AI outsourcing
Reading claims documents accurately at claim volume
A UK insurer was drowning in claims paperwork. AI extraction and classification, with human quality control, sped intake without sacrificing accuracy.
Results at a glance
- 97%
- Extraction accuracy
- -69%
- Intake turnaround
- -52%
- Cost per document
- 81%
- Straight-through rate
high-confidence fields
2.6 to 0.8 days
no human keying
The challenge
Every claim arrived as a bundle of documents in no fixed shape: forms, invoices, photographs, medical notes and correspondence. Staff keyed the important fields by hand and sorted each document by type before a claim could move. At claim volume this was slow, and manual keying introduced errors that surfaced later as rework and complaints.
The insurer had tried to speed intake by adding people, but headcount alone could not keep up with peaks after storms and other event surges. It wanted the routine reading and sorting handled automatically, with accuracy high enough to trust and a clear check where the machine was unsure.
Pain points
- Key claim fields were keyed by hand, which was slow and error-prone
- Documents arrived in mixed formats with no consistent structure
- Intake could not absorb surges after weather and other events
- Downstream rework and complaints traced back to intake errors
Our approach
- Built an extraction model to pull key fields such as policy number, claim date, amounts and claimant details from mixed document types
- Added a classification step that sorted each document into type so the right handler received the right pages
- Set a confidence threshold so high-confidence extractions flowed straight through while uncertain ones were routed to a human checker
- Ran quality control on a daily sample of straight-through cases, feeding corrections back to retrain the model and hold accuracy
Results at a glance
The results
- Extraction accuracy reached 97 percent on high-confidence fields, verified against the daily quality-control sample.
- Average intake turnaround fell from 2.6 days to 0.8 days as routine reading and sorting stopped being manual.
- Cost per document processed dropped sharply, and downstream rework fell with fewer keying errors.
The routine reading just happens now, and it is right. Our people spend their time on the claims that actually need judgement.
Common questions
A confidence threshold sends uncertain cases to a human checker, and a daily sample of straight-through cases is reviewed for quality. Corrections feed back into the model so accuracy holds over time.
These are representative engagements. Client identities are anonymised and the metrics are illustrative of the type of results this work delivers, not audited figures for a named client.
Tell us what you need to outsource
Share your requirement and model your savings. We reply within six hours with a clear next step.
Six-hour response. UK GDPR compliant. Named UK accountability.