Yes. Arranging a callback is scheduling, and an agent can do it at depth 3 once it is connected to the team's real calendar or queue. The conditions: it books only slots that exist, calls back only numbers on file or verified, records why the customer wants the call, and promises nothing about what the call will decide.
What decides whether the agent can book it?
Real capacity. "Someone will call you within the hour" is a promise. The agent should book a slot from the scheduling system or place the customer in a callback queue with the queue's own estimate, never invent a time.
A known number. Calling back the number on the account is safe. A new number given in a chat needs verification, and the person making the call should verify the customer again before discussing anything.
A reason on file. The person who calls should not start from nothing. The agent records the question, what it already told the customer, and why it could not resolve it.
The right team. A callback about a complaint, a hardship request or suspected fraud goes to the team trained for it, often with priority.
When is a callback the wrong answer?
When the agent already knows the outcome. If a request is outside policy and the person who calls back will say the same thing, booking the call only delays the refusal and raises hopes. A clear refusal with a reason, and an offer of a call if the customer still wants one, serves the customer better; the argument is set out in a refusal with a reason is good service.
The opposite mistake matters too. An agent that never offers a person, even when asked, will frustrate customers who have every right to speak to someone. The test is whether the call can change anything for the customer.
What may the agent do, and what stays with a person?
| Step | The agent | A person |
|---|---|---|
| Offer a callback | Offers it when a person is needed or asked for | None |
| Choose the slot | Books from real availability | Adds capacity |
| Confirm the number | Uses the number on file, or verifies a new one | Re-verifies the customer on the call |
| Brief the caller | Writes the reason, history and what was said | Reads it before calling |
| Set priority | Applies the routing rules | Overrides priority |
| Promise an outcome | Never | Decides on the call |
| Rearrange or cancel | Moves the slot at the customer's request | Handles repeated no-answers |
What does this look like in practice?
Take a consumer lender. A customer asks the agent to remove a late fee because their salary was paid late. The agent's role scope does not include waiving fees, so it says plainly that it cannot remove the fee and why, and that the collections team can review the request. The customer asks to speak to someone.
The agent books the next collections slot, confirms the mobile number on the account, and writes a brief: fee date and amount, the customer's reason, and that this is the first late payment in two years. It does not say the fee will probably be waived. Whether it should be is a judgement for the team, as covered in whether an AI agent can waive a late fee. If the customer had mentioned that they could not pay at all, the agent would have marked the callback as a possible hardship case and routed it accordingly.
Which controls make it safe?
- Booking only through the scheduling system, so every slot is real.
- Callbacks to the number on file by default; new numbers verified first.
- A required brief for every callback, filed with the conversation.
- A blocked list of phrases that predict outcomes ("they'll sort that for you").
- Monitoring of missed callbacks, since a broken promise of a call is a complaint in waiting.
The principle that the agent says no clearly, and hands over well, is the core of the guide to AI customer support agents.
Where Pryme fits
In Pryme, booking a callback is its own capability with its own certificate, separate from any decision the call might lead to. An agent can be allowed to book calls at depth 3 while fee waivers and retention discounts stay refused or held.
Questions people ask
Can the agent make the callback itself?
Some agents can call customers by voice. That is a different capability, needing AI disclosure, identity checks on the call and its own certification.
What if the customer misses the callback?
The agent can offer a new slot by message. After a set number of missed calls, a person decides whether to close the request.
Should the agent tell the customer who will call?
It can name the team. Naming an individual risks a broken promise if rotas change.