AI In Insurance: Openness Without Trust
A new J.D. Power survey shows customers are open to AI, but they don't think it serves them first. Nearly 70% say carriers gain far more value than policyholders. Only 26% believe benefits are shared evenly. As automation spreads across underwriting, service, and claims, that gap is growing.
Generative tools like ChatGPT pushed AI into the mainstream. That awareness cuts both ways for insurers: customers will use AI, but they're wary of how it's applied. The signal is clear-adopt, but earn trust as you go.
Comfort vs. Concern: Where Customers Draw the Line
- Comfort: status checks, billing updates, and basic service tasks.
- Concern: high-impact choices. Nearly 50% feel uneasy with AI evaluating claims. Pricing triggers even stronger resistance.
- Guardrails: roughly a third want AI pricing paused until bias controls are proven. Another 30% accept it only with strict fairness and explainability. Just 15% support unrestricted use.
Acceptance Isn't Approval
People will try AI tools. They just assume those tools prioritize corporate outcomes. That's the misread. Treat "openness" as a starting point, not a green light.
What Insurers Should Build Now
1) Segment by Risk and Keep Humans in the Loop
- Automate low-stakes tasks end-to-end. Keep human review for claims decisions, pricing, and coverage determinations.
- Define escalation triggers: dollar thresholds, injury indicators, liability disputes, and documentation gaps.
- Log every override and feed learnings back into model updates.
2) Explainability at the Point of Use
- Short, plain-language reasons for decisions and price impacts.
- Clear appeal paths and response SLAs. Track overturn rates and fix root causes.
- Publish a simple model change log customers can understand.
3) Bias Controls That Stand Up to Scrutiny
- Pre-launch fairness testing, drift monitoring, and periodic third-party reviews.
- Limit proxies for protected characteristics. Validate feature importance and stability over time.
- Use a model risk framework aligned with NIST's AI RMF.
4) Data Discipline and Vendor Accountability
- Minimize attributes, document consent, set retention limits, and enforce purpose binding.
- Demand model cards, monitoring APIs, fairness reports, and audit rights in vendor contracts.
- Run red-team tests for prompt injection, synthetic identity, and data leakage.
5) Show Customer Value, Not Just Cost Savings
- Speed: time-to-first-contact and time-to-payment for AI-assisted vs. standard workflows.
- Accuracy: supplemental payment rates and appeal overturn rates.
- Fairness: periodic test results, remediation steps, and outcome trends in plain language.
- Choice: human review by default for high-impact decisions; opt-in automation for routine tasks.
Operations Checklist by Function
Underwriting
- Use AI for data prep and triage; require human sign-off for edge cases and declinations.
- Send clear adverse action notices with reason codes and next-step guidance.
Pricing
- Document features, expected lift, and validation results before rollout.
- Run backtests, stress tests, and stability checks across segments and time.
Claims
- Straight-through processing for low-severity, low-complexity claims only.
- Adjuster review for injury, liability, and high-dollar losses; capture rationale for decisions.
Compliance and Governance
- Central inventory of models, owners, risks, controls, and monitoring SLAs.
- Incident playbooks, audit trails, and transparent reporting to leadership and regulators.
The Strategic Choice
AI will change how P&C business gets done. Regulation and market pressure will intensify. Carriers that prove customer benefit and fairness now will find the next wave of changes far easier to push through later.
Upskill Your Teams
Actuaries, underwriters, product, and claims leaders need shared standards for prompts, oversight, and measurement. If you're building that capability, explore role-based programs at Complete AI Training.
Your membership also unlocks:
