Accuracy Explained - ClaimScore

25 Single point of failure criteria typically used to review claims

Since the 25 individual criteria are independent of the others, each one must allow valid claims to pass, while still catching fraud. To do so, they must lean towards conservative, making them easily bypassed by well engineered programatic fraud attacks.

Compared to combining 25 criteria in an AI scoring system

When combined, the criteria leverage each other to catch fraud with more than 2 million unique result combinations. This allows each individual criteria to be more strict, since valid claims can fail individual criteria yet still pass the overall analysis.

ClaimScore combines 65+ criteria in an AI scoring system

When the number of points exceed 65, the combination of unique criteria results is over 35 quintillion (that is 18 zeros). This level of granularity is nearly impossible to bypass, and catches even the most sophisticated programatic fraud attacks.

Precision

99.5%

The percentage of correctly identified valid claims out of the total number of valid claims (i.e. true negatives). In other words, how accurately ClaimScore identifies valid claims.

Recall

97.5%

The percentage of correctly identified fraudulent claims out of the total number of fraudulent claims (i.e. true positives). In other words, how accurately ClaimScore identifies fraudulent claims.

Overall Accuracy

98.5%

Typically, fraud detection accuracy in class action settlements is reported simply as the percentage of valid claims passing the fraud detection method. However, the number of fraudulent claims passing does not affect the reported accuracy. In that scenario, if accuracy is reported as 99.5% and the percentage of fraudulent claims getting through undetected climbs from 10% to 80%, the accuracy is still reported as 99.5%. Using this method does not actually report fraud detection accuracy, but is simply a measure of valid claims to passing.

Conversely, ClaimScore takes a two pronged approach to measuring and reporting accuracy. Not only is the percentage of valid claims being approved measured by precision, the percentage of fraudulent claims is being measured by recall as well. This allows ClaimScore to maintain a minimum benchmarked rate of approval for valid claims (99.5%+), while simultaneously pushing the percentage of fraudulent claims being caught as high as possible (97.5%+).

Accuracy based on Granularity