Review the Evidence

At Quavo, we employ an 18-point investigation process to ensure that we are taking a systematic approach to data gathering. Among the points investigated are spending patterns, denied attempts (and the reasons they were denied), geo locations, devices provisioned, and charge velocity, but the data alone cannot provide everything necessary to determine whether the dispute should be written off due to fraud or denied; it needs context. 

Add Context to your Evidence

The context surrounding the data reveals the narrative of what happened on the account.  The image becomes clearer as we add details such as card status. 

For instance, when a fraudster steals a card or acquires the card data, they typically try to drain the account. They conduct transactions as quickly as possible until they begin receiving declined attempts, which signals to the fraudster that they can no longer extract money from that account, either due to it being depleted or the card being locked. When you observe this pattern, it generally indicates “true fraud.” 

Now consider this: If someone loses their card and it’s found by someone who wouldn’t typically engage in fraud, the finder might commit a crime of opportunity by making a few unauthorized transactions. Typically, fear then overtakes them, and they abandon the card before they are caught. With this in mind, examining that pattern in relation to the stolen card suggests that it was not true fraud since there was no attempt to drain the account or maximize the benefit of having the card information, despite it being true fraud.  

These details illustrate why it is essential to consider the nuance of the scenario and provide context around the data to support pay or deny decisions.   

How do you teach AI nuance?

Nuance, as humans experience it, is based on empirical data that we have observed repeatedly. If you can quantify that empirical data, you can establish parameters around that data to teach AI a form of nuance. 

What we are currently working on is creating parameters based on the cardholder’s previous behaviors and the status of the card. This data is then processed by artificial intelligence to extract patterns and insights that would normally take significant time for a human to locate and identify. Then, this data will be presented to Fraud Specialists, allowing them to focus on the important task of interpreting this information to determine whether it is “true fraud” or “not true fraud” (friendly fraud), rather than spending their time gathering data.  

By confirming that we are employing context in our understanding of fraud investigations, we are ensuring accurate investigations. When combined with context-supported data gathered by the AI, we are also saving agents processing time, saving on disputes costs, and providing an error-free experience for cardholders.