Underwriting fairness in Pakistani insurance

Three ways to see it

1. 1

   Insurance fairness has features insurance regulators care about more than banking regulators do. Geographic discrimination: charging higher prices in a region not because of actuarial signal but because of correlated factors. Health-data sensitivity: using genetic, mental health, or chronic disease data in ways the customer would not expect. Gender pricing: in life and motor insurance, gender pricing is permitted in Pakistan but tightly watched in Europe. Proxy discrimination: postcode, ethnicity-correlated names, religion-correlated names. Each is a regulator litmus test.

2. 2

   Way one to audit Pakistani underwriting: pull six months of issued policies and computed premiums. Compute the loss ratio expected under the AI's pricing and compare to the realised loss ratio across slices: district, gender, age band, occupation, declared income. Where the AI is over-pricing a slice (loss ratio < expected target), that slice is being charged unfairly. Where the AI is under-pricing (loss ratio > target), the insurer is bleeding. Both findings deserve action.

3. 3

   Way two: health-data discipline. If your AI uses any health data, declare what data was used, for what purpose, with what consent. Customers, when they discover their old hospital records were used in pricing, sometimes consent retrospectively, often complain. The retrospective complaint is the dangerous one. Pakistani insurers should adopt the principle of explicit informed consent for any health data feature before SECP requires it.

The why-tree

Try this with Claude

AI-edge prompt to try: 'You are an EIOPA-aligned actuarial fairness reviewer. For a Pakistani health insurance pricing model that uses age, gender, district, declared income, and recent hospital visits, list five fairness concerns ranked by regulator priority. For each, propose a test and a remediation.' Use as a draft scope.