By Anand Shirur, Senior Vice President, Klear.ai

CURRENT ESTIMATES ON LOSSES due to Fraud Abuse and Waste range from $60 to $80 billion a year. You may see around 3 to 5% of your claims having tell-tale signs of fraud, which could add up to over 15% of your total incurred. Traditionally the claim teams have relied on the acumen and diligence of claim examiners to spot suspicious claims. With experience, the examiners develop a ‘hunch’ for suspicious behavior. However, this strategy has its limitations in modern-day claims adjudication. High caseloads do not afford the luxury of focusing on one claim at a time.

ENTER AI Recently, Artificial Intelligence and Machine Learning technologies are proving to be very effective in flagging potentially fraudulent claims and transactions. They can draw upon the multi-dimensional relationships hidden in your structured data and adjuster notes and use advanced statistical models to make accurate predictions. Self-learning algorithms observe its own prediction till the claim plays out to its logical closure. And based on hits and misses, the model ‘retunes’ itself. Result – Accuracy of predictions improves with time.

The underlying models, predicting fraud, evolve in line with organizational maturity. One may start with encoding the ‘red flags’ into ‘business rules. Over 50 red flags have been observed, like ‘No witness to injury’ or a ‘termination notice recently served to the claimant’. However, the business rules need to be updated periodically, to keep up with newer and bolder deceptive fraudsters.

Higher on the maturity curve is the ‘Anomaly Detection technique, which statistically isolates an ‘Outlier’ behavior. Identifying ‘Overbilling’ by a provider could be a good use case, where the system compares the price of a CPT against an average price or a fair market price of a service. The technique isolates a line item if the price is over (say 80%) of the average price. If you have sizeable historic claims data, then you may look at more advanced ‘Predictive Modeling. Using advanced statistical models, the predictive model accurately forecasts potentially fraudulent claims. The models give you the level of confidence as a percentage with every prediction and notify the ‘top drivers’ of the fraud score. Put together, the insight points the adjuster towards a line of investigation that could be pursued to confirm the prediction.

New-age cutting-edge solutions have AI ‘baked’ into the transactional systems. The tight integration of intelligence and transaction ensures that ‘insights drive action’. Fraud predictive insights can block vendor payments before it is too late. Or let an adjuster deny or delay a claim acceptance till a proper investigation is concluded. With these systems, you have the liberty to categorically define the manner and extent to which insights shall drive your actions and transactions.

Anand Shirur leads the ‘Product Management’ function for Klear.ai. You can reach him at anand.shirur@klearai.com