Definition
Measures the percentage of claims where the AI triage model's complexity classification (simple/moderate/complex) matched the actual handling complexity revealed during adjustment, detecting whether the model's training-era complexity definitions still reflect current operational reality.
The Measurement Gap
Claims triage models classify incoming claims by complexity to route them to appropriate handling tracks: simple claims to STP or junior adjusters, complex claims to senior staff or specialist teams. But insurance complexity is not static. New coverage forms, evolving endorsement structures, shifting legal precedent, and regulatory changes continuously reshape what makes a claim complex. A triage model trained on last year's claims may classify a cyber endorsement claim as "simple" because cyber claims were rare in the training data, when current coverage forms and litigation patterns have made them genuinely complex. Legacy metrics cannot detect this classification drift. First contact resolution measures whether claims close in one interaction, not whether the routing was appropriate. Average handling time tracks how long claims take without distinguishing correct routing from misrouting that eventually self-corrects. Claims reopening rate captures some misrouting signal but only when the error is severe enough to require reopening. Adjuster workload metrics track volume distribution without assessing whether the distribution matches actual complexity. SIU referral rate tracks fraud investigation routing but not general complexity routing accuracy.
Decision Supported:
Is the claims triage model routing claims to the right handling tracks, or has coverage evolution rendered the model's complexity classifications outdated, and which claim types need reclassification?
Formula
Plain language: For each claim, record the triage model's predicted complexity tier. After the claim is resolved, record the actual complexity tier based on handling characteristics (number of coverage questions raised, need for coverage counsel, payment adjustments, litigation involvement). Compare predicted to actual.
Notation:
TCCI = Correctly_classified_claims / Total_claims_sampled x 100
Under_triaged_rate = Claims_classified_simple_but_actually_complex / Total_claims_classified_simple x 100
Over_triaged_rate = Claims_classified_complex_but_actually_simple / Total_claims_classified_complex x 100
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Correctly_classified_claims | Claims where predicted complexity tier matched actual complexity tier | Count |
| Total_claims_sampled | All claims in the measurement sample | Count |
| Under_triaged_rate | Simple-classified claims that required complex handling | % |
| Over_triaged_rate | Complex-classified claims that were resolved with simple handling | % |
Track TCCI overall and the under/over-triaged rates separately. Under-triaging is more dangerous because claims get auto-paid with wrong coverage interpretation.
Proposed formula, pending validation pilots.