ELECTRONIC ACCOUNT AUDITOR

CONTEXT

At one health insurer, a considerable number of accounts sampled by the audit had problems and inconsistencies. The audit needed tools that would help identify issues by increasing the account recovery rate.

 

SOLUTION
Structured data were collected from accounts, beneficiaries and several other relevant variables. Combined anomaly detection and reinforcement learning (Multi-armed bandit) models were applied to the data, which selected the best models according to their results for each type of account.

 

RESULTS
The recovery of anomalous accounts by the audit area increased 2.4 times.