An Airline Blamed Its Own Chatbot in Court. It Lost.

In 2024 a man named Jake booked a flight to his grandmother's funeral. Before booking, he asked Air Canada's chatbot whether he could claim a bereavement discount. The chatbot said yes, confidently and in detail. It was wrong. No such policy existed in the form the chatbot described. Air Canada refused the refund, and when it reached a tribunal the airline argued something remarkable: that the chatbot was a separate entity, responsible for its own actions. The tribunal disagreed. It is your chatbot. You are responsible for what it says. Nothing crashed. No error appeared. The automation ran exactly as built, sounded completely reasonable, and cost the company money and a headline. That is what this video is about: not automations that break, but automations that work perfectly and do the wrong thing. Most businesses watch the demo and say yes. The testing is the part that actually decides whether this works. Got a question about any of this? Put it in the comments. I read and reply to all of them. CHAPTERS 0:00 The chatbot that promised a discount that did not exist 0:32 Jake won. The airline paid. 0:56 Inside a real make.com scenario 1:08 Where this workflow can go wrong 2:30 The catch-all that asks a human 2:54 Why every reply goes to drafts 3:26 The Air Canada case in full 3:51 "The chatbot is responsible, not us" 5:05 Giving automation more authority than it earned 6:30 The four ways automations get it wrong 7:14 Why AI fails differently from normal software 7:57 Hidden instructions inside job descriptions 8:28 The ACE framework 8:44 A: adversarial testing 9:18 Vague enquiries, angry customers, contradictions 10:18 C: the circuit breaker 11:17 E: easy escalation, and what to pass on 11:46 Matching safeguards to risk 12:15 Medium risk: review before it publishes 12:30 High risk: money, pricing, discounts 13:07 The better question to ask 13:12 Human in the loop does not mean manual 13:52 The three questions before you automate 14:44 Who I am and what we do Source for the Air Canada case: https://www.bbc.co.uk/travel/article/... --- I am Uwa Ujam, founder of Autore. We build AI systems for small and medium businesses across property, charity, food and beverage, manufacturing and professional services. Our current focus is speed to lead. When someone calls, fills in a form or messages your business, they get a response within 60 seconds, day or night, without you handing your reputation to an automation nobody checked. If this was useful, take the scorecard on the website. It takes about three minutes and tells you how ready your business actually is for AI, or book a call and we will go through it properly. Website: https://autore.ai Email: hello@autore.ai AI Opportunity Scorecard: https://autore.ai/scorecard Slides from this video: https://gamma.app/docs/z476vd97upyqi22 #aiautomation #makecom #aiguardrails #ChatbotFail #businessautomation