TD Bank’s bold bet on AI

Register now

The advent of practical artificial intelligence technology that brings smarts and efficiency to many areas of a bank — customer service, customer intelligence, marketing, fraud detection, cybersecurity, lending and compliance, to name a few — has firms scrambling to find the expertise and resources to quickly adopt this technology. After all, the banks that get this right will have a distinct competitive advantage over their peers and the ability to compete with AI-savvy fintechs.

TD Bank made an astute move in this direction last week with its acquisition of Layer 6.

Layer 6, an AI startup with 17 data scientists, is about 10 minutes from the bank’s Ontario headquarters. The Layer 6 team has built an AI prediction engine that can ingest a variety of data types — customer profiles, transaction histories, phone calls, images (e.g. photos and documents) and video — and be trained to make predictions, such as what the next best action for a customer is (e.g., offer her a mortgage). It can also be applied to detecting fraud, scoping out cybersecurity threats and underwriting loans.

“Instead of having to deal with different data types separately and fuse the results, it can all go into the same engine and all be trained to produce a particular set of results in real time,” said Layer 6 co-founder Jordan Jacobs, who is now also chief AI officer (business and strategy) at TD Bank Group.

With this acquisition, TD Bank has a solid base to build upon its efforts to harness artificial intelligence technology. It is already testing Kasisto’s AI chatbot and it financially supports AI programs at the Toronto-based Vector Institute (which was co-founded by Layer 6 founders Jacobs and Tomi Poutanen), the University of Toronto and Western University. It’s advancing its use of big data and analytics, which at many banks has stalled due to lack of resources and interest. And it’s hiring tech talent that’s notoriously hard to attract to traditional banks.

Still, it's rare for a bank to bring aboard an entire team of data scientists. But TD Bank was already working with Layer 6 on a few projects.

“We really liked the team and its capabilities,” said Michael Rhodes, group head, innovation, technology and shared services, at TD Bank Group. “These guys are world-renowned, world-class in what they do. They won a global award for being the best recommendation engine. As a reference point, Alibaba won the year before.”

What AI will do for TD Bank

Like many banks, TD Bank has sought a way to enrich its data analytics efforts.

“We’ve invested a lot in our data infrastructure as just about every bank has,” said Rhodes. “We have a good toolset for traditional modeling, but as the proliferation of data and computing horsepower has increased, it’s given rise to the need for a new tools. We know what traditional modeling can do for you and we know the power AI can bring on top of that. So we’re quite bullish on what this can mean for us.”

TD Bank plans to use Layer 6’s AI engine for marketing, customer service, fraud detection, cybersecurity, and underwriting, with possibilities for more use cases later.

Rhodes hopes AI will help recreate the personal bonds that branch bankers used to create with customers who frequently stopped in.

“In the 1970s, I used to visit the National bank of Washington Branch with my mother and father,” Rhodes said. “They had a small business in the community. The branch manager knew me, always gave me a lollipop, knew my family and my parents, the age of the kids, where they were in school, the business.”

Today, customers use mobile and online banking, ATMs and phones, and occasionally visit branches. The bank also collects customers’ credit bureau reports, transactions and loan applications. All of these touchpoints create data exhaust that can be mined for ideas about what is relevant and personal to customers.

“In order to extract great insight out of all that data, you need machine learning as a tool,” Rhodes said. “You can’t just rely on someone walking into a store any more. You need to have a much deeper understanding by leveraging all your interaction points.”

Such knowledge can produce more personalized digital experiences.

“Today, most financial institutions will have a relatively static representation in their mobile and digital properties with respect to what their screens are and what the flows look like,” Rhodes said. “Try to imagine a world where those interactions or journeys can vary based on what we know about you. The way to get there is by using AI.”

Better understanding of customers can also help the bank create that covetednext best action” — a product recommendation, a piece of advice, or some other interaction that moves the customer relationship forward.

“What should we be talking to this customer about? That’s very valuable,” Rhodes said.

Cybersecurity, fraud detection and lending

Several banks have already begun using AI to winnow the thousands of false positives that are spun out of cybersecurity and fraud detection systems every day. AI can help identify previously unseen anomalies indicating nefarious behavior that humans can’t detect. It can also conduct superfast research on transactions that show they’re not criminal behavior.

“If you have a customer whose transaction behavior looks like it fits a certain trend, then clearly AI is a great tool to help identify and surface that as something that deserves a recheck,” Rhodes said.

Many nonbank lenders use AI to make quick loan decisions, even where there’s a thin or no credit bureau file for a potential borrower. Banks are gingerly tiptoeing into this area, concerned about complying with fair lending and disparate impact rules. BankMobile recently began using Upstart’s AI lending software to make loans to recent graduates, for instance.

“Underwriting is a place I know a lot of people will look at AI, ourselves included, to understand the adjudication process, particularly for thin files, when the FICO score and job history may be thin, and bring data to make an informed decision to make sure you make the right loan to the right customers,” Rhodes said. “That’s good for them and good for the bank.”

Becoming bankers

Jacobs never anticipated becoming a banker, but he said his first week has gone well. “We’re retaining our integrity as a group and a brand.”

Will Layer 6 remain independent or become integrated into the bank’s world?

“The words ‘independent' and ‘bank’ don’t often fit together well,” Rhodes acknowledged.

Though Layer 6 employees are now TD Bank employees, their office will continue to function “more like an earlier-stage organization than a 200-year-old bank,” Rhodes said. “They will maintain their brand. They’re great AI scientists and we want them to do their work.”

Jacobs agreed. “We’re certainly going to be a deeply integrated part of the bank, but we will be able to retain our brand which is great for recruiting, and we will work with people at the bank to identify the things that are priorities,” he said. “There’s a bit of a funnel that will come to us so we’re not just hit by 1,000 different people at the same time.”

Often fintech startups for business reasons have to settle on one use case that they then have to sell to multiple customers. In this case, business units at the bank will bring Layer 6 all kinds of use cases, which should be more intellectually stimulating, Rhodes said.

Jacobs noted that the Layer 6 team includes several experts on natural language processing and image recognition, which could have uses in banking. For instance, NLP can be applied to voice recordings of call center interactions to conduct sentiment analysis.

“It can understand when someone is getting frustrated, or are they happier at the end of the call than they were at the beginning,” Jacobs said.

Image recognition could be used to deter fraud; facial recognition can be used to check a current selfie against one on file more accurately. Loan applications and other documents are also images that can be analyzed for signs of fraud.

“There’s no doubt there will be applications for it,” Jacobs said. “It’s a little early. We’re on day five.”

Editor at Large Penny Crosman welcomes feedback at

For reprint and licensing requests for this article, click here.
Artificial intelligence Machine learning Analytics Predictive analytics Data mining Fraud detection Lending Fintech M&A TD Bank