So how does this work? Corporate card feed and cash/card receipts are received by the system, using which the expense report gets generated on its own. Employee validates the report, post which the AI engine, once again looks for any aberrations such as - duplicate claims, policy exceptions, overstated expenses, personal expenses or fictitious claims. In addition to looking at each individual claim in question, machine learning algorithms recognize patterns in data looking at - historical claims, trends of expenses for employees in similar roles, non-conformance to Benford's law etc. to flag off suspicious claims. Other elements of the user profile, such as notice period, percentage of round value expenses, nature of claims, role, etc. are also used as inputs to the AI model. Supervisor intervention is sought only for a small set of flagged expenses - and these explanations are also incorporated into the AI engine to further refine the expense pattern for the given user. In situations where the comments given by employees to justify the claims are insufficient, supervisor inputs are sought and claims processed or rejected.
So what business value does this generate? The obvious benefit is around increased process velocity with mean cycle time for claim processing reducing from 15 days to 24 hours as a result of removing redundant checks. The solution also helps achieve enhanced risk controls with false positives dropping by close to 80%, without missing out on suspicious transactions. In addition, the solution helps organizations manage and reduce costs in employee time and back-office time as well as improve user experience through reduction in user clicks. The employees hence minimize their involvement in an exercise they would much rather avoid.
Wipro has built the Apollo™ platform for Fraud Control using Big Data Analytics that has been deployed in various use cases in multiple industry domains. For details on the platform and underlying philosophy, please visit the Apollo™ webpage.