As the inaugural head of RBC’s new AI group, Bruce Ross spends his time seeking out opportunities to use and develop AI tools across the bank’s five business segments and implementing them.
Ross, who stepped into the role in February, reports directly to CEO Dave McKay. He spent 12 years as group head of RBC technology and operations and previously worked on technology for HSBC Bank Canada. Prior to joining RBC, Ross was IBM’s general manager of global technology services in North America, responsible for the then-emerging fields of cloud, mobility, social and analytics.
His mandate at RBC is to speed AI adoption and the benefits that come with it.
“Transformation is defined by two variables: the quality and scale of the people you commit to it, and the money you put behind it,” Ross said. “We committed to [delivering] $700 million to $1 billion of net benefit to RBC, our shareholders and investors through AI execution on a run rate basis [by the end of] 2027.”
Every year, RBC spends upwards of $5 billion in technology, including AI. As part of its push to adopt AI across the board, Ross said the institution is investing in proprietary technology, productivity in wealth management and partnering with regulators to manage risk.
To steepen the bank’s AI adoption curve, RBC Borealis — the bank’s in-house research institute — employed more than 100 researchers to build a proprietary foundation model, Atom, in addition to using commercially available models. Atom, short for asynchronous temporal model, was trained on large-scale financial datasets and used across 15 RBC products and processes last year.
For example, RBC can extract insights from the transactional data of tens of millions of clients to create personalized offers and highlight next best actions. That data is unique, Ross said. While competitors can buy AI models from vendors, they won’t have the same capability without the same kind of data to train it on.
Ross also claimed that RBC Lumina, its internal enterprise data and AI platform, is built on Canada’s largest private sector graphics processing unit farm for AI training — second only to the federal government.
Keeping advisors in the AI loop
According to Ross, there are three layers of AI for financial institutions: one, improving the productivity of individual employees; two, broader implementations that boost revenue or reduce costs; and three, client-facing tools.
Most financial institutions are at level one, some are at level two and very few are at level three, Ross said. RBC is approaching level three cautiously.
“When you’re dealing with people’s financial wellbeing, … you have to be very, very careful about that,” Ross said. He wants to have a human in the loop for all AI processes.
In wealth management, it means the advisor remains client-facing rather than the technology. For example, an advisor might use AI to help make portfolio management decisions and scenario plans, but the human advisor will still present recommendations to clients.
Thus, RBC isn’t looking to offer interactive AI wealth planning for clients — employees will always be involved.
Balancing risk, managing compliance
While RBC is “aggressively” looking at how AI can help in various business segments, it’s also taking a measured approach by treating regulators as partners and building strong internal governance, Ross said.
“[Banks and regulators] are educating one another about the characteristics of what good compliance looks like in an AI environment because everybody is learning at the same time,” Ross said. “We put such a huge amount of investment into AI that we should be able to bring some intellectual capital to our regulators as well.”
As AI moves faster than regulators can react, Ross isn’t waiting for regulation to catch up. Instead, RBC has built three internal lines of defence before taking an AI feature to market. There are staff knowledgeable in AI in each business group as the first line, an AI-specific risk management team as the second line and the bank’s audit group as the third.
Equipping staff with the right AI knowledge and setting rigid standards for AI use speeds up adoption, as it means the rules are consistent and well-understood, Ross said.