Threll.ai
Liability and binding agreements

The checklist before your agent says something that binds you

"The AI said it, not us" has been tried as a defence, and it failed. Six questions that should have an answer before your voice agent speaks about price, dates or coverage.

5 min read
A woman in her forties sits on a low stool in a glass-walled phone booth, having just set her mobile down on the shelf with her other hand resting on a printed sheet, seen through the open glass door as afternoon light through venetian blinds lays bands across the floor

An airline tried that defence, and the outcome is worth knowing.

A customer asked Air Canada's chatbot whether he could apply for a bereavement discount after flying. The bot said yes. No such policy existed. When the customer took the matter to British Columbia's civil resolution tribunal, the airline argued that the chatbot should be treated as a separate entity, responsible for its own statements. The tribunal rejected that: the company answers for the information on its own surfaces, whether it comes from a static page or from a bot. The customer was awarded the difference.

The case is Canadian, it concerns text rather than speech, and it binds no Nordic court. But it states a principle Nordic law arrives at from another direction: a trader answers for the information its business gives its customers. A voice agent is not a third party to blame. It is you, on the phone.

That doesn't make the agent dangerous. It only means someone has to have decided what it may speak to — before it answers the first call, not after the first complaint.

Six questions

The question Why it deserves an answer
May the agent quote a price? A figure given on the phone is something the caller will act on. If your prices vary, the caveats belong in the script, not in the head of whoever wrote it
May it promise a date or a delivery time? "We'll be there tomorrow" is easy to say and hard to take back. Decide whether the agent gives times at all, or only says that someone will call with one
May it say what's covered? Warranty, insurance, service, complaints. An approximate answer here is worse than none, because the caller hears a promise
May it say yes to a claim? Refunds, compensation, exceptions to terms. The line should be absolute and low: the agent records the claim, it does not grant it
What happens when it doesn't know? With nowhere to go, the agent settles for a plausible answer. It needs permission to say it doesn't know, and somewhere to transfer to
Who reads what it actually said? None of the above holds if nobody looks. Ten calls a month, read by someone who knows the terms

Most companies answer the first two rows themselves. The next three are the ones usually left open. The last one decides whether the other five mean anything.

What is your agent allowed to promise?

Fifteen minutes in which we work through the six questions against your own setup, and write down what the agent must not speak to.

Book Meeting (opens in a new window)+47 24 12 66 30

15 minutes · no obligation · pick your own time

A recording of "yes" is not written acceptance

If your agent calls consumers rather than answering them, this is where the ground shifts — and it shifts differently in each Nordic market.

Norway's angrerettlov § 10 says an agreement arising from an unsolicited sales call is not binding until the consumer has received the offer in writing on a durable medium after the call and has accepted it in writing; the consumer authority is explicit that an audio recording does not satisfy that. Sweden's distance contracts act has carried the same written-acceptance requirement for telephone sales since 2018 — an oral agreement is simply invalid. Denmark goes further and bans unsolicited sales calls to consumers outright, with a short list of named exceptions.

For an outbound deployment the consequence is the same in all three: the agent cannot close in the conversation. What it can do is establish interest and trigger the written offer. In Denmark the prior question isn't acceptance at all — it is whether you were allowed to dial. And the number you dial from is a separate regulatory question in the same exercise.

A wrong price is not just a misunderstanding

Nordic marketing law follows the same European rule: commercial practice must not give false information or otherwise mislead the consumer, including about price, where that is liable to affect a purchasing decision they would not otherwise have made.

An agent quoting a price that isn't right fits that description rather precisely. The point isn't that the agent is a special risk — an employee saying the same thing does exactly the same thing. The point is that the agent says it identically to every caller, every day, until somebody looks.

Some sectors have had the line drawn for them. At an estate agency, the one answer the agent must never give is written into the rules. Most others have to write the list themselves.

Whoever doesn't know must be allowed to say so

Every question above ends in the same place: the agent needs somewhere to go when the answer isn't there.

That is why escalation is cheaper to take seriously than to postpone. What triggers the transfer, and what the human is told, is the same decision that determines whether the agent guesses. And the numbers that look best on a report don't show whether it guessed — an agent that answered confidently and wrongly looks, in the statistics, like an agent that resolved the case.

Threll.ai builds voice agents in Norwegian, Swedish and Danish. The question is rarely whether the agent can answer. It is what you have said when it does.

Related Articles

Continue reading more articles

See all articles