AI spend is up. Measurable impact is scarce.
Insurers are allocating 3-8% of 2025 IT budgets to AI initiatives. Yet fewer than 5% have published clear, measurable financial impact. The early wins are in efficiency and underwriting quality, not headcount cuts, per the Swiss Re Institute's Sigma research.
What this means for carriers
- Set ROI targets that tie to the P&L: combined ratio, loss ratio lift, expense ratio, cycle time, written premium per FTE.
- Move beyond pilots. Productionize 2-3 underwriting and claims use cases where data is ready and governance is feasible.
- Address the pricing challenge "without prior assessment": build pre-bind checks, proxy signals, and scenario-based rating to reduce blind spots.
- Put model risk management in place now: data lineage, bias testing, human review gates, and audit trails.
Market context to plan against (2025-2027)
High inflation, active industrial policy, demographic shifts, and fast AI progress are setting the tone. Despite that mix, industry profitability holds, with premium growth expected to stay steady through 2027.
Reindustrialization and public spending are lifting demand in engineering, property, and liability lines. At the same time, supply chain rewiring is creating more concentrated and correlated exposures.
Investment and ALM implications
Since 2012, government intervention has tripled and policy moves are pushing long-dated inflation and bond yields higher. That changes reinvestment math and risk budgets. Expect more focus on duration, inflation hedges, and liquidity so claims and collateral needs are met without forced sales.
Capital remains solid, with solvency ratios above 200% and strong liquidity buffers. Use that strength to fund selective transformation while keeping long-term solvency planning front and center.
Product mix is shifting with aging populations
Demand is drifting from family protection toward retirement income, long-term coverage, and health insurance. That brings new ALM requirements, longer liabilities, and different lapse and morbidity dynamics.
A practical AI roadmap for insurance leaders
- Pick four high-confidence use cases: two in underwriting (risk triage, submission ingestion, pricing support) and two in claims (fraud triage, document automation).
- Allocate budget with intent: ~70% to deployment and change management, ~30% to models and tooling.
- Define success upfront: points of combined ratio improvement, loss cost accuracy, quote speed, and straight-through processing rate.
- Build the data spine: policy, loss, external signals, and consent management. Reduce manual inputs at the source.
- Upskill cross-functional teams (underwriting, claims, actuarial, risk, IT) on prompt design, oversight, and metrics. Consider structured learning if you need a fast start: AI courses by job.
- Create guardrails: escalation paths for model disagreements, documentation, and periodic performance reviews.
Underwriting and pricing: act on partial information
The hard part is pricing new or shifting risks without full prior assessment. Tactics that help: enriched third-party data, surrogate variables with clear governance, and early-warning signals in portfolios to catch drift before it hits loss ratios.
Keep sight of correlation risk
As supply chains cluster and exposures correlate, update aggregation limits and stress tests. Model compound events across property, liability, and specialty, not in silos.
Bottom line
Spend on AI is real, but the scoreboard is thin so far. Pick use cases with measurable impact, get them into production, and keep capital discipline while the macro picture evolves. That's how you turn a line item into outcomes.
Your membership also unlocks:
