← AI, commerce and societyA6 · LESSON 3 OF 7

A6, Lesson 3: When AI answers wrong and the bill is yours

Tags: Sourced stories / Our view / Illustrative examples. Self-paced reading and practice. No math or code. Every case or news claim carries its source and the date we read it; invented figures are labeled invented.

1. The story

In November 2022, Jake Moffatt's grandmother died. On Air Canada's website, he asked the airline's chatbot about bereavement fares. The chatbot told him he could book now and apply for the reduced fare afterwards, within 90 days. That was wrong: Air Canada's own policy page said bereavement fares cannot be applied after travel. Relying on the chatbot, Mr. Moffatt booked two flights for $1,630.36. When he applied for the refund, Air Canada refused, and later argued before the tribunal that it could not be held liable for what its chatbot said.

In February 2024, British Columbia's Civil Resolution Tribunal, a small-claims tribunal in Canada, disagreed. Air Canada had argued that its chatbot was, in effect, a separate legal entity responsible for its own actions. The tribunal rejected that argument in this case: the chatbot "is still just a part of Air Canada's website," it found, and "it makes no difference whether the information comes from a static page or a chatbot." It found Air Canada liable for negligent misrepresentation and ordered it to pay Mr. Moffatt $812.02. One tribunal ruling in one province is not a universal law, and this lesson is not legal advice.

2. The one idea

When your AI speaks to a customer, its words are your business's words. That was the tribunal's finding in this one case; we treat it as an operator habit everywhere, not a legal rule.

3. The kitchen-table version

Two price tags on one shelf. The customer is charged the higher amount and shows you the lower tag. You honor the lower one, because the tag is yours - it does not matter which stock boy wrote it. A chatbot is the tag-writer that never sleeps.

THE REAL NUMBERS, from the decision itself. Mr. Moffatt paid $1,630.36 for the two flights. The tribunal found the bereavement fares would have cost $979.48, so it awarded the difference, $650.88, plus $36.14 of pre-judgment interest and $125 in tribunal fees: the $812.02 order, payable within 14 days. Small money for an airline. The expensive part was the principle, now public forever.

4. Worked example

Our invented candle shop has a 30-day return policy on its official page - and an old FAQ draft in a forgotten folder that says 60. Connect a chat tool that can read both, and the numbers write themselves. Suppose 40 customers ask about returns in a week, and the bot quotes the 60-day version to 9 of them - about one in four. If each wrongly promised return costs the shop an invented $24 in stock it cannot resell, that is 9 x $24 = $216 in a week, and $216 x 52 = $11,232 in a year, from one forgotten file. Figures invented and script-checked; staff time, shipping and reputation are excluded on purpose. The arithmetic is small. The shape echoes Moffatt: the bot and the page disagreed. In our fictional shop, nobody was checking that disagreement.

Instructor: "So delete the old file and move on?" Learner: "Delete it, yes - but the habit is the audit, not the cleanup. There will always be another forgotten folder: an old marketplace listing, a PDF menu, a seasonal banner. The question is whether anyone checks what the bot is reading."

5. What it costs and what it breaks

One source of truth per policy takes an afternoon to build and a standing habit to keep - every promotion, marketplace listing and seasonal page is a new copy that can drift. Human review of money-touching answers costs real minutes every day. Logging what the bot says costs storage and attention. Skip them and the cost shows up the Moffatt way: one customer, one wrong promise, and a dispute where your own website is the evidence against you.

6. Monday checklist

Name the one true home of each customer-facing policy (returns, shipping, price promises). Search for stray copies: old FAQ pages, marketplace listings, PDFs, social bios. Decide who alone may edit the official policy. Turn on logging for every answer your bot gives. Spot-check five real bot answers against the official page, and write down what you found.

7. Try it yourself (five minutes)

Pick one policy in a business you know, returns or shipping. Ask any AI chatbot about it as if you were a customer, then compare the answer with the written policy. Every difference you find is a Moffatt moment waiting to happen.

8. Self-check

  1. What did the tribunal hold in this case? Answer: That the chatbot was part of Air Canada's own website, that Air Canada was liable for negligent misrepresentation, and it ordered $812.02 paid within 14 days.
  2. How does $812.02 break down? Answer: $650.88 fare difference ($1,630.36 paid minus $979.48 due) plus $36.14 pre-judgment interest plus $125 tribunal fees.
  3. Was the failure that the chatbot spoke at all? Answer: No - on the decision's facts it contradicted the airline's own page, and the disagreement went uncorrected until a customer relied on it.
  4. In the candle-shop example, what produced the wrong promises? Answer: Two contradicting policy documents the tool could read, with no audit of which one it used.
  5. Which of your own policies currently lives in more than one place?

Sources

  • Moffatt v.
  • Air Canada, 2024 BCCRT 149, decided 14 February 2024; all quotes and figures from the decision itself (paras 15, 17, 27, 32, 34, 40, 42-44), read 3 Nov 2026.
  • Wick and Ember, the 40-chats week, the 9 wrong answers and the $24 figure are invented, illustrative and script-checked, not a reported story.
  • Review flags: "its words are your business's words" is the tribunal's finding in one Canadian small-claims case, presented as our operator habit, not a universal legal rule; the habit trio is ours.
  • Go deeper, free: Harvard's session 8, Risks of Generative AI, covers four categories of risk and a healthcare case study. Watch it here.

Review flags

  • "its words are your business's words" is the tribunal's finding in one Canadian small-claims case, presented as our operator habit, not a universal legal rule; the habit trio is ours.

Claim ledger

Claim or exampleBasisChecked
Canadian tribunal, 2024 BCCRT 149; negligent misrepresentation in this case.Moffatt decision3 Nov 2026
$650.88 damages + $36.14 interest + $125 fees = $812.02; order within 14 days.Moffatt decision3 Nov 2026

Independence

Independently developed by Commerce Stories, not affiliated with Harvard. We link to the archived 2024 course; we do not reproduce its videos, slides or exercises. The linked videos were not watched for this written-material review.