What "65% automatable" underwriting looks like inside an MGA
Accenture says up to 65% of underwriting hours can be automated or augmented. In a small commercial MGA, most of those hours are spent assembling evidence, not deciding.
Accenture's Underwriting rewritten surveyed 430 senior underwriting executives in 11 countries. AI adoption in underwriting is expected to rise from 14% today to 70% within three years, and Accenture cites research that up to 65% of underwriting hours are subject to automation or augmentation. The top challenges respondents named were ineffective systems (65%) and lack of information and analytics (42%).
Numbers like that are easy to quote and hard to picture. Here is what they look like in a commercial MGA, from our work building underwriting tools for one.
Where the hours go
For a small-business submission, the underwriter's decision may take minutes. Getting to the point of deciding takes much longer:
- reading the application and any attachments,
- checking the business exists and does what it says,
- looking at the premises (street imagery, building age, occupancy),
- checking the class of business against appetite,
- looking for red flags in public records and reviews,
- writing it all up in the file.
That is research and documentation, which is where language models are strongest.
What we built
An underwriting assistant with three parts:
- Retrieval over the MGA's own rules. Appetite guides, underwriting manuals and referral rules, so the assistant answers with the MGA's own words and cites the page.
- Research agents. Given a business name and address, they gather web, image and registry evidence and assemble it into a draft underwriter report, every claim linked to its source.
- A broker workstation that carries the file from quote to bind, so the research lands where the decision is made.
The underwriter still decides. The difference is that they start from an assembled file instead of an empty one.
Governance is part of the product
Insurance regulators are paying attention. As of March 2025, 24 states had adopted the NAIC Model Bulletin on insurers' use of AI. Whatever the jurisdiction, the same design choices help: the assistant recommends and documents, a licensed person decides, and every recommendation can be traced back to its evidence.
Automating 65% of the hours does not mean automating 65% of the decisions. In underwriting it mostly means never again starting from a blank file.
Sources
- Underwriting rewritten: Harnessing the power of gen AI for data-driven risk management, Accenture, August 25, 2025.
- Nearly Half of States Have Now Adopted NAIC Model Bulletin on Insurers' Use of AI, Quarles & Brady, April 2, 2025.