AIG says Claude cut insurance submission review time by more than 90%
A new Anthropic case study describes faster risk summaries and quoting, with results dated March 31, 2026. Claims-service work remains in early pilots.
Loading page…
A new Anthropic case study describes faster risk summaries and quoting, with results dated March 31, 2026. Claims-service work remains in early pilots.
Listen to this story
AIG’s Claude workflows are intended to remove document preparation as a bottleneck in underwriting: where deployed, they cut submission review time by more than 90% and let staff review every applicable submission. The rollout covered eight business lines by March 31, 2026, but the figures are AIG-reported results published by Anthropic, not an independent evaluation. The practical test is whether broader review capacity translates into more viable quotes; AIG reports gains in one property business, while its claims results remain early-pilot findings.
AIG began developing Underwriting by AIG Assist in 2023, initially using Claude 2.0 to turn incoming materials into risk summaries with audit trails.
At AIG Lexington Middle Market Property, underwriters reviewed more than 30% more submissions and quoted them 55% faster as of March 31, 2026.
Claims pilots reportedly shortened client response times from days to hours and coverage reviews from hours to minutes; these are not claim-settlement times.
AIG’s underwriters can start reviewing an insurance submission with an AI-generated risk summary instead of spending weeks gathering information first. A new Anthropic case study says the insurer’s Claude-based workflows cut submission review time by more than 90% where deployed, while allowing staff to review every applicable submission—not just those they previously had time to examine.
The disclosure describes a deployment that had reached eight AIG business lines as of March 31, 2026, rather than a rollout beginning now. Its performance figures are AIG-reported results published by the company’s AI supplier, not an independent evaluation. The central change is how much preparation happens before an underwriter begins assessing a submission.
AIG began working on the problem in 2023. Information arrived from insurance brokers, clients and third parties in many formats, leaving employees with substantial manual work before they could assess risk. The insurer said it lacked the resources to review every applicable submission and viewed that constraint as a source of missed business opportunities.
Working with Anthropic, it built Underwriting by AIG Assist, initially using Claude 2.0. The workflow processes and analyzes incoming submission data, then produces a curated risk summary with a full audit trail. That changes the underwriter’s starting point: instead of gathering information for weeks before beginning a summary, the employee receives one at the outset of the review.
AIG designed the system to process every submission received in the business lines where it operates. The company says the deployed workflows now enable review of 100% of applicable submissions. That figure concerns the scope of review within those deployments; it is not a claim that every AIG business uses the system.
The case study gives a closer view of AIG Lexington’s Middle Market Property business. As of March 31, its underwriters were reviewing more than 30% more submissions and quoting them 55% faster, according to the account. These are separate measures from the reduction in submission review time: they describe the volume handled and the speed of producing a quote.
For AIG, that addresses the bottleneck it originally identified: submissions that staff could not get to represented potential business left unexamined. The reported gains therefore concern both faster preparation and greater capacity to consider incoming business, rather than simply making the same number of document reviews take less time.
AIG has also extended Claude into its claims organization, though the results described are from early pilots. Claims by AIG Assist reduced the time to respond to clients after a claim was filed from days to hours, the company says. Coverage reviews fell from hours to minutes. Those measures describe response and review times, not the time required to settle a claim.
Two further uses remain exploratory rather than deployed capabilities:
One scenario would assign separate agents to extract submission data, check risks against underwriting guidelines and compare pricing with portfolio targets. A fourth agent would bring their work together. In that proposed setup, underwriters would oversee every decision—not hand the insurance decision to the agents.
Loading discussion...
Join the conversation
Explain whether broader consideration would change your comfort with AI’s role.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.