Airwallex today shared data on the impact of a new generative AI tool that enhances the speed and efficacy of its “Know your Customer” (KYC) and onboarding processes. The new tool reduces ‘false positives’ by 50 percent on average, while boosting the number of customers that pass through the onboarding process without human intervention by 20 percent.1
“With generative AI, we’re honing our KYC tools to be much more accurate and context-aware,” said Jacob Dai, co-founder and chief technology officer at Airwallex. “We’re making it faster and easier for our customers to get up and running with Airwallex – often within minutes – while enhancing our ability to detect and prevent fraud on our platform. These enhancements translate directly to our customers – they can unlock the value of our platform more readily and accelerate their own global growth.”
The tool is attuned to nuances and sensitivities of various languages and cultures, a key feature for a company as globally distributed as Airwallex. “We work with businesses from incredibly diverse industries and cultural settings,” said Dai. “Having an intelligent AI co-pilot, trained on vast amounts of linguistic data, helps us navigate those intricacies more quickly and effectively.”
Addressing the KYC challenge with smarter AI tools
“KYC is that vital first step in our relationship with a customer,” said Dai. “We use a combination of intelligent technologies and human expertise to establish that new customers are who they say they are, that they’re being truthful about the nature of their business, and that their products or services are legal and in line with Airwallex’s acceptable use policies.”
Historically, Airwallex has used rules-based analytics and Natural Language Processing (NLP) to scan new customers’ websites. Those tools are very effective in identifying high-risk keywords, but also tend to generate a significant volume of false-positive alerts – which can slow down the KYC process unnecessarily.
“Generative AI helps Airwallex’s models to be much more sensitive to the context and meaning of keywords,” said Dai. “When scanning a customer’s website, the model can better distinguish between a retailer selling a ‘military-style’ jacket and a merchant selling prohibited military goods, or a ‘champagne-colored’ dress and actual champagne, or smartphone accessories versus actual smartphones being sold illegitimately.”
