Portfolio Concentration Risk Index (PCRI)

GMI-INS-001InsuranceIntegration-Required

Definition

Measures the degree to which AI-driven risk selection has concentrated the insurance portfolio into correlated risk segments, revealing hidden catastrophic exposure beneath a healthy aggregate loss ratio.

The Measurement Gap

AI underwriting models optimize for loss ratio by selecting risks that individually appear profitable. Over time, this optimization can quietly concentrate the portfolio into correlated risk classes, such as wood-frame residential construction in coastal zones or a narrow band of "preferred" commercial occupancy types. The aggregate loss ratio looks healthy because no individual segment is losing money. But a single correlated event (hurricane, wildfire corridor, industry-specific recession) can devastate the entire concentrated book simultaneously. Legacy metrics cannot detect this. Combined ratio measures profitability after the fact, not portfolio composition. Loss ratio tracks aggregate severity but cannot distinguish a diversified 60% loss ratio from a concentrated 60% loss ratio. Claims frequency measures how often losses occur, not whether those losses are correlated. Retention rate tracks customer persistence, not whether the retained book is diversifying or concentrating. None of these metrics operate in the dimension of portfolio correlation structure.

Decision Supported:

Is the AI underwriting model building a diversified book, or is it quietly cream-skimming into risk classes that will be devastated by a single correlated event, and should risk appetite guardrails be tightened before the concentration becomes a solvency issue?

Formula

Plain language: Measure the effective number of independent risk segments in the portfolio using a Herfindahl-Hirschman concentration index, then compare against the theoretical maximum concentration. A declining index means the AI is concentrating the book into fewer correlated segments.

Notation:

PCRI = 1 - (HHI_current / HHI_max)

Where: HHI_current = Sum of (s_i)^2

Variables:

Variable Definition Unit
s_i Share of total insured exposure (sum insured or premium) in risk segment i Dimensionless (% of total portfolio)
HHI_current Herfindahl-Hirschman Index of current portfolio across risk segments Dimensionless (0 to 1)
HHI_max Theoretical maximum concentration: entire portfolio in one segment (= 1.0) Dimensionless
Risk segments Defined by geography x construction class x occupancy type (P&C), or age band x health status x coverage tier (health/life) Classification

Companion metric, Concentration Trend: PCRI_trend = (PCRI_current - PCRI_prior_year) / PCRI_prior_year. Negative trend values indicate concentrating (deteriorating); positive values indicate diversifying (improving).

Proposed formula, pending validation pilots.

At a Glance

Unit
Index (0 to 1)
Trend Direction
Toward 1.0
Reporting Frequency
Quarterly
Category
AI Underwriting and Pricing Integrity
← Behavioral Gaming Detection Rate (BGDR)Pricing Disparate Impact Index (PDII) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment