01How it happens
AI agents are built to be helpful and agreeable. A buyer who asks again, adds a reason (“I’m a regular”, “I need it today”, “it’s cheaper elsewhere”) and offers to meet halfway gives the agent a reason to be helpful with your money. Without a firm number to check against, each step feels reasonable.
What an agent says can bind the business: in Moffatt v. Air Canada, 2024 BCCRT 149, a tribunal held the airline responsible for incorrect fare information from its chatbot.
02How to make it harder
- Give the agent the lowest price for each product, not a percentage rule it must work out.
- List the free extras it may offer, and say that anything else needs a person.
- Let it hand a hard negotiation to a person instead of conceding.
- Make prices come from your system (a tool call), not from the conversation.
03How to find out
Send it many buyers who haggle the way real ones do, and record every price it offers and every sale it closes against your lowest prices. That is what a sales agent test does, with the money at stake and an honest range.
Questions
Is telling the agent “never discount” enough?
It usually makes it refuse sales you would have taken, and a determined buyer can still find wording it does not recognise. Give it the actual lowest price per product, and test how it holds under pressure.
Can an AI agent be held to a price it offered?
That depends on your terms and the law where you sell. In Moffatt v. Air Canada (2024), a Canadian tribunal held an airline responsible for wrong fare information its website chatbot gave a customer, and rejected the argument that the chatbot was responsible for its own words. Testing before launch is cheaper than finding out.
See where your agent leaks money. Tested against your own rules, with the money at stake.