Insurers set sights on AI in underwriting
AI is familiar territory in claims. The next push is underwriting, prevention and sales - areas where better decisions move the P&L. A new Sollers report signals the shift, backed by a telling stat: 82% of firms say they're using large language models at work, but only 30% have AI in underwriting workflows and just 15% are using it for contract analysis.
The interest is there. The bottleneck is architecture, data plumbing and governance. Without those, pilots stall and value never shows up in the numbers.
Why progress feels slow
- Disconnected systems and messy data make ingestion and feature creation hard.
- Underwriting decisions need traceability. Black-box outputs without audit trails won't pass reviews.
- Models aren't the product - the workflow is. If human-in-the-loop and escalation paths aren't designed, adoption lags.
As Piotr Kondratowicz, business architect at Sollers, put it: "Insurers that fail to establish governance structures to support AI transformation risk falling behind in a market that is moving fast toward AI, automation and predictive decision-making." He expects expansion into underwriting, pricing and customer-facing services over the next three years.
High-value underwriting use cases
- Submission intake and triage: auto-extract from emails, ACORDs and attachments; route by appetite, value and complexity.
- Pre-fill and risk summaries: pull third-party data, generate risk overviews and highlight missing info.
- Portfolio-aware pricing support: suggest factors, surface comparable risks and show historical outcomes.
- Coverage and clause review: flag exclusions, conflicts and endorsements; compare against playbooks.
- Referral management: auto-justify referrals with evidence; capture underwriter decisions for continuous learning.
Minimal architecture that actually ships
- Data layer: reliable ingestion (docs, emails, APIs), a basic feature store and lineage.
- Model layer: mix of predictive models and LLMs with prompt libraries, templates and guardrails.
- Workflow layer: APIs into policy admin/CRM, queues, human-in-the-loop, and clear override rules.
- Observability: monitoring for drift, bias, latency and business KPIs; replay capabilities.
- Security and privacy: PHI/PII controls, redaction, access policies and encrypted logs.
Governance that clears audits
- Defined use-case inventory with risk tiers and approval checkpoints.
- Model risk management: validation, challenger setups, and periodic reviews.
- Explainability thresholds by decision type; documented playbooks for overrides.
- Retention, consent and vendor assessment policies that match your jurisdictions.
Useful references: the NIST AI Risk Management Framework and the NAIC principles on AI.
90-day plan for underwriting leads
- Pick one product and one decision (e.g., small commercial submission triage). Define acceptance criteria and guardrails.
- Stand up document intelligence, a simple feature store and a model gateway; integrate with intake and referral queues.
- Ship a limited-scope workflow with clear human overrides and full logging. Measure baseline vs. post-launch.
- Expand features and decisions only after hitting targets on accuracy, cycle time and loss ratio impact.
Metrics that matter
- Quote turnaround time and touch count per quote.
- Bind rate by segment and submission quality.
- Underwriter capacity (submissions reviewed per FTE).
- Loss ratio lift vs. control and leakage reduction.
- Referral volume, approval time and reasons distribution.
Customer-facing upside
- Smarter product recommendations and clearer coverage explanations.
- Digital services that reduce risk (alerts, checklists, benchmarks) tied to pricing incentives.
- Faster endorsements and renewals with pre-filled changes and clause checks.
The window is open. Firms that get their data, workflow and governance in order will move from pilots to production and see underwriting results improve. Those that wait for a perfect plan will watch others set the standard.
If your teams need hands-on training to build these workflows, see our curated AI courses by job for insurance-focused skills.
Your membership also unlocks:
