CIBC Mellon administers $3.5 trillion in assets in a market where a single day can cost hundreds of millions. Here’s how the firm is using AI to make sure it’s always the one that gets there first.

On February 18, 2021, the first Bitcoin ETF in North America launched in Canada and pulled in over $500 million in its first week. The very next day, a second Bitcoin ETF launched on the same exchange. The second, according to Mal Cullen, CEO of CIBC Mellon, pulled in $35 million. Although it was a similar product in the same market, launched only one day apart, there was a big difference in the numbers they each pulled.

That contrast is the kind of thing that defines Cullen’s world. CIBC Mellon is one of Canada’s largest asset servicing firms, responsible for the administration of $3.5 trillion in assets. In that kind of environment, a single day’s delay costs real money — sometimes, hundreds of millions of it.

“What is the value of one day in your business?” Cullen asked the audience at Appian World 2026 in Orlando last month. “In our business, it can be massive.”

That question is now driving one of the more consequential AI deployments in Canadian financial services, and the lessons from it reach well beyond capital markets.

The Problem With 30 Years of Process

CIBC Mellon has been in business for 30 years. It is a joint venture between BNY — which administers more than $59 trillion in assets globally — and CIBC, one of Canada’s largest banks. That parentage brings enormous scale and institutional credibility. It also brings complexity.

“There’s only one thing better than being owned by a bank,” Cullen said, with a measured smile, “and that’s being owned by two banks. Two banks means two compliance teams, two risk teams, and two opinions on almost everything.”

Getting serious about AI for CIBC Mellon meant resisting the urge to move fast. Before a single tool was deployed, the firm went back to basics. The team mapped every workflow, identified where capacity was constrained, where risk was highest, and where manual work was concentrated. What they found surprised them.

“The people doing the work were not the problem,” Cullen noted. “It was how work was flowing between teams that was causing the constraints.” In other words, the technology was never going to fix what the process had broken.

From Assembly Lines to AI

The first major deployment involved fund accounting — a process CIBC Mellon runs at enormous scale. The firm produces roughly 350,000 fund valuations every month, each one subject to tight deadlines and strict accuracy requirements.

For years, the process ran vertically: One accountant owned a file from start to finish. It was a process built on individual expertise, which meant it was also built on individual limits and nearly impossible to scale. The firm redesigned it horizontally, distributing work across specialized teams. But that created a new problem — handoffs between teams became a source of friction and delay. Supervisors had no visibility into where work stood without asking.

A fund accountant with over a decade of experience, who knew the process better than anyone in the building, used Appian’s low-code platform to build what Cullen calls a “control tower” — a workflow system that gives every team real-time visibility into where work is in the process, automating the handoffs that had been causing the delays.

The result was a 34% efficiency gain on a single process. At 350,000 valuations a month, that compounds quickly.

“He told me he designed out everything he didn’t like about his job,” Cullen said. “When you get people who understand the process to work on it, they don’t automate the manual things that were there before. They re-engineer the process and make it better.”

The ETF Problem

The second example goes back to that Bitcoin ETF story. When an ETF launches or distributes returns to unit holders, it involves a complex web of counterparties — the fund manager, the custodian, the exchange, the market maker, and the transfer agent. Each one needs to be notified. Each one has a role. Getting that fund out a day earlier requires all of them to move in sync.

An ETF product expert at CIBC Mellon built a workflow on Appian that brings transparency across all those counterparties in one place — turning a fragmented, email-heavy process into something automated and auditable.

Three weeks before Appian World, CIBC Mellon demonstrated the application to Canada’s largest ETF providers at a client user group in Toronto. “The room got quieter,” Cullen recalled. “People leaned in. One of our largest clients said to their peers: That just saved me a material amount of time in my day.”

The Governance Question

None of these developments happened quickly, and Cullen is direct about why. CIBC Mellon is not using AI in anything client-facing yet. Every AI deployment to date is internal — contained within defined workflows, auditable, and reviewed by humans before any output affects a client.

“We can only move with AI as fast as our clients’ comfort level,” he said. “We’ve intentionally not embedded AI into anything client-facing because we don’t feel the governance is there yet.”

The numbers from the broader market confirm what Cullen already knew. According to a new study by Harvard Business Review Analytic Services, sponsored by Appian and released at the conference, 92% of organizations agree that AI agents need rules-based guardrails to operate safely — yet fewer than half have actually defined them. CIBC Mellon is among the organizations that have chosen to build the foundation before scaling the deployment.

Inside the organization, that caution is shaping how the firm is preparing its people. The firm has designated 100 employees out of roughly 2,000 as AI champions. These champions are given early access to tools, time to build use cases, and a mandate to pressure-test applications in sandboxes before anything moves to production. They run weekly internal sessions called “Artificially Speaking,” bringing in firms like Snowflake and Microsoft to share what’s working and what isn’t.

Cullen has watched this pattern play out before. Twenty years ago, he was having conversations with CTOs who said cloud was a fad and would never be trusted with sensitive data. Then hybrid cloud emerged, giving organizations a middle path — the efficiency of cloud infrastructure without the perceived loss of control. He expects the same arc with AI.

“I think you’ll see hybrid AI next,” he said. “Contained, governed, but moving.”

What the Rest of the Market Can Learn

CIBC Mellon’s story is not a story about a technology breakthrough. It’s a story about organizational discipline applied to a powerful tool. Measure before you build, put the people who know the process closest to the problem, and govern before scaling.

Those lessons apply well beyond asset servicing. In a market where only 16% of organizations report realizing meaningful value from AI, the organizations getting real results are the ones that treated governance as a feature, not a constraint.

“Don’t start with the technology,” Cullen told the audience at Appian World. “Measure everything first.”

In an industry where one day can mean the difference between $500 million and $35 million, that kind of patience turns out to be its own competitive advantage.