Start with the work around the decision
A broker submission rarely arrives as a neat, complete record. The story may be spread across an email, a proposal form, a schedule and a follow-up attachment. Before an underwriter can assess the risk, someone has to bring that information together.
That is a useful place to start a conversation about AI. The goal does not have to be autonomous underwriting. A more practical starting point is helping the team reach a clear, reviewable view of a submission with less repetitive handling.
Separate extraction from interpretation
Finding a declared value in a document is different from deciding what that value means for the risk. A good workflow makes that distinction visible.
Extracted information should remain connected to its source. If a proposed business activity comes from the broker’s email, the reviewer should be able to see that email. If two documents contain different values, the system should surface the conflict instead of quietly choosing one.
Missing information should stay missing until it is supplied or explicitly resolved. A confident-looking answer is no substitute for evidence.
Build a review point that matters
Human review works best when the reviewer can see what needs attention. Presenting a long generated summary with a single approval button can simply move the effort from reading source documents to checking generated text.
A more useful review experience brings together the key facts, unresolved questions and supporting documents. It should make it easy to correct a field, explain an exception or send the submission back for clarification.
Consider a fictional property submission where the proposal and schedule list different sums insured. The useful AI contribution is to identify both values and their sources, then flag the discrepancy. The underwriter decides how to resolve it.
Measure the quality of the handover
Time saved is worth tracking, but it should not be the only measure. Look at how often extracted information needs correction, whether important omissions are caught, and how much work remains before a reviewer can make a decision.
Start with one submission type and a clear definition of a good result. Review examples with the people who handle them every day. Use their feedback to improve the workflow before expanding its scope.
Keep the outcome in view
Our perspective is that AI should give underwriting teams more room for judgement. That means making information easier to inspect, uncertainty easier to spot and the next action easier to understand.
This is the problem space behind Quill: bringing submissions, risk information and underwriting workflows into a connected workspace. The conversation starts with how your team works and where better information could make the greatest difference.
