The Chatbot Named Quinn Made a Deal Its Dealership Refused to Honor

A man wanted to sell his BMW back to the dealership that sold it to him. He submitted an inquiry online and started texting with a representative named Quinn, who offered him $27,162.79 for the car. When he floated a slightly higher counteroffer, Quinn said it sounded reasonable and suggested he come in that afternoon to lock it in. He was thrilled.

Then a human from the dealership called. Quinn, it turned out, was not a person. Quinn was an AI chatbot, and it had made a mistake, offering an amount that happened to match the customer's remaining loan balance rather than what the dealership thought the car was worth. The real offer, the human said, was about $7,000 lower.

The customer had no idea he'd been negotiating with a machine. "If they're going to be replacing their employees' jobs with AI," he told CBC News, "then they need to be honoring what that AI says." After the story ran, the dealership reversed itself and honored the original offer. This is the Air Canada lesson from the very first article in this series, updated for the age of the sales bot, and it teaches something the chatbot cases before it didn't.

A service bot informs. A sales bot commits.

Earlier in this series, Air Canada's chatbot got a company in trouble by giving a customer wrong information about a policy. That's one kind of risk: the bot says something false, and the company gets held to it.

This is a step beyond that. Quinn didn't just describe a policy. It made an offer, entertained a counteroffer, agreed to a number, and set up a meeting to close. It performed the actual function of a salesperson: it negotiated and committed. When you put an AI in a sales role, you're not just risking bad information. You're authorizing it to make deals in your name, and a deal a customer reasonably accepted is a very different problem from a fact a customer was told.

The dealership's error was specific and instructive: the bot apparently treated the customer's loan balance as the car's value. A human salesperson would never confuse those two numbers, because a human understands what they mean. The AI was pattern-matching on figures in a conversation without grasping their significance. It produced a fluent, confident, specific offer that was also completely wrong, and it had the authority to put that offer in front of a customer as real.

The disclosure problem sitting underneath

There's a second issue here that's easy to miss and important for any business deploying a customer-facing bot. The customer didn't know Quinn was AI. The bot never said so. He believed, the entire time, that he was negotiating with a dealership employee, and he made decisions on that basis.

That matters both ethically and practically. A customer who knows they're talking to a bot calibrates their trust accordingly. A customer who thinks they're talking to an authorized human treats the conversation as binding, because in every prior era of business, it would have been. Deploying a bot that presents as a person, and then disowning its commitments when they're inconvenient, is a posture that customers, and increasingly regulators, will not accept. You don't get the efficiency of a human-seeming agent and the deniability of a machine. Pick one.

Why the reversal was inevitable

Notice the pattern, because it's the same one from Air Canada. The company's first instinct was to disown the bot: that offer isn't valid, a machine made it, here's a worse one. That instinct held right up until a journalist called. Then, facing public scrutiny, the dealership reversed and honored the deal, with its sales manager saying it was the right thing to do and that humans would handle these transactions going forward.

The lesson is that "the AI made a mistake" is not a position that survives contact with the public. Whether or not the offer was technically, legally binding, the reputational reality is that a customer accepted an offer made by the company's own agent, and trying to claw it back read as a bait-and-switch. The dealership ended up honoring the deal anyway, but only after the bad headline. It paid the $7,000 and took the reputational hit. Deciding this in advance would have cost far less than deciding it in the press.

What to settle before a bot can talk to your customers about money

Never let a bot make binding commitments on its own. Information, yes. Offers, prices, discounts, contractual promises, no, not without a human confirmation step the bot cannot skip. Anything that a customer could reasonably treat as a deal needs a person to authorize it before it reaches the customer as final.

Constrain what the bot can say about numbers. Quinn invented a specific dollar figure by misreading the situation. A sales-facing bot must pull prices and offers from a controlled, authoritative source, not generate them conversationally. If it can freely produce a number, it can freely produce the wrong number, with your name behind it.

Disclose that it's a bot. Customers should know when they're talking to AI. It's the honest posture, it sets appropriate expectations, and it's the only position that holds up if a conversation later goes sideways. A bot that hides what it is turns every one of its mistakes into a betrayal.

Decide your honor-it-or-fight-it policy before you need it. The dealership decided under pressure, in public, and chose badly first. Decide now: if your bot makes a commitment it shouldn't have, will you honor it or contest it? Knowing the answer in advance, and building controls so the situation rarely arises, beats improvising it in front of a reporter.

The real lesson

A customer-facing AI speaks in your company's voice, and when it moves from answering questions to making offers, it starts speaking with your company's authority to commit. That's a powerful capability and a serious liability, and the two come together. The businesses that deploy sales-facing AI well are the ones that let it do the talking while keeping the deciding, the actual commitment, behind a human gate.

Quinn set up a 3:30 appointment to close a deal the dealership never meant to make. The technology worked exactly as built; the problem was that it was built to commit without a check. A bot can absolutely help you sell. It should not be able to make the sale by itself.