The $30M that looked like fraud
A health system that also runs an insurance plan was paying out far more than its clinical volume could explain. Leadership assumed fraud. The SQL logs showed a date-format mismatch that made the system pay for the same procedure twice.
- Organization
- EY, for a health system with its own insurance plan
- My role
- Staff Technology Consultant, Data and Platform Analysis
- When
- 2021 to 2023
- Result
- $30M+ in overpayments identified; losses cut 23%
The problem
The client was losing money fast, and leadership assumed fraud. While I was working on speeding up their claims process, I noticed payout volume was well above what the clinical volume could explain. I needed to find where the money was going and prove the cause.
What I did
During the day I was building and running anti-fraud training. At night I dug through the SQL logs. The pattern in the raw records wasn’t fraud. Hospitals were sending claim updates with a date format the payer system didn’t recognize, so it treated each update as a brand new claim and paid for the same procedure twice.
I sized the impact at over $30 million and took it to the C-suite to get the fix funded. We built a layer that recognized the update flag so revised claims stopped paying out as new ones.
Result
Losses dropped 23%, and the C-suite sponsored scaling the fraud detection work.
What I took from it
In automated systems, an efficiency problem can hide a big financial one. The answer was in the raw data, not in the assumption everyone started with.