A grounded case for AI-assisted claims intake extraction
Prepared by CDW with Heartland Mutual Insurance
The problem, in Heartland Mutual Insurance's words
“Intake sits six days, mostly manual PDF reading.” — Michelle Dorsey
“The 15% the model would get wrong are the same ones my team finds hard.” — Michelle Dorsey
“We handled Q1 volume by paying overtime, not by hiring.” — Dana Reyes
What it costs
400 × 2 × $38.75 = $31,000 / day
400 claims per day × 2 avoidable days × $38.75 handling cost. At 250 working days, that is $7,750,000 per year.
Inputs confirmed by Michelle Dorsey and Dana Reyes.
The agreed constraint
“Audit trail required on any automated decision; assisted extraction is acceptable.” — Robert Osei, Compliance Officer
The pilot keeps human review on low-confidence extractions.
What this case does not yet prove
Extraction accuracy on Heartland's own forms, including handwritten adjuster notes. Whether the 15% Michelle flagged behaves as her team expects. Actual review time once fields are pre-filled. The pilot exists to answer these.
The proposed pilot
- Scope
- AI-assisted extraction from 500 anonymised claims; reuses CDW's prior document-pattern pilot spec
- Duration
- Six weeks
- Owner
- Alex Chen
- Success
- Process 500 anonymised claims with an audit trail and human review for low-confidence fields
The ask
Karen Whitfield, CFO: fund the six-week pilot and allow Alex Chen’s team to prepare 500 anonymised claims.
CDW will carry the evidence into the funded pilot and keep Heartland Mutual Insurance’s operating team in control of the next step.
Ravi Menon · CDW