Transforming Banking with AI: Hyper Personalization Catches More Fraud with Less Friction

Imagine this: You’ve scored the world’s most popular concert tickets online, despite fierce competition from millions of other fans. The tickets of your dreams are in your cart and the clock is ticking -- but when checking out, your Company One credit card is declined, for no apparent reason.

Still, you scramble and manage to pay for the tickets with your Company Two card, thrilled at the outcome but furious that the first card was declined. By incorrectly flagging your transaction as fraudulent, Company One created unnecessary friction. Company One stressed you out, badly, and you’re not sure if you’ll ever use that card again.

Even AI Needs Help

The vastly different customer experiences in this true story come down to one thing: Both card companies use AI-powered fraud detection models, but only Company Two’s model is enhanced with hyper personalization. This advanced analytic technique can enable pinpoint accuracy in fraud detection, greatly reducing ire-inducing false positives like the one above. 

Specifically, Company One and Company Two’s AI models both can determine whether a transaction appears to be highly uncharacteristic -- which, in the case above, happens to be a first-ever ticket purchase of over $1,000. But only Company Two couples this result with AI-fueled hyper personalization, which provides insight that the transaction is, in fact, highly plausible. The result: Company Two approves the transaction as legitimate while Company One incorrectly flags it as fraudulent.

Let’s explore what hyper personalization is and how, when coupled with AI fraud detection, it can transform customer experiences in fraud decisioning.  

AI Becomes Smarter About You With Hyper Personalization

AI fraud systems have evolved from using generic population-risk statistics to make their decisions, to needing to know you by the way you transact. Today the most effective AI-powered fraud models make hyper-personalized decisions at an individualized level, incorporating transaction profiles that capture a customer’s unique transaction history -- including past recurrences such as common merchants, recurring payments, spending in merchant categories, recency of their last spend, and location corridors (regions between cities) -- to understand and anticipate future transactions.

In this way, hyper-personalized transaction profiles provide insight to override AI models’ tendency to make a generic prediction. A general transaction may score as having a high probability of being fraudulent, but hyper personalization further enriches the fraud detection model to interpret it as plausible based on the customer’s past. Over time, the model can learn to expect which risky-looking transactions are indeed legitimate. These capabilities allow the AI model to see the vast majority of transactions as safe and concentrate on the difficult transactions to determine whether they are fraudulent.

In our case above, even though the ticket buyer’s first time $1,000-plus purchase is risky, hyper personalization incorporates the customer’s history of occasionally buying other concert tickets. Their spending on online travel and entertainment categories further informs the fraud detection model that the transaction is likely legitimate.

Hyper Personalization Predicts the Future for AI

Hyper personalization is not just about an individual customer’s past transactions. It also predicts the future based on how similar customers have behaved. Collaborative profiles tokenize the purchase histories of a huge corpus of customers and then, optimally, group the purchases into a pre-defined number of archetypes. These archetypes are discovered though an unsupervised modeling process that associates the probability of a token appearing in one archetype versus another.      

Archetypes can have powerful meanings, for example quantifying purchases more likely for a business traveler, a cash-strapped new family, or a Gen Z consumer with disposable income to buy expensive concert tickets.

When an individual customer presents a new behavior, such as a first-time ticket purchase of over $1,000, a fraud detection model with hyper personalization leverages what it knows of the customer’s past purchases, as well as collaborative profiling informed by hundreds of thousands of other customers that are of a similar archetype distribution. In doing so, the AI model can determine if these similar customers have made similar types of time-pressurized, high-stress purchases in the past. The collaborative profile ascertains that the ticket buyer’s peers do, and so the fraud model determines that the transaction is likely not fraudulent.   

AI Reduces Friction

Hyper personalization that incorporates both a customer’s recurrence patterns, and collaborative profiles presents invaluable individualized insights to the neural networks within fraud detection systems, the mechanisms that score transactions based on their likelihood of being fraudulent and provide reasons why. It’s critically important to have smarter AI, and AI that focuses on the individual, to produce better decisions and less friction. Ultimately, AI-powered fraud detection removes unnecessary friction, creating extraordinary experiences for customers like my concert-going friend.