1. Personal data from historical benefit claims fed into a machine learning algorithm
6. Review outcome is fed back into the algorithm
2. Algorithm searches for patterns that correlate with claims containing fraud or error
5. Claims considered correct after review are paid, those thought potentially fraudulent are investigated further
3. Model scores new claims for potential fraud and error
4. High scores are flagged for review by a human, while payment of benefit paused
1. Personal data from historical benefit claims fed into a machine learning algorithm
6. Review outcome is fed back into the algorithm
2. Algorithm searches for patterns that correlate with claims containing fraud or error
3. Model scores new claims for potential fraud and error
4. High scores are flagged for review by a human, while payment of benefit paused
5. Claims considered correct after review are paid, those thought potentially fraudulent are investigated further