The corporation had already invested in extensive research to correctly classify several hundred payments as “state-defaulted” or “company-serviced.” Starting with this “training set” of unidentified accounts, An A&I principal applied a statistical algorithm for classifying payments according to the few known facts, such as the payment mode, amount and date. This “learning algorithm” was then “trained” to determine a classification rule.
The A&I algorithm correctly classified more than 80% of the payments in the training set, and achieved a success rate of over 60 percent when applied to other payments.
The algorithm estimated the classification for unidentified payments, and the probability that a payment was “state-defaulted.” Applying these percentages to the payment amount determined the amount due the state, and the balance due the company.
The A&I principal satisfied state auditors that the statistical method divided the unidentified payments with objectivity and reasonable accuracy, and helped the client recoup millions of dollars.