Foteini Agrafioti spent the late 2000s collecting and analyzing heartbeat signals for her PhD project. Working from the University of Toronto’s Bahen Centre for Information Technology, she hoped to decode how to identify humans via their heartbeats, just like how other biometric features, from fingerprints to retina scans, are used to tell people apart.

“If you look at heartbeats by the naked eye, they more or less look the same,” she said. “Then we started using machine learning that uncovered unique patterns for every person hidden in the signals. We were able to prove that your heartbeat is effectively as unique as a fingerprint.”

Agrafioti’s PhD project was her first foray into machine learning, which revealed the “power of machine learning,” she said. “I fell in love with (the technology) at that moment and how it can show you things you can’t see and prove things that no one thought would be possible.”

Post-PhD, she created and sold a smart wristband company that identified users based on their heartbeats. She then co-founded and started leading Borealis AI Inc., the

Royal Bank of Canada

’s

artificial intelligence

(AI) research institute.

Instead of studying heartbeats, Agrafioti and her 950-strong team of data scientists and machine learning engineers now spend their time analyzing RBC’s reams of financial transaction data and the latest advancements in AI and machine learning to devise new AI applications to improve the bank’s services and products.

AI was a nascent technology when RBC launched Borealis AI in 2016, with the launch of ChatGPT eight years away. RBC chief executive

Dave McKay

tasked Borealis AI with researching and advancing machine learning. Agrafioti credited McKay with giving free rein to the team “to pursue topics … that wouldn’t have application within a year. It was a big investment made early … and a leap of faith.”

The financial services industry has been among the foremost adopters of AI, but as the technology burrows into the mainstream — with some warning that the boom is now morphing into a bubble — banks are under increasing pressure to prove the technology’s real-world merit and that their capital-intensive AI efforts will bear fruit.

RBC is operating under the same scrutiny, with McKay pushing a target to reach up to $1 billion in revenue growth and cost savings from its AI investments by 2027, he said during the bank’s March 2025 Investor Day call. The stakes are especially high for Canada’s biggest company, which has annually pumped $5 billion into its technology investments and staked its future growth on the ability to leverage data and AI.

Nothing’s a given

Last year, RBC ranked among the world’s top three banks in AI maturity for the third straight year, surpassing all its Canadian rivals, according to the Evident AI Index, a scorecard published by benchmarking platform Evident Insights Ltd. that tracks AI adoption across financial services.

The index, which assesses lenders based on various metrics, gave RBC high scores for its depth of innovation, such as research and patents, and its tech capability and development.

Toronto-Dominion Bank

was the only other Canadian bank to crack the top 10 and notched decent scores in talent and innovation. TD is a founding sponsor of Toronto’s Vector Institute, one of Canada’s three national AI institutes, and it has partnered with the organization to train talent and to conduct joint research on AI use cases.

But success in AI wasn’t a given for RBC despite its rank as Canada’s largest bank, according to Evident chief executive Alexandra Mousavizadeh.

“We don’t find that there’s a correlation between AI capability and size,” she said, adding that big legacy banks holding massive amounts of fragmented data “have to move mountains before they can even run AI on it.”

RBC’s focus on AI research and talent helped.

Borealis AI “made sense” because RBC required a conduit to transform its data into something useful and a vehicle to attract technical talent, Jonathan Aikman, a finance lecturer at Queen’s University in Kingston, Ont., and founder of AI consulting firm NordAI Analytics Corp., said.

From the start, the Borealis AI team had access to internal datasets and the computational capacity required to “do interesting work,” which gave them the flexibility to innovate, Agrafioti said.

The institute in 2020 inked a partnership with chip titan

Nvidia Corp.

and cloud provider Red Hat Software Inc. that guaranteed it a steady supply of the world’s most advanced and coveted graphics processing units (GPUs) — the building blocks of AI systems. RBC has upgraded and expanded its GPU clusters several times since then.

“We always buy the latest hardware and restock,” she said.

Still, Canada’s talent flight posed a challenge. Talent was “at a deficit” during the institute’s early years, Agrafioti said.

“There weren’t a ton of PhD-trained candidates who weren’t leaving the country,” she said. “It didn’t occur to anyone to take a job in Canada because nobody thought that we would be doing serious AI here.”

Silicon Valley and its tech giants had snatched away some of Canada’s brightest AI minds. For example, Google LLC scooped up the “godfather of AI,” Geoffrey Hinton, and his then-graduate students Ilya Sutskever and Alex Krizhevsky from the University of Toronto in 2012.

“When we lost the big academics in Canada, their labs went with them,” Agrafioti said, adding that Borealis AI adjusted its recruitment strategy to search worldwide. “We changed. We were very specific about building these channels internationally and for many years. We (now) recruit internationally.”

She said the institute is well-connected with the leading technical schools across South and East Asia. Borealis AI has also teamed up with institutions in the United States. It announced in July a partnership with the Massachusetts Institute of Technology’s computer science and AI lab that will give it access to MIT’s tech talent and cutting-edge research.

“We have a lot of students who will finish their degrees anywhere in the world and come to work with us,” Agrafioti said.

The strategy seems to have worked, according to Mousavizadeh.

“What RBC has done well is get that deep talent stack … and supply that expertise across the bank, going from ideation to use cases pretty quickly,” she said. “When you have a big applied AI research arm intertwined with your operations … and lined up with your business heads, that’s the magic.”

AI tools of the trade

In some ways, RBC’s path of AI development has mirrored that of Canada’s. Like Canada, the bank first built up its research capacity in AI. But for all of Canada’s research firepower, the country has struggled with commercializing AI research, and losing talent and intellectual property to the U.S.

But “we had faith that we would be able to commercialize it,” Agrafioti said of RBC.

Nearly 10 years after launching Borealis AI, RBC is in the thick of trying to wrangle $700 million to $1 billion in value from AI. McKay has said that RBC can hit that number once its eight major AI use cases are all deployed and fully scaled.

RBC’s AI tools have been built for the purposes of helping the bank and its clients supercharge productivity and revenue and avoid losses, Agrafioti said.

In 2023, RBC announced ATOM, an AI model trained on billions of client financial transactions, which it uses to peer more deeply into a borrower’s credit history, something it said will allow it to extend loans to people with less credit history.

RBC has also applied ATOM to its Avion Rewards program, where it is used to tailor recommendations for its 1.3 million monthly users. This application has generated “cost savings” and boosted the number of users redeeming their points, according to the bank. RBC is now reviewing the use of ATOM in other aspects, such as personalizing digital banking platforms and fraud detection.

“Our AI is pushing work around fraud detection that’s getting better at protecting clients from incurring fraud and (related) losses,” Agrafioti said.

In 2020, RBC launched Aiden, an AI platform that allows clients and staff to automate electronic trading, streamline workflows and generate data-driven research. The use of Aiden has grown 28 per cent annually and has led to a 60 per cent improvement in research analysts’ speed in generating insights from company earnings.

This May, RBC launched a new AI team for its capital markets division, which has been using and scaling the platform.

RBC’s models generate insights that “you just can’t buy,” Bruce Ross, RBC’s group head of technology and operations, said on RBC’s Investor Day in March. “You’re not going to be able to get it from OpenAI or anyone else.”

Analysts say the lender is well-placed to achieve McKay’s $1-billion revenue and cost-saving target. RBC has “shared a credible path” to deliver its financial goals via scale and technology, BMO Capital Markets managing director Sohrab Movahedi said in a March note.

Its offerings that leverage AI create “convenient and hyper-personalized experiences” for clients, which could help it retain its leading position in Canada and spur greater earnings and profitability, he said.

While McKay has outlined eight major use cases of AI, the bank’s internal AI projects likely number a thousand, Aikman said.

“If you look at the efficiencies they’ve created internally, you’d see it’s actually vastly more” than $1 billion, he argued.

Its AI capabilities that focus on efficiency, personalization, and risk management will be crucial for the lender as it pushes its global expansion, according to Aikman. RBC had the foresight to “see what their counterparts in London, New York, Hong Kong and Singapore were doing, and they understood … what they needed to do to be able to compete,” he said.

A double-edged AI sword

Generating a return on AI investments won’t come without hurdles. Lenders that have bet big on the tech face challenges related to regulation, competition and talent.

For example, complying with current and future AI rules “may limit the speed and scope of AI adoption” for RBC, Aikman said. He added that new entrants, from fintech startups to retail giant subsidiaries such as President’s Choice Financial (

Loblaw Cos. Ltd.

) and Canadian Tire Services Ltd., are eroding the big banks’ traditional customer bases and increasing the competition for AI talent.

McKay framed the risks as “macro forces of change,” saying the bank is staring down “heightened regulatory scrutiny; disruptive threats from private capital, big tech and fintech; and accelerating technology disruption … and an AI arms race that every business needs to keep up with.”

He also warned about the economic headwinds from the Canada-U.S. rupture, noting that the bank is closely watching

Canada-U.S.-Mexico Agreement

(CUSMA) renegotiations.

Any success from RBC’s AI efforts, however, also serves as “a double-edged sword,” said Kai Li, a Canada research chair in corporate governance and finance professor at the University of British Columbia.

“It might make the shareholders happy,” she said, but AI-driven operational efficiencies will almost certainly lead to job cuts and shrinking employment opportunities.

Agrafioti said that RBC “does not expect (job losses) at scale from AI at this moment. This is a strategy that’s focused on… opportunity for new revenue.”

Li also cautioned that the bank will need to overcome the biases often found in data.

“For example, because of my cultural background and gender, I may be advised to play it safe and to not invest in the stock market, but to buy bonds or money market funds,” she said. “If you don’t overcome the biases in your data, some people in the community will be underserved or wrongly served.”

Agrafioti said Borealis AI is operating at full speed to reach McKay’s target, and Canada’s progress on AI has been supporting the lender’s goals.

Ottawa has poured $4.4 billion into AI over the past 10 years, and the

Mark Carney

government has only doubled down on the technology, incorporating the tech into the public service and appointing Canada’s inaugural

AI minister

in May.

Industry and government brainstormed for years on how to build up the country’s AI ecosystem to curb the brain drain, said Agrafioti, who co-chairs the federal advisory council on AI.

“I want to believe that we reversed (the brain drain) quite a bit,” she said. “I don’t have to pitch Canada anymore. People get it now.”

Canada’s three national AI institutes in Toronto, Montreal and Edmonton have become a “magnet for researchers who want to study AI,” she said.

That reinvigoration of Canada’s AI talent pool has meant that Borealis AI’s recruitment is increasingly “trending more toward Canada,” Agrafioti said. The institute is staffed by 100 PhD scientists and 850 AI developers and data engineers from around the world. This January, Borealis AI also teamed up with homegrown Canadian AI startup

Cohere Inc.

to co-develop a generative AI (GenAI) tool focused on improving productivity and efficiency for financial services.

Agrafioti said her team will continue to focus on creating and tinkering with GenAI applications that drive efficiency, which will free up the bank to have “more meaningful” interactions with clients.

“It’s not just about (using) AI to reset a password or find a transaction,” she said. “We want to change the relationship we have with our clients by removing mundane interactions using technology. We’re lenders. This is who we are. And these are meaningful domains for us.”

Still, “given how fast AI technology evolves, it’s hard to tell how any single bank can stay on top of this,” Li said. “Time will tell which is the winning strategy: to have an in-house AI institute or to outsource.”

• Email: ylau@postmedia.com