Same underwriting headcount, 3.1x submission throughput.

Submissions came in as PDFs, ACORD forms, and broker emails. Underwriters spent two days a week just getting to the point where they could underwrite. We rebuilt the front of the funnel and gave them that time back.

What was actually broken.

Submission intake was an artisanal process: an underwriter would open a broker email, pull the ACORD attachment, find the loss runs in a separate attachment, copy the entity name into the policy admin system, run the loss-run through an OCR tool that fell over on 30% of documents, then start the actual underwriting. The team of 14 underwriters was capacity-constrained — submissions sat in the queue for 11 days on average before a first touch, and the broker channel was complaining. The CUO had a hiring requisition open for two years that could not be filled in the local market.

What we did, and what we deliberately did not do.

We built a submission-ingestion agent that read the broker email, the ACORD form (structured), the loss runs (OCR with a vision fallback for handwritten endorsements), and the supporting documents — then produced a normalized submission record with every field populated, every document linked, and every prior-claim coded against the carrier's claim taxonomy. The underwriter opened a fully-prepared file. Anything ambiguous was flagged with a confidence score and the source passage. We did NOT change the underwriting decision logic. We changed the work that came before the decision.

“Two years of trying to hire fourteen more underwriters in a market that did not have them. FirmAdapt's stack gave my existing fourteen people three times the throughput. The broker channel stopped complaining. The hiring requisition is closed.”

What the numbers did, twelve months in.