AI customer service:faster replieswithout losing customers

AI customer service means answering enquiries faster: the AI drafts replies from your knowledge base, summarises and sorts incoming requests and answers simple questions directly in chat. There are two routes: AI in the background that makes suggestions to your team, and AI in direct contact with customers as a chatbot. If you take the second route, Article 50 of the EU AI Act has required you since 2 August 2026 to make clear that an AI is answering, and you need a reliable way through to a person.

Nikolai Schöbel and Jeremias Burger

Co-founders of Scalableloops GmbH. Nikolai Schöbel leads online marketing and AI strategy, Jeremias Burger the AI architecture. Both build AI systems and train teams on them in their own agency work.

Chatbot or assistant for your team: which fits?

The key difference is who reads the AI’s answer first: your team or your customer.

AI in the background. The AI suggests replies, summarises and searches the knowledge base, and a person sends the message. Zendesk describes a Copilot for service agents that suggests replies while the agent sends the message themselves. General AI chats such as ChatGPT, Claude or Gemini can be used the same way, but only with a business account for customer data. Mistakes stay internal, which makes this the more cautious start.

AI in customer contact. A chatbot or AI agent answers enquiries itself, on your website, by email or on the phone. Examples include Zendesk AI Agents, Fin from Intercom, HubSpot’s Customer Agent and the Service Agent in Salesforce Agentforce. Intercom says Fin works across channels such as chat, email and phone and hands conversations over to your team with the full history. Our article What is an AI agent? explains how an AI agent differs from a simple chatbot.

A rule of thumb: start in the background, track how often the suggestions fit without changes, and only let the AI answer customers directly for the questions where that is reliably the case.

How do you write AI replies that suit the customer?

If your team works with an AI chat or an assistant, the prompt decides the quality. A good prompt states the role, the enquiry, the source and the limits.

For example: “You are replying on behalf of a furniture retailer’s customer service team, friendly and professional. Here is the customer’s enquiry and our text on delivery times. Answer only with information from this text. If the text does not answer the question, say that we will get back to them, and do not make anything up. No more than 100 words.”

For upset customers, a separate prompt helps: “The customer is annoyed because their delivery has been postponed for the second time. Write a reply that takes the frustration seriously, without excuses, and gives the next concrete step from my notes. Do not promise any compensation.”

For summaries, this is often enough: “Summarise this thread in three sentences: what is the request, what has been done so far, what is still open?” That helps with every handover in the team.

For enquiries in other languages, translation services such as DeepL are an option. According to DeepL, texts on DeepL Pro are only stored for as long as technically necessary for the translation and are not used for training. Our guide Use AI for translation goes into more detail.

Who is liable when a chatbot gives wrong information?

A well-known example comes from Canada. Air Canada’s chatbot told a customer he could apply for a reduced bereavement fare after booking. The airline’s policy did not allow that. In February 2024, the Civil Resolution Tribunal of British Columbia ruled that Air Canada was liable for the wrong information and awarded the customer 650.88 Canadian dollars.

The decision is based on Canadian law, but the lesson applies broadly: customers do not distinguish between information from a member of staff and information from a chatbot on your website. What the chatbot says is attributed to your business.

Three rules follow from this. The chatbot answers only from approved content. For money, deadlines and goodwill, it hands over to a person. And someone on your team regularly reviews a sample of conversations so that mistakes are caught before a customer complains.

How do you keep quality high when AI is part of the team?

AI in customer service needs the same attention as a new team member. Decide who approves suggested replies, who maintains the knowledge base and who reads samples of chatbot conversations. Watch for complaints about answers that miss the point, and for how often customers ask for a person.

Article 4 of the EU AI Act has applied since 2 February 2025 to organisations that use AI systems. Since 27 July 2026 it applies in amended form: you have to take measures that support the development of AI literacy, but you do not have to guarantee a specific level for each person. No certificate is required. In customer service, that means the team should know which data may go into the AI, where it tends to get things wrong and when a case belongs with a person. Details are in our article on the AI literacy obligation under Article 4.

Customer service and sales often overlap, for example when an enquiry turns into a new order. Our guide Use AI for sales shows how AI helps there, and for retailers Use AI for ecommerce is worth a look too.