By Mo Kahlain
In short
- HBR argues that customers increasingly experience the chatbot as the brand. For a smaller company, that makes the chatbot an operations decision.
- A British Columbia tribunal ordered Air Canada to pay $812.02 after its chatbot gave a customer wrong information about a fare.
- Before launch, we would pick the questions customers really ask, name who owns each answer, and write what the chatbot may never promise.
Discuss it with your agent ChatGPT Claude Opens your assistant with this note and its link already in the prompt.
Share Email it LinkedIn Post on X PDF
Julian De Freitas teaches marketing at Harvard Business School. In the November-December issue of Harvard Business Review, he describes a room of executives. He showed them data on how many people say they feel emotionally closer to a chatbot than to their friends. Mouths fell open, he writes.
His article is what started this note. His argument is that customers increasingly experience the chatbot as the brand itself. So managing the relationship is moving from a marketing challenge to a product design challenge.
We read that from a different chair. Most of the leaders we talk to run companies of 5 to 250 people, and they have no product design team. They have a quote from a vendor and a chat window that could go live next month. For them, we think the chatbot is first an operations question: what can this thing say on our behalf, and who answers for it?
A tribunal has already answered part of that
In February 2024, a tribunal in British Columbia ruled on a small case that many owners should read. Jake Moffatt asked Air Canada's website chatbot about bereavement fares after a death in the family. The chatbot told him he could apply for the reduced fare after the trip. The airline's policy said otherwise, and it refused the refund.
Air Canada argued that the chatbot was "a separate legal entity responsible for its own actions." The tribunal member, Christopher Rivers, rejected that. "It should be obvious to Air Canada that it is responsible for all the information on its website," he wrote. "It makes no difference whether the information comes from a static page or a chatbot." The order came to $812.02.
The amount is small, and this was one small-claims tribunal in one province, so we would not read it as settled law. Ask your lawyer about your own case. We keep it for a plainer reason: the customer did what we all do. He asked the thing in front of him and believed the answer.
Your chatbot repeats what your company knows
A chatbot gets its answers from what you give it. In most companies we have seen, the real answer to a hard customer question sits in someone's head. The refund rule has an exception that one employee knows about. The delivery promise depends on the season. The price list on the shared drive is from last spring.
A person at the front desk handles that gap every day. She checks, she asks a colleague, she says "let me get back to you." A chatbot trained on your documents has only the documents. Where they are old or silent, it can still produce a confident sentence, and your customer reads that sentence as your word.
So the work comes before the vendor. You can't automate what you don't understand, and customer answers are a good place to find out what you don't understand.
What we would decide before signing
We would start with a list of the 20 questions customers really ask, taken from your inbox and your phone notes. The vendor's demo script stays aside. For each one, we would write the correct answer and the name of the person who owns it. This is a Map exercise, and it usually takes days. The first surprise tends to be how many answers depend on who you ask.
Next comes the short list of things the chatbot may never promise on its own. Refunds, delivery dates, prices outside the list and anything to do with a contract are the usual candidates. It can explain these subjects and hand them to a person.
Then we would decide how that handover works, because this is where the customer judges you. They should reach a named person or a real time window, with the conversation attached, so they never have to repeat it.
Last, someone in the company reads the conversations every week. It does not need to be a big job. That person owns the list from the first step and corrects it, because your prices and your rules will change after launch.
What we don't know
We do not have a number for how much a good chatbot helps a company your size. We are wary of the ones we see quoted. De Freitas's finding on emotional closeness comes from his research. We have not seen it tested on people buying from a local distributor or a clinic.
What we can say is narrower. A tribunal has already treated a chatbot as the company behind it. If your customers do the same, its answers deserve the care you give a contract or a price list. An owner can decide that this week, before any quote is signed.
Get the next field note by email. Subscribe to The Signal









