How do you automate customer service with AI?

Can you automate customer service with AI?

Yes, you can automate a large part of customer service work with AI. AI can read a question, understand what it is about, look up the right data and write an answer. That works best for questions that come up often and have a fixed answer.

Automating does not mean that no person is involved any more. For most businesses the best set-up is a mix: AI does the groundwork, and an employee decides on everything that is sensitive or complicated. Are questions piling up with you? Then also read customer questions are left unanswered.

Which customer questions can you automate?

Mainly questions that come up often and whose answer is in your own systems. Think of:

• Where is my order? AI looks up the order and gives the status.
• Opening hours, delivery times and terms. The answer is on your site or in your manual.
• Making or changing an appointment. AI offers a slot from your calendar.
• Returns or exchanges. AI explains the steps and creates a return.
• Sending a question to the right person. AI reads the email and assigns it to the right colleague.

Less suitable are complaints, negotiations and questions where someone is angry or upset. In those cases a customer wants to speak to a person.

Through which channels can you automate customer questions?

Through any channel where questions come in as text. The most commonly used are:

• Email. The easiest start. AI reads the email and prepares a draft reply in the mailbox.
• Web form. The question already comes in through fixed fields, so AI can sort it well.
• Chat on your website. Here the customer talks directly to a chatbot. In that case, pay attention to the notice that it is AI.
• WhatsApp. Many customers prefer sending a message to sending an email. You can have those messages sorted and answered too.
• Phone. That is possible with speech technology, but it is the most difficult channel. Do not start there.

It helps if all channels come together in one overview. Then your team sees in one list what AI has prepared, wherever the question came from.

Chatbot, draft reply or AI agent: what fits?

There are three ways, from cautious to far-reaching. Which one fits depends on how much control you want to keep over each answer.

1. Draft reply. AI reads the question and prepares an answer in your email or customer service system. An employee reads it, adjusts it and sends it. This is the safest start.
2. Chatbot. A chatbot on your website answers simple questions itself. If it does not know, it passes the question to a person.
3. AI agent. An AI agent does more than answer: it can also arrange something, such as scheduling an appointment or creating a return, within rules that you set.

My advice is to start with draft replies. You then see exactly what AI does well, before you let it answer on its own.

How does AI know the right answer?

AI knows the right answer when it is given your information, not just what it knows itself. A language model does not know your orders, prices and terms. That is why you connect it to your own sources: your webshop, your customer system and your manuals.

A widely used technique for this is called RAG. AI first searches for the relevant pieces in your information and writes the answer based on them. That way the answer stays close to what is in your own information.

Even then AI can make mistakes. It can make up a detail that looks convincing, a hallucination. With draft replies, the employee catches that. If you let AI answer on its own, limit it to questions whose answer you can check well.

Do you have to say that a customer is talking to AI?

Yes. Since 2 August 2026 the transparency rule from the European AI Act, Article 50, has applied. If someone talks directly to an AI system, such as a chatbot, it must be clear that it is AI. There is an exception if that is already obvious to an attentive user, but do not count on that too quickly.

In practice this means: put a short sentence at the top of the chat, such as "You are talking to a digital assistant". With draft replies that an employee checks and sends, the customer is not talking directly to AI. You can read more about the rules under the EU AI Act.

This is an explanation in plain language, not legal advice.

What remains human work?

Everything that needs judgement or empathy. Specifically:

• Complaints and disappointed customers.
• Exceptions, such as a goodwill arrangement or a non-standard delivery.
• Decisions about money, such as a refund outside the rules.
• Checking what AI prepares, especially at the start.

Also make sure a customer can reach a person at any time. A chatbot that keeps going round in circles annoys customers and actually creates more work.

How do you start with AI in your customer service?

Start small and measure what happens. A good order is:

1. List which questions come in most often, and where the answer is.
2. Choose one type of question, for example the status of an order.
3. Have AI create draft replies for it, and watch along for a few weeks.
4. If the drafts are correct, let AI answer those questions itself, with a clear notice.
5. Then choose the next type of question.

Pay attention to privacy as well. Customer questions contain names, addresses and sometimes sensitive information. Work with business accounts and make agreements in a data processing agreement. You can read more about the rules under the GDPR.

How do you know whether it works?

By keeping track of a few things, from the first week. For example:

• How many draft replies an employee sends unchanged, and how many they have to adjust.
• How long a customer waits for an answer, before and after the start.
• How often a customer asks for a person in the chat.
• Which questions AI does not answer well. Information is often missing there, which you can add.

That way you decide which questions AI may answer itself based on what you see, and not on gut feeling.

What does automating customer service cost?

That depends on how many questions you have, how many types of question there are and which systems are involved. If AI also has to look up data, such as the status of an order, an integration is added. With me it starts with a business analysis. In it I look at which questions come up most and where the answers are. Then you get a proposal with a fixed price, so you know in advance where you stand. You can read more about what I do with AI on AI and automation. Also read how your price is put together.

Frequently asked questions

Sources: Regulation (EU) 2024/1689 (AI Act), Article 50 and Dutch Data Protection Authority (Autoriteit Persoonsgegevens): Data processing agreement (in Dutch). Consulted on 19 October 2026.