Features | From Pivot Magazine

Artificial intelligence will change the way you bank, here’s how

Banks are betting big on AI. Can consumers rely on an algorithm to safeguard their savings? 

A Facebook IconFacebook A Twitter IconTwitter A Linkedin IconLinkedin An Email IconEmail

photo illustration of a corded debt machine, with card in slot and a digital happy face on screenMastercard is just one of many financial institutions across the world using AI to fight fraud (Bank terminal by Deposit Photo; cord by Alamy)

Every minute, Canadians make more than 9,000 credit card transactions. They swipe, insert or tap, and a few seconds later, if all goes as planned, they walk away with their coffee or new clothes.

A lot happens in those few seconds. Out of sight, a machine fetches data from the cardholder’s bank and credit card company to make sure nothing’s amiss. It knows where they are, what they’re buying and whether it’s in line with their spending habits—and the habits of millions of others who spend like them. With a Mastercard, the process may involve more than 400 data points, including the cardholder’s IP address and the brightness of their device’s screen if they’re shopping online. In the time it takes for them to put away their wallets, artificial intelligence algorithms crunch all those tiny details and decide whether to approve the purchase or spit out that dreaded alert: declined.

Since rolling out an AI fraud-detection tool called Decision Intelligence in November 2016, Mastercard has detected 30 per cent more instances of credit card fraud and nearly halved the number of false declines—legitimate transactions that get rejected. For businesses, those declines are a huge problem—merchants pay a fee whether or not a payment goes through. The financial research firm Aite Group estimates that, in 2018, false declines in the U.S. alone totalled $331 billion (U.S.). “False declines add an extra layer to a process that should be simple,” says Ajay Bhalla, Mastercard’s president of cyber and intelligence solutions. “Because of an incorrect rejection, consumers may even switch their banks.”

Inadequate protection, on the other hand, can also affect the bottom line: according to CPA Canada’s 2019 annual fraud study, two in five respondents were so concerned about fraud that they refused to use their credit cards with certain merchants.

Data is the fuel that powers AI decision-making, and banks have a lot of it

Mastercard is just one of many financial institutions across the world using AI to fight fraud, cut down on false declines, provide customer service and automate rote back-office work. According to a report by Swiss financial services company UBS, three-quarters of respondents at banks with more than $100 billion in assets already employ AI.

“This is one of the most natural areas that AI can succeed in,” says Paige Dickie, senior engagement manager at the Vector Institute, an AI research hub in Toronto. She leads a country- and industry-wide project—involving some of Canada’s largest banks, regulators, researchers, police and tech firms—to help financial institutions harness the power of AI.

Banks are a natural first mover. Data is the fuel that powers AI decision-making, and banks have a lot of it. They keep detailed records on their customers, and regulation requires them to report suspicious transactions, large transfers and new account details.

To make the most of all that data, many of Canada’s big banks have founded, bought or partnered with AI labs. In 2016, RBC established its own institute, Borealis AI. And last January, TD acquired Layer 6, a startup created by two of Vector’s founders. It’s a natural pairing: banks have the data, and AI startups have the experts who understand how to use it.

The results of such partnerships are already trickling into the market. Bank of America’s AI-powered virtual assistant, Erica, is one of many “chatbots” developed by financial institutions. It acts as a cardholder’s round-the-clock attendant, reminding them when bills are due and alerting them if a recurring expense, such as a gym membership, is higher than usual. A customer can also text Erica to check their account balance or book an in-person banking appointment.

Wells Fargo has its own chatbot, which users can talk to over Facebook Messenger, as well as a suite of AI-powered predictive banking tools. They’ll analyze past activity, such as regular expenses and deposits, to anticipate financial troubles (for example, that a customer is headed for overdraft territory), identify possible culprits (all those Uber Eats deliveries) and propose solutions (budgeting tips).

One bank’s AI application analyzed legal documents—a job that used to take 360,000 human hours—in mere seconds

Banks are also using AI to secure their customers’ accounts. “Usernames, passwords, security questions—I think those will be things of the past,” Dickie says. Instead, in the near future, they’ll rely on a combination of facial recognition, voice identification and other biometric identifiers. 

Dickie argues that, to win customers’ trust, banks will need to be transparent about how they use their data and allow people to opt out of services that rely on it. She adds that consumers will likely decide whether or not to use these services depending on how useful they are, not on how much data they have to surrender to use them. “Look at Google or Uber,” she says. “People give up their data in a heartbeat to use their products. I think the same will happen with banks.”

Artificial intelligence is reshaping banks’ back offices, too. At JPMorgan Chase, an AI application called COIN can analyze legal documents and extract key data—a job that used to take loan officers and lawyers 360,000 hours—in seconds.

That inevitably raises the question: will there be any work left for humans? According to Autonomous, a London-based financial sector research firm, 2.5 million American financial services jobs will be “exposed” to AI by 2030. Some of those jobs will inevitably disappear, but a 2018 Accenture report about AI’s impact on the banking sector makes the case that many positions will instead morph: “Using AI, people will be able to spend more time on exceptional work: the 20 per cent of non-routine tasks that drive 80 per cent of value creation.”

In other words, the robot apocalypse isn’t exactly nigh. For the time being, humans still trust humans. “There still needs to be people within organizations with analytical minds who can apply skepticism appropriately,” says Tashia Batstone, CPA Canada’s vice-president of external relations and business development. She says there’s a role for CPAs in assuring data is robust, accurate and unbiased before—and after—it runs through AI algorithms. “If we do this right, we will be the individuals who understand the data. We will understand the system.”

In a highly regulated sector like banking, Dickie agrees flesh-and-bone advisors won’t vanish any time soon. “Having humans in the loop for some time will be important, especially in high-risk situations,” she says. “The day that everyone trusts AI without a doubt, that will be a scary world.”