From paper pile to one click
A P&C insurer was drowning in documents. Claims that could be triaged in minutes were taking days.
A P&C insurer was drowning in documents. Claims that could be triaged in minutes were taking days.
Claims arrived as PDFs, photos, and scanned forms. Adjusters manually re-typed data into the claims system and chased missing pieces. Straight-through processing was a dream; in reality, every file needed human eyes. The insurer wanted to speed up intake and triage without breaking audit trails or regulatory expectations.
We built an intake and extraction pipeline that accepts uploaded documents, extracts key fields (policy, loss type, dates, amounts, contact info) using vision and NLP models, and pushes structured data into their claims platform. Low-confidence extractions are flagged for human review; high-confidence ones can auto-create the claim and route by rules. Everything is logged and auditable. We trained on their form types and legacy data so the system speaks their product and compliance language.
Average time from submission to first touch dropped by more than half for standard claims. Adjusters spend less time on data entry and more on complex cases. The same pipeline is now being extended to FNOL and correspondence so the whole front end runs with less friction.