AI agents at banks are facing a familiar human pressure: prove they can make money for their employers, or risk being replaced. 

Mitch Bouchard, general manager of Canada for the California software company ServiceNow, said in an interview that the company is working with Canada’s big banks to assess the business value of their AI agents. He said the Big Five have moved from an era of expensive AI experimentation to a more disciplined phase, where AI agents need to prove their worth or face the robot equivalent of layoffs.

“We’re seeing banks turn things off,” he said. “We’re seeing them bring more discipline to how they actually assess the value of AI.”

Banks have poured massive amounts of money into AI. Tech research firm IDC projects financial services companies worldwide will spend about $160 billion on AI tools in 2026, a figure it expects to grow by 35.5 per cent over the next three years. But while BMO, RBC and TD have set targets for generating a return on that investment, only TD has put a dollar figure on what AI is currently delivering, saying it generated roughly $195 million in what it calls “AI value” in the first nine months of its fiscal year.

Alexandra Mousavizadeh, co-founder and co-CEO of Evident Insights, which publishes an annual report tracking AI adoption across the financial services sector, said she thinks that’s about to change. “We’ve done the experiments, we’re starting to see what’s going to pay off,” she said.

In an emailed statement, Greg Brown, BMO’s head of enterprise AI, said “the measure of progress won’t be how many agents we deploy, but the outcomes they deliver.” RBC, TD and Scotiabank did not provide comments by deadline, and CIBC did not respond.

Reuben Piryatinsky, a consultant who works with Canada’s big banks on AI, said the ones that have deployed the technology most successfully have several things in common: making the business team, not the tech team, accountable for financial outcomes; creating centralized AI standards for the entire bank; and making sure every step of a process is optimized before automating it.

“If your processes are broken, and you’re building AI agents to follow these processes, they’re just going to create even more of a mess,” Piryatinsky said.

Bouchard said the banks have been thinking carefully about how to put guardrails in place for AI agents, which perform work like responding to employee HR requests or reviewing mortgage applications. When systems are in place to track, manage, and turn agents off if something goes wrong, banks can take on more interesting work, he said. “Once that’s built, you can accelerate innovation again.”

Mousavizadeh said the gap is widening between banks that have prioritized AI investments and those that haven’t. All five major Canadian banks are well-positioned to pull ahead in the global AI race, with each building or already operating advanced platforms, she said.

Mousavizadeh agreed that underperforming AI agents are at risk of being put out of work as banks look to cut spending with low returns. She said that for now, the jobs of the AI agents’ human co-workers are comparatively secure.

“If anything, you’re working longer hours and hiring more people,” she said. “Those 50 banks that we’re following are adding headcount rather than reducing it.”