How financial services operators are dialling up conversational AI to catch out fraudsters

Organisations are using new technology to analyse the voices of those posing as customers in real time while reducing false positives
Illustration of conversational AI being used to fight fraud

Great Britain is the fraud capital of the world, according to a Daily Mail investigation published in June. The study calculated that 40 million adults have been targeted by scammers this year. In April, a reported £700m was lost to fraud, compared to an average of £200m per month in 2021. As well as using convincing ruses, scammers are increasingly sophisticated cybercriminals.

If the UK does go into recession, as predicted, then the level of attacks is likely to increase even further. Jon Holden is head of security at digital-first bank Atom. “Any economic and supply-chain pressure has always had an impact and motivated more fraud,” he says. He suggests that the “classic fraud triangle” of pressure, opportunity and rationalisation comes into play. 

Financial service operators are investing in nascent fraud-prevention technologies such as conversational AI and other biometric solutions to reduce fraud. “Conversational AI is being used across the industry to recognise patterns in conversations, with agents or via chatbots, that may indicate social engineering-type conversations, to shut them down in real time,” continues Holden. “Any later than real time and the impact of such AI can be deadened as the action comes too late. Linking this to segmentation models that identify the most vulnerable customers can help get action to those that need it fastest and help with target prevention activity too.”

This last point is crucial because educating customers about swindlers is not straightforward. “Unfortunately, there will always be vulnerable people being scammed,” Holden says. “The banks are doing a lot of work to identify and protect vulnerable customers, but clever social engineering, often over a long period, will always create more victims of romance scams, investment scams, or purchase scams when victims send money for goods never received.”

How AI can help fight fraud

AI is a critical tool to fight fraud. Not only does it reduce the possibility of human error but it raises the flag quickly, which enables faster, smarter interventions. Additionally, it provides “far better insight of the cyber ecosystem”, adds Holden, “almost at the point of predictive detection, which helps with both threat decisioning and threat hunting”. 

Jason Costain is head of fraud prevention at NatWest, which serves 19 million customers across its banking and financial services brands. He agrees it is vital for conversational AI to join the chat. Because the call centre is an important customer service channel and a prime target for fraudulent activity – both from lone-wolf attackers and organised crime networks – he resolved to establish more effective security mechanisms while delivering a fast, smooth experience for genuine customers. 

In late 2020, NatWest opted for a speech recognition solution by Nuance, a company which Microsoft recently acquired. It screens every incoming call and compares voice characteristics – including pitch, cadence, and accent – to a digital library of voices associated with fraud against the bank. The software immediately flags suspicious calls and alerts the call centre agent about potential fraud attempts.

Since our initial implementation of AI three years ago, the improvements to alert quality have been incredible

Before the end of the first year of deploying the Nuance Gatekeeper system, NatWest had screened 17 million incoming calls. Of those, 23,000 led to alerts and the bank found that around one in every 3,500 calls is a fraud attempt. As well as a library of ‘bad’ voices, NatWest agents now have a safe list of genuine customer voices that can be used for rapid authentication without customers needing to recall passwords and other identifying information. That knowledge enables the bank to identify and disrupt organised crime activities to protect its customers and assist law enforcement.

“We’re using voice-biometric technology to build a clear picture of our customers’ voices and what criminal voices sound like,” Costain says. “We can detect when we get a fraudulent voice coming in across our network as soon as it happens. Using a combination of biometric and behavioural data, we now have far greater confidence that we are speaking to our genuine customers and keeping them safe.”

He estimates the return on investment from the tool is more than 300%. “As payback from technology deployment, it’s been impressive. But it’s not just about stopping financial loss; it’s about disrupting criminals.” For instance, NatWest identified a prolific fraudster connected to suspect logins on 1,500 bank accounts, and an arrest followed.

“For trusted organisations like banks, where data security is everything, the identification of the future is all about layers of security: your biometrics, the devices you use, and understanding your normal pattern of behaviour,” adds Costain. “At NatWest, we are already there, and our customers are protected by it.”

Benefits of investing in conversational AI

There are other benefits to be gained by investing in conversational AI solutions. Dr Hassaan Khan is head of the School of Digital Finance at Arden University. He points to a recent survey that indicates almost 90% of the banking sector’s interactions will be automated by 2023. “To stay competitive, organisations must rethink their strategies for improved customer experience. Banks are cognisant that conversational AI can help them be prepared and meet their customers’ rising demands and expectations,” he says.

This observation chimes with Livia Benisty. She is the global head of anti-money laundering at Banking Circle, the B2B bank relied on by Stripe, Paysafe, Shopify and other big businesses, responsible for settling approximately 6% of the world’s ecommerce payments. “With AML fines rocketing – the Financial Conduct Authority dished out a record $672 million (£559m) in 2021 – it’s clear that transaction monitoring cannot cope in its current state,” Benisty says. “That’s why adopting AI and machine learning is vital for overturning criminal activity.”

She argues, however, that many in the financial services industry are reluctant to invest in the newest AML solutions for fear of being reprimanded by regulators. “If you’re a bank, you come under a lot of scrutiny and there’s been resistance to using AI like ours,” she says. “AI is seen as unproven and risky to use but the opposite is true. Since our initial implementation of AI three years ago, the improvements to alert quality have been incredible. AI alleviates admin-heavy processes, enhancing detection by increasing rules precision and highlighting red flags the naked human eye could never spot.”

Even regulators would be impressed by the results revealed by Banking Circle’s head of AML. More than 600 bank accounts have been closed or escalated to the compliance department, thanks to AI-related findings. Further, the solution “dramatically reduces” the so-called false positive alerts. “It’s well known the industry can see rates of a staggering 99%,” adds Benisty. “In highlighting fewer non-risky payments, fewer false positives are generated, ultimately meaning more time to investigate suspicious payments.”

As the economy weakens, and criminals grow stronger, financial services operators would be wise to dial up their conversational AI capabilities to improve customer experience today and pave the way to a password-less tomorrow.