Customer service: what an AI agent handles alone, and what it passes on
Routine requests, order tracking, escalation, complaints: where an AI agent helps your customer service, and where a person must stay in charge.
Customer service gathers the questions that come round every day: where is my order, how do I maintain this product, what does the warranty cover. It is natural ground for an AI agent — provided the line between what it handles alone and what it passes on is clearly drawn.
What an agent can handle alone
- Frequent questions, from your knowledge base: maintenance, warranty, opening hours, terms.
- Order and delivery tracking: querying the carrier, giving the status, warning of a delay.
- Asking for missing information: order number, photo, product reference.
- Qualifying and routing: quote, complaint, technical question, to the right person.
Example: a manufacturer receives the same questions every day about the maintenance and warranty of its products. The agent answers from the knowledge base, asks for missing information, and hands over to the technician when the fault is outside its framework.
What it must pass on
- An unhappy customer: the agent detects the tone and the number of follow-ups, and passes the case to the account manager with a summary of the history.
- A request outside its scope, or an answer it is not sure of.
- A delicate complaint, a situation of distress: a person takes over.
- Any commitment: a refund outside the rules, a commercial gesture beyond the defined framework.
The rules of the game
Saying it is an AI
Since 2 August 2026, the European AI regulation requires people to be told when they are dealing with an AI. It is also a matter of trust: a customer who finds out afterwards feels deceived. See AI Act: what changes for SMEs and mid-sized companies.
Answering from your sources, not from memory
A customer service agent must only answer from your information: your terms, your products, your lead times. Outside these sources, it passes on. It is the best remedy against invented answers. See Can an AI agent make mistakes?
Keeping a person visible
A customer must always be able to reach a person. The agent absorbs the volume; it does not close the door.
What you need to get started
- A knowledge base, even an imperfect one: frequent questions, terms and conditions, product sheets.
- Access to the mailbox or support tool, and to order tracking.
- Hand-over rules: to whom, in which cases, with what information.
- A sample of real requests, including the most difficult ones.
A process, or a whole domain?
Handling routine requests is a process: a Process Agent takes it on. When it is the whole department — routine requests, escalation, order tracking, complaints, appointments —, it is a domain: that of Business Agents, up to five processes, with the domain’s knowledge written down for the agents. The complete domain: Customer relations and support.
Frequently asked questions
Will my customers know they are dealing with an AI?
Yes, they must know: it has been an obligation under the European AI regulation since 2 August 2026. We build it in by default in everything that interacts with people.
What happens when the agent cannot answer?
It says so, and passes the request on to the right person with a summary of the exchange, instead of inventing an answer. The hand-over rules are written during scoping.
Can the agent handle complaints?
It can qualify them, reconstruct the history and prepare a reply, or even a commercial gesture within a defined framework. The reply is reviewed and approved by a person: a delicate complaint needs a human.