RBC Is Putting AI Where You Can Least Afford to Get It Wrong

In January, RBC bought a position inside the website where most Canadians go to look for a house.

REALTOR.ca, owned by the Canadian Real Estate Association, carries more than 60 percent of the country's online real estate traffic. In a nation of 41 million people it reports over 110 million unique visitors and 630 million visits a year. That is not a website. It is the front door to the housing market, and RBC now has a desk inside it. The first phase, live for the spring buying season, put co-branded financial literacy content and a direct mortgage inquiry path onto the platform, with REALTOR.ca earning referral fees on qualified leads. Its CEO, Scott Neil, described the goal as a consistent flow of mortgage leads to RBC. The second phase adds AI-powered property search and financial guidance.

Erica Nielsen, the group head responsible for RBC's consumer banking business across Canada, the Caribbean and the US, wrote about that partnership on LinkedIn this week. She described AI as already making a real difference for clients "at the turning points that shape their lives," and announced that she will be publishing a running account of how the technology is landing inside her business.

Take the warmth out of that phrase and it says something quite specific. RBC is putting AI at the moments when its customers have the most money on the line and the least idea what they are doing.

That is not a gotcha. It is a description, and Nielsen would probably accept it. She names the moments herself: buying a house, renewing a mortgage, opening an investment account. In order, that is the largest debt most people will ever take on, the moment that debt gets repriced, and the point at which a person's savings stop being cash. Banks call these turning points because that is what they are, and because they are the only three or four occasions in a decade when a retail customer is paying full attention.

They are also where the leverage sits. At each one, someone is making a decision worth hundreds of thousands of dollars with a fraction of the information held by the institution on the other side of the screen. Nielsen led with that, which is more candour than most banks bother with. What her post does not address is whose interests the software serves at the moment it arrives.

The other two

Homebuying is the loudest of the three. The other two are quieter and, in at least one case, more commercially significant than it looks.

Mortgage renewal. Nielsen mentions renewals almost in passing, as something clients can now complete in the mobile app with AI working in the background. Renewal is the one turning point where the incumbent's entire job is to stop the customer from shopping around. AI applied to renewal is, in plain terms, retention technology. It may well produce a smoother experience. It is not a neutral one.

Investment account opening. The most benign of the three, and probably the most honest description of where retail banking AI actually lives right now: document extraction, eligibility checks, identity verification, next-best-action prompts. Boring, high volume, and where most of the near-term money is.

Why the turning points, and why now

There is a number behind all of this. At its 2025 Investor Day, RBC committed to C$700 million to C$1 billion in net enterprise value from AI, on a run-rate basis, by the end of 2027. Bruce Ross, who took over the bank's new AI Group in February and reports directly to CEO Dave McKay, has repeated that figure as the measure he is held to.

You do not reach a billion dollars through chatbots. You reach it through conversion and retention at exactly the moments Nielsen listed. The turning points are not where AI happens to be useful. They are where the value is, and the strategy is built around them.

RBC has the machinery to do it. Borealis was founded in 2016, well before this became a mainstream corporate subject. Close to 27,000 employees now use RBC Assist. Around 8,000 people in Capital Markets use Aiden. Lumina, the enterprise data and AI platform, runs on one of the largest private sector GPU clusters in the country. ATOM, the bank's proprietary financial foundation model, was used across fifteen products and processes last year. RBC ranks third globally and first in Canada on the 2025 Evident AI Index, ahead of every other Canadian institution.

What the rest of the industry already knows

The uncomfortable part is that the sector's own practitioners have identified the two things that go wrong when AI is deployed at high-stakes decision moments, and neither is addressed in Nielsen's post.

The first is trust, and it is thin. Experian Chief Innovation Officer Kathleen Peters told CIOnews in May that a lack of trust is the main brake on AI adoption, noting that just 25 percent of Americans trust AI in a retail setting. Retail is low stakes. A mortgage is not. Experian's answer was to build a verification layer, Agent Trust, with Visa and Cloudflare, on the premise that people will not hand consequential decisions to systems whose provenance they cannot check. A turning point is where trust matters most and is hardest for the customer to verify.

The second is measurement, and this is where Nielsen's post is weakest. Wendy Turner-Williams, Chief Data and AI Officer at SymphraAI, told CIOnews in June that organisations are tracking the wrong things entirely. "Very few people are measuring things like decision quality or decision velocity," she said, arguing that leaders keep reporting deployment while producing no evidence the business actually moved.

Look at what RBC publishes: 27,000 users, 15 products, fifteen hundred patents, a GPU cluster. Every one of those is a deployment metric. At a turning point, the number that matters is whether the customer's decision was better, and nobody in banking is publishing that. Turner-Williams's point about baselines cuts hardest here. What was the before state? What did the mortgage decision look like without the AI? Without that, there is no credible claim in either direction.

The colleague at the roundtable

The second half of Nielsen's post turns to staff, and the writing gets looser. She relays a colleague describing teams building project dashboards, mapping workflows and using AI to pressure-test their thinking before meetings, with work that took days now taking hours.

That is a nice anecdote and no kind of evidence, and the industry has a more precise account of what actually happens. Vikas Krishan, Chief Digital Business Officer at Altimetrik, told CIOnews in June that the real bottleneck in banking is neither the models nor the pilots but the operating model underneath them. His observation about adoption is the one worth holding against the roundtable story: junior staff take to AI natively, senior leaders approach it as veterans of earlier technology cycles, and the middle layer struggles, because their work spans people management and individual execution in ways that do not map onto agent-assisted workflows. Reorganisation, not retraining, is the work nobody has started.

Krishan also described where this ends up. "At the moment, we're using people as a guardian function," he said, before outlining a model in which that inverts. That is the conversation no bank wants to have in public, and there is already a data point in the sector: Klarna has claimed its internal customer service AI performs the work of 853 full-time agents, saving roughly $60 million.

Nielsen's framing is that AI frees people to spend their time on clients. Every bank says this. If work that took days now takes hours, the arithmetic surfaces somewhere eventually, and a first-person account of an AI rollout that never touches it is not really a first-person account.

The test

There is no reason to doubt Nielsen means what she says. RBC's head start is real, the infrastructure is real, and a group head narrating a rollout in public rather than through a press office is unusual and worth encouraging.

But she chose the frame. If the story is AI at the turning points that shape people's lives, the series has to eventually cover a turning point where it went wrong: a recommendation that suited the bank more than the client, a renewal that would have gone better elsewhere, a piece of guidance the customer had no way to interrogate. Those cases exist in any deployment at this scale.

The good version of this series answers Turner-Williams's question and publishes a baseline. The ordinary version gives us eleven more posts about colleagues at roundtables. We will know which one it is by about instalment four.