When a service business decides to automate something, the instinct is usually to start with the part customers actually see: a chatbot on the website, an AI-answered phone line, an email agent that replies to inquiries overnight. It feels like the highest-leverage move, since it's the interaction people notice first. Two things are worth knowing before you point your first automation project there. It usually isn't where the payoff shows up fastest, and it's the one place where a mistake becomes your problem, not your software vendor's.
A tribunal already decided who owns the chatbot's mistake
In 2024, a passenger asked Air Canada's website chatbot about bereavement fares after a family death. The bot told him he could apply for a reduced fare retroactively within 90 days. That wasn't the airline's actual policy, and Air Canada refused to honour it, arguing in front of the BC Civil Resolution Tribunal that the chatbot was "a separate legal entity that is responsible for its own actions." The tribunal didn't buy it. It ordered Air Canada to pay the fare difference plus damages, and tribunal member Christopher Rivers wrote the line that's now cited in almost every piece written about AI liability since: "It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot," according to CBC News.
That ruling is Canadian, it's small claims court rather than a landmark statute, and the amount involved was modest. None of that changes what it establishes for a service business today: whatever your AI tells a client about pricing, deadlines, refunds or exceptions to policy, that's your word. It doesn't matter which vendor built the underlying model, and it won't matter that the answer was a plausible-sounding mistake rather than a deliberate one.
Customers can tell when they've been deflected
The liability question is only half the argument. The other half is that client-facing AI, done badly, doesn't even deliver the efficiency it promises. Qualtrics surveyed more than 20,000 consumers for its 2026 Consumer Experience Trends Report and found that nearly one in five people who'd used AI for customer service walked away with no benefit at all, a failure rate close to four times higher than for AI use in general. Consumers in the same report ranked AI customer service among the worst uses of the technology for convenience, time savings and usefulness.
"Too many companies are deploying AI to cut costs, not solve problems, and customers can tell the difference," Isabelle Zdatny, head of thought leadership at Qualtrics' XM Institute, said of the findings.
That tracks with what we see with clients. A bot that's built to deflect volume, rather than actually resolve a client's question, tends to make a business look worse than having no automation there at all. A caller who gets looped through a phone tree that can't answer a real question remembers the friction, not the technology behind it.
Automate the plumbing before you automate the conversation
None of this means client communication can't eventually be automated well. It means the order matters. Start with everything that happens before or after the conversation: pulling up the client's record automatically, drafting a reply for a person to review and send, logging the interaction, triggering the follow-up task. All of that saves real time without ever putting an unsupervised answer in front of a client, and it's where a lot of the actual hours get clawed back. The Canadian Federation of Independent Business found that small businesses using generative AI gain back roughly an extra hour for every hour they spend with it, with a typical 29 percent productivity gain in the first year, and that return shows up well before anyone lets a model talk to a customer unsupervised.
When you do move the actual conversation to AI, put a hard boundary on what it's allowed to promise. Money, deadlines and exceptions to policy should route to a person, or at minimum get flagged for same-day review, the way you'd handle a new hire's first few weeks on the phones. Read a sample of transcripts every week for the first stretch, not just the ones that generated a complaint. And make sure there's an obvious, fast way for a client to reach a human the moment the bot is out of its depth. The Air Canada case wasn't costly because the airline used a chatbot. It was costly because nobody was checking what the chatbot was telling people, and nobody had drawn a line around what it was allowed to say.