Financial institutions would be much more effective at fighting fraud if they could pool their private transaction data to train artificial intelligence models that can find patterns that reveal criminal activity. The challenge is allowing them to avoid sharing their actual raw data. 

Insight Partners’ portfolio company Featurespace just won funding from Innovate UK and the National Science Foundation in the US to develop an artificial-intelligence system to help banks and payments services providers catch money laundering and other financial crime, while protecting data privacy. It was one of only 12 organizations to win the prize and has until late January to build its prototype.  

AI has proven effective at flagging the subtle patterns that reveal bad actors – Featurespace is one of a handful of firms that deploy machine learning models to flag fraud. However, doing so effectively across banks and borders typically requires the kind of data sharing banks are wary of, due to regulation or privacy concerns. Featurespace will use an AI technique called federated learning to build its prototype (confidential computing is another approach with similar goals).  

“This type of privacy-preserving, collaborative AI is a hard problem that no one has yet solved,” Featurespace director of innovation Dr. David Sutton told Insights Distilled, adding that the firm will productionize its prototype, if successful.  “We understand the real-world problem, which puts us into a great position to bring this into the market with real-world data and constraints.” 

Featurespace’s current customers include NatWest, HSBC, Turkey’s AKbank, and Danish Danske Bank.