The democratization of artificial intelligence has had a dramatic impact on how businesses approach identity risk. Prior to its introduction, there were already an influx of data points streaming into digital businesses, from site behavior, to device IDs, to digital signatures like IP addresses. Traditional rules-based approaches to onboarding customers were overwhelmed by new access to data. Artificial intelligence, then, through the adoption of machine learning, became a needed tool. Compared to rules-based systems, machine learning offers a way to take advantage of all known data points available on a customer transaction in order to hone the signal, reduce the noise, and provide a better customer experience.
Machine learning has also afforded a shift to focus on good customers. Fraud can be isolated through binary signals with low recall and high precision that are effective in rules, but leave the majority of customers in an unactionable “not fraud” bucket. This has two consequences: first, it leaves no ability to treat your best customers better, and second, it forces many good customers through unnecessary friction due to false positives, triggered by blunt force rules.
A rules-based approach to customer decisioning reduces to a series of if/then statements. Often implementation is from a binary data point where, if true, X action is taken, and if false, Y action is taken. Teams can buy or build platforms that combine rules using multiple data points, or move from binary data to field arrays. The most complex rules systems leverage scorecards where any number of criteria might be met order to trigger a specific action. At the end of the day, however, these systems are only leveraging a small portion of the data available—data with linear combinations based on historical analysis to come to a decision point. The best decisions made with these systems only decide whether or not something is fraud, however.
The knowledge that something is not fraud does not infer no risk at all. While fraud risk can be flagged with blunt signals like a risky IP address or new email, to confirm the inverse requires a more holistic understanding of all data available. Not only do individual factors need to be true or false, validity needs to be measured in relationship to the validity of multiple other data points. Machine learning understands the linkages between all data points, and all the permutations of those links in real-time. This provides a holistic understanding of an identity, creating profiles of both fraud and customers. Decisions can then be made across the spectrum of risk, both positive and negative, creating opportunities for businesses to treat good customers like their best customers while still mitigating fraud.
Kurt Weiss is the Director of Strategic Accounts at Ekata.