In my previous role, I worked on something similar to this. We built a model to predict which claims(CMS 1500) would be denied by the carrier(Insurance Provider) before they were submitted and allowing the hospital staff to adjust the Diagnosis/Procedure codes and validate the combination. The client had 20% denials, our model brought that denial rate down to 10%.

We wanted to compare the performance of our model with commercial third party offerings and we enrolled into an early beta for a similar offering from ChangeHealthcare one of the big players in the healthcare industry, their model was wildly off(<50% accuracy), we thought we were missing something and went back and forth with their team, turns out they didnt know what they were doing, they just slapped AI on the API and tried to see if anyone would actually use it. That model never saw the light of day. After studying the data model and seeing the number of variables involved I would be skeptical of prediction models retaled to insurance in the healthcare industry.