Aventine Health
Cutting claims handling time from eleven days to four
Median handling time fell from 11.2 days to 3.8. No assessor roles were cut; the team absorbed a 30% volume increase without hiring.
- faster median handling
- 66%
- volume absorbed, no new hires
- 30%
- unreviewed automated denials
- 0
Problem
A 40,000-claim monthly backlog handled by 120 assessors, with cycle time drifting upward for six consecutive quarters.
Solution
Document extraction and triage in front of the existing claims system, with human review kept on every decision above a confidence threshold.
Result
Median handling time fell from 11.2 days to 3.8. No assessor roles were cut; the team absorbed a 30% volume increase without hiring.
Aventine had automated claims before. Twice. Both programmes stalled at the same point: the models performed well on the sample and badly on the tail, and nobody could say by how much because there was no baseline to compare against.
How the work ran
- Six weeks establishing a measured baseline — cycle time, rework rate and assessor effort per claim type — before any model was trained.
- A thin slice covering the three highest-volume claim types, running against live traffic in shadow mode for four weeks.
- A confidence threshold set from the shadow run, above which claims route straight through and below which they queue for an assessor with the extraction pre-filled.
- Full rollout across eleven claim types over five months, with the evaluation harness wired into the deployment pipeline.
What made the difference
The decision that mattered was not technical. It was agreeing up front that no claim would be denied without a human looking at it, which removed the objection that had killed the previous two attempts.
“Our claims backlog had defeated two previous programmes. The difference this time was that they measured the baseline properly before touching anything.”
Samuel Adeyemi
Head of Operations, Aventine Health