A year ago, insurance companies boasted that AI investments had emerged from fledgling pilot projects into production enhancements that increased efficiency and improved profitability. What a difference a year makes. Comments on most recent conference calls by Travelers, Chubb, Hartford, AIG, and others position last year’s deployments as this year’s competitive moats that help keep competitors at bay.
AI-fueled advances in underwriting and claims were discussed on recent conference calls. Travelers cited a digital quoting platform that now processes over a million transactions annually. It has helped the firm's agent distribution produce a new quarterly record, in part due to new underwriting capacity attributed to technology. A faster, more predictable platform is likely to attract agents to process even more business.
Across other firms:
Chubb highlighted that AI has helped accelerate its underwriting of small commercial business, historically underwritten manually because of unprofitability at scale.
Hartford’s personal lines business has experienced a revamp of its underwriting process.
AIG provided numbers on its improved underwriting, which now processes 4x submissions with a 20% improvement in the submissions that are bound.
Increased underwriting volume increases bound policies and the loss experience data collected, which can be fed back into AI models to further improve risk selection. The moat widens.
Claims efficiency is a second area of improvement AI is driving. Traveler’s highlighted that over half of claims now qualify for straight-through processing, which produces a paid claim without human interaction. Staffing has been reduced by 30%, and operations have been consolidated into two centers from four.
Hartford’s AI effort has accelerated the summarization of medical records in underwriting. The model advances with every set of medical records it summarizes by operating with improved consistency and precision, which translates into margin resilience.
The lower claims costs for insurance companies translate into lower combined ratios, which give companies a choice to price lines at more competitive rates, which attracts more volume, which produces more claims data that can be used to improve the AI model. The moat widens.
Interestingly, AI only recently appeared as a category of risk that companies must underwrite as well as deploy. Cyber, professional indemnity, and liability risk is now joined by AI risk, which Travelers mentioned is a formal underwriting consideration in cyber products. Today's straight-through processing sits just outside this specific risk.
Agentic systems that carriers anticipate developing carry AI risk with direct operational significance. A presentation by AI researcher Ellie Pavlick of Brown University explained a scenario called “Schrodinger’s Chain-of-Thought” problem that agentic systems may introduce AI risk.
As agentic models are executed with longer autonomous chains of reasoning without human reviews to check each step in the process, a problem manifests. Agentic AI makes underwriting, claims adjudication, and fraud flagging decisions that produce a visible routing chain, but the chain may not drive the answer. As a result, the actual computational path that produced the answer remains opaque and may pose serious governance issues.