Plainly and accurately. When a customer asks "did a computer decide this?", staff should say what the agent did, what a person decided, why the outcome is what it is, and how the customer can ask a person to look again. They should never hide the agent, and never blame it. The record gives them the facts to do both.
What does the customer actually want to know?
The question "was this decided by AI?" usually stands in for three others: was I treated fairly, can someone explain it, and can I get it changed? A good answer addresses all three, briefly. A defensive answer, "it's just the system", fails all three and usually leads to a complaint.
Staff can only answer well if they know how the agent works in their business: what it does alone, what it drafts for a person, and what it is never allowed to do. That is a training point, not a script point.
What should staff say in each situation?
| What happened | What staff can say | What to avoid |
|---|---|---|
| The agent completed a routine action alone | "Our system handled that automatically under our standard rules. Here is what it did and why. If something looks wrong, I can review it now." | "The computer did it, nothing I can do." |
| The agent drafted, a person approved | "Our system prepared it and a member of our team checked and approved it. The reason was…" | Claiming no automation was involved |
| The agent held it and it is waiting | "This needs a person to approve it. It is with our team and should be decided by…" | Promising an outcome before the person decides |
| The agent refused | "This request falls outside what our system can do, so it has come to me." | "The system says no." |
| A person declined it | "A member of our team reviewed it and decided not to go ahead, because…" | Blaming the agent for a person's decision |
Two rules sit behind the table. First, be honest about involvement. People must be told when they are interacting with an AI system, and customers increasingly ask. Second, own the outcome. The agent works for the business; the business is responsible for what it does.
How do staff find the reason?
The decision record shows the verdict, the facts the agent used, and which bound decided. A trained member of staff can read it in a minute and translate it: "The refund was held because it is above the amount our system can approve on its own, so it is waiting for a manager." That is accurate, specific and reassuring. Explaining the reason for a particular decision is covered in more depth in how to explain an AI agent's decision to a customer.
Where a decision based solely on automated processing has legal or similarly significant effects, GDPR's Article 22 is relevant, and customers may be entitled to have a person review it. Staff do not need to quote the law. They need to know how to put a decision in front of a person with authority, and how quickly.
An example: a declined refund
Take an online retailer. A customer calls, upset that a refund was refused, and asks whether "a bot" decided it. The member of staff opens the record. The agent drafted a refusal because the item was returned outside the return window, and held it; a team leader approved the refusal with the reason "outside 30-day window, no fault reported".
The staff member explains: "Our system prepared the decision and a team leader checked and approved it, because the return arrived after the 30-day window." The customer says the item was faulty. That is new information. The staff member says they will ask the team leader to look again on that basis, and logs it. The customer leaves knowing who decided, why, and what happens next. Market detail is in AI agent training for customer support agents in e-commerce in Mexico.
Which controls support good explanations?
- Staff can read the record for any decision about the customer they are speaking to.
- A route exists to put a decision before a person with authority, with a time standard.
- Explanations given are logged with the call.
- The review queue is staffed to meet that time standard; see how to staff an AI agent review queue.
The full role map is in AI agent skills by role.
Where the Academy fits
Pryme Intelligence Academy's Foundations track, about four hours and free, gives every customer-facing member of staff the vocabulary to explain an agent accurately: the decision chain, depth, and what held, refused and withheld mean. Staff who also approve held items go on to the Operator track, which teaches them to read the record and decline with a reason.
Questions people ask
Should staff volunteer that an agent was involved?
If the customer is interacting with the agent directly, they should be told. When staff explain a decision, mentioning the agent's part is honest and usually helps.
What if staff do not know why the agent decided something?
They should say they will find out, open the record or escalate, and come back. Guessing is worse than a short delay.
Can a customer insist on a person deciding?
They can ask, and in some situations rules give them a right to human review. The business should have a clear route either way.