While discussing approaches of identifying process discrepancies and detecting fraud, a key question that comes up is around the right approach for proactive fraud control. While there is no disputing on the value of leveraging data to make decisions, there are a wide variety of philosophies around this. On one hand, risk and compliance teams have created a bank of excel-based routines and heuristics that could be used for alerting. A cottage industry of products have sprung up targeting specific areas like duplicate payments and vendor frauds but have missed out areas that need more data correlation such as IP protection and FCPA detection.
Our approach behind ApolloTM is to leverage the power of big data and artificial intelligence to address the challenge. We will also address sceptics around whether we are using a sledgehammer to crack an egg.
In my last article, I spoke around how we need to be able to correlate between multiple data sources - many of which would be messy and difficult to handle. The ability to manage unstructured data at speed will call for a different kind of technology and fortunately, with ApolloTM, we now have the ability to use the same technology that powers leading companies who have built mega-businesses around insights from data. I am referring to the cloud platform for cheap and flexible infrastructure, the data handling capabilities in open source packages like Hadoop and the ability to develop advanced algorithms in packages like R and Python.
Let me talk a bit more on this. In context, one installation would process 100mn records every day and over 200 rules on scheduled basis. A traditional installation would have taken at least three months to provision and extremely expensive licensing. In addition, I am not even sure if we could get some of our current rules to operate on that platform. However, on the big data platform, we were able to get the platform executed in a week and at no software purchase cost.
Once this initial element of correlation at scale and speed is done, we still face the challenge of ensuring a good coverage of all red flags and accuracy (precision and recall in statistical terms). And to make investigations efficient, we need to ensure that we keep incorporating the learning's to tighten the rules. Doing this manually is tedious and would be overwhelming and this is where machine learning rules enable a feedback loop from actual results from investigations. The nature of the rules framed by the algorithms could be extremely complex incorporating over a 100 parameters with complex logic elements that cannot be manually crafted.
Another key challenge is in detecting new frauds or unusual policy violations - that are more often rare, unique and hence not pre-defined. In such cases, the way to detect is by identifying abnormalities based on transactional or individual behavior that is out of pattern. This could either be on a stand-alone basis, compared to past data or on a peer comparison. I guess either ways, we can do with some help to pick out the needles from the hay-stack!
Wipro has built the ApolloTM 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 ApolloTM webpage.