cases within the 1-day target9%71%+62%
Rework (multiple reviews)11%4%−7%
Credits per case4124−41%
Annual run cost (40,800 cases)$251k$147k−$104k

What changed the conversation

  • The wait was visible, not guessed at. Splitting time per case into machine, human wait and idle showed that 2.0 of the 2.9 days were a timer setting — not a staffing problem.
  • Cost traced to a step. The 38% overrun was one step on an over-specified model: classification alone accounted for 25 of the 41 credits a case consumed.
  • Fixes were one click, not a project. All three changes were proposed as actions and published the same week they were approved.

Conclusion

Eight weeks after go-live the process was completing every case and still missing its target on nine cases in ten. The map showed where the days and the credits actually went; three changes, none of them a rebuild, took the median from 2.9 days to 0.9, cut the run cost by $104,000 a year.