In 2024, Air Canada was held responsible for the wrong information its website chatbot gave a customer about bereavement fares. The tribunal rejected the idea that the chatbot was a separate service the airline was not answerable for. The company owned what the bot said. The principle for any regulated business is simple: you are accountable for what your AI tells your customers.

That matters more as AI agents move into Canadian customer service. The market is moving fast. Salesforce just paid around $3.6 billion for a customer-service AI agent, and Canadian banks are rolling out their own assistants quickly.

Against that backdrop, a survey of more than 2,500 senior decision-makers found that about three quarters of enterprises have already pulled back or shut down a live customer-facing AI agent. The pullback rate was higher, around 81%, among the organizations with the most mature AI governance. That looks backwards until you see why. Those organizations have the monitoring, escalation paths, and audit trails to notice when an agent gives a wrong answer about a bill, a benefit, or a claim, so they catch it and act. The rollback is the quality control working, not the technology failing.

81%

of organizations with mature AI governance had pulled back a live customer-facing agent, against about 75% overall. Better controls mean you notice the problem sooner. (Sinch survey of 2,500+ decision-makers, 2026.)

The costly mistake in a regulated business

In a regulated business, the costly mistake is rarely the case you left with a human. It is the automated answer you cannot stand behind later, in front of a regulator or a tribunal.

Canada has no single AI law to point to. The proposed federal act died in 2025, so the rules already on the books are the ones that bind you. Quebec's Law 25 requires you to tell people when a decision is made by automation alone and to let them ask for a review. New OSFI guidance makes banks and insurers responsible for the risks of their AI models. And the Air Canada case showed that the company, not the bot, answers for a wrong answer.

Customers feel this too. Their preference for dealing with a person is still rising, not falling.

What to do with this

None of this argues against AI in customer experience. It argues for deploying it the way a regulated operator should. If you are deploying AI to help with customer experience in a regulated industry, consider the following:

  1. A rollback is not a bad thing. It usually means your quality controls are working, and with further training your agentic experience will improve. Treat each pullback as a normal step in building the capability rather than a failure.
  2. Always pair quantity with quality. The percentage of cases an agent solves on its own, as a stand-alone metric, doesn't show the quality of the work behind it. Think of it as AHT without FCR. You want to look at that solve rate alongside how quickly you are catching mistakes, bringing a person in to take over from the AI, and closing the loop.
  3. Get your workforce trained for the hard cases. Train your people to handle the complex and sensitive problems, bringing the human touch, empathy, and experience where it is needed most.

The firms that do this well will not be the ones that automate the most. They will be the ones that can tell when an agent is getting it wrong, step in without hesitation, and keep training it until it earns more of the work. That is how you build an AI capability you can stand behind.