AI in insurance: assist now, decide later
At the Bermuda Risk Summit 2026, industry leaders were clear: artificial intelligence and machine learning are here to support people, not replace them. The panel agreed that AI's best use today is freeing teams to focus on client relationships, product innovation, and closing the protection gap.
Natanje Holt of DXC Technology framed AI across three stages: analyzing and gathering data, making decisions, and acting. Analysis is mature and widely adopted. Decision support has limited appetite. And "acting" is mostly basic process automation-useful, but still assistive.
Where AI is working today
Data ingestion, normalization, and triage are delivering consistent gains. Think document classification, loss-run extraction, policy data validation, and first-pass underwriting checks. These are the heavy lifts that bog people down; AI handles them well so your experts can handle the nuance.
Holt emphasized that end-to-end decisioning isn't ready for prime time. Simple actions like address changes can be automated, but high-impact calls still need human judgment.
Fraud detection is the clearest win
Andrew Johnston of Gallagher Re highlighted claims fraud detection as a standout. AI confirms flagged cases and, more importantly, surfaces signals humans may miss. The best systems keep people in the loop so models keep learning from real outcomes.
This "affirmation in the loop" approach shortens investigation time, reduces leakage, and improves referral quality for SIU teams.
Guardrails before scale
David Flett of KPMG Bermuda warned that weak governance is already causing problems for some carriers. The fix isn't flashy: risk and control frameworks, clear ownership, auditability, testing, and monitoring. If you can't explain the model, you can't stand behind the outcome.
For reference frameworks, see the NAIC's Principles on AI and NIST's AI Risk Management Framework for practical oversight guidance.
Human-first underwriting and claims
Relm Insurance COO Henry Paddison was blunt: outcomes drive tool choice. AI and automation improve speed and quality, but people moderate risk. Relm evaluates insureds' AI through four lenses-physical harm, reputational, regulatory, and financial risk-and applies the same thinking internally.
Translation: AI is a tool, not a black box making underwriting or claims decisions. Keep experts in control and use AI to reduce cycle times and noise.
Jobs and the future of work
Holt expects AI to create more roles, even if the exact titles are still forming. That means reskilling is the play-analysts who can oversee models, adjust workflows, and connect insights to action will thrive.
The work doesn't disappear; it shifts. Less swivel-chair processing. More client time and higher-quality judgment calls.
Practical steps for insurers
- Start where maturity is highest: data extraction, triage, and enrichment across underwriting and claims. See: AI for Insurance
- Prioritize fraud: combine rules, anomaly detection, and human review to improve hit rates and reduce leakage.
- Define your risk taxonomy for AI decisions: physical, reputational, regulatory, and financial impact. Map controls to each.
- Stand up governance early: model inventory, approvals, explainability standards, bias testing, monitoring, and incident playbooks. Executive accountability is non-negotiable. See: AI for Executives & Strategy
- Keep humans in the loop: set thresholds where automation stops and expert review begins. Document decision overrides and feedback.
- Pilot, measure, then scale: track cycle time, loss ratio impact, leakage reduction, and customer outcomes-not just cost savings.
- Vet vendors like critical infrastructure: data lineage, security, model update cadence, and audit rights should be contract terms.
- Upskill your teams: claims, underwriting, and operations need prompt skills, model oversight basics, and process redesign capabilities.
The headline
AI is proving its value as an assistant-fast, tireless, and good at finding patterns-while high-stakes decisions stay with people. Build the guardrails, focus on measurable outcomes, and use the time savings to close the protection gap and deepen client relationships.
Your membership also unlocks:
