HomeAI customer service
AI customer service and automation for chat and email Answers that are right. And a person when it matters.
Anyone who wants to automate customer service does not need a chatbot that answers everything alone. Our AI agents take over the recurring questions by chat and email from your knowledge, look up order and contract data live and pass anything sensitive on to your team.
Free, no strings attached. If you like, with a written assessment of your project.
- 01Question by chat or emailaround the clock
- 02Answer from your knowledgewith a source, clearly an AI
- 03Orders and contracts liveonce identity is confirmed
- Sensitive cases to your teamwith all the details
Automating customer service: does this sound familiar?
- The same questions arrive every day: delivery status, invoice, returns, product data, opening hours.
- Knowledge is scattered across manuals, emails and people's heads, and new colleagues need months to master it.
- Enquiries outside office hours wait until the next day.
- An earlier chatbot gave wrong answers and annoyed customers.
- Questions about data protection for AI and the AI Act slow down every decision.
A good AI customer service does not solve this by answering everything automatically. It answers what it can answer with certainty, and prepares the rest so that your team can decide quickly.
What an AI customer support agent must be able to do.
A chatbot that answers quickly is easy to build. An agent that your team and your customers trust needs these eight things.
-
Answers only from your knowledge, with sources
FAQs, manuals, policies, product data and earlier answers form the knowledge base. Every answer names its source. If the agent does not know something, it says so and hands over instead of guessing.
-
Live data from your shop and ERP system
Order status, shipment tracking, contract and invoice data are fetched by the agent from your systems through interfaces, and only once it is clear who it is talking to.
-
Actions with limits
Change the address before dispatch, create a return, open a ticket: the agent may do what you allow. Refunds and exceptions are possible only within a set limit and with approval from a person.
-
Handover to people, with context
Your team receives a summary, the history, the data found and a suggested reply. Your customers can ask for a person at any time.
-
Web, email and WhatsApp
Chat on your website, the email inbox and WhatsApp share the same knowledge and the same rules. The agent replies in the language of the enquiry.
-
Measurement and test sets
Before launch we test the agent on real past tickets. In operation we measure how many enquiries it resolves on its own, how often it hands over, which answers your team corrects and how customers rate the answers.
-
Costs under control
Upper limits per conversation and per month, small models for simple questions, larger ones only where needed. You see the cost per enquiry.
-
GDPR and operation in the EU
The agent runs on servers in the EU, identifies itself as AI at the start and stores data only for as long as you decide. You receive a data processing agreement.
Channels in detail
The agent shares the same knowledge and the same rules across all channels. Each channel has its own particularities, which we describe in guides of their own.
-
Email
Incoming emails are sorted and passed to your team with a draft reply. This is the safest way to start. More: Answer customer enquiries and emails with AI.
-
Chat on the website
The chat answers questions straight away and hands over to your team when needed. More: Website chatbot.
-
WhatsApp
Many customers prefer to write there. WhatsApp has its own rules for templates and consent. More: WhatsApp chatbot for business.
-
Telephone
A phone assistant answers calls and forwards them. We advise starting with chat and email and adding the phone afterwards. More: AI receptionist for small businesses.
Which programs the agent works with
Connect, do not replace: the agent works with what you have. You do not need a new ticket system or a new shop for it.
-
Helpdesk and ticket system
The agent creates tickets, adds summary and history and classifies them. Your ticket system remains the system your team works in.
-
Shop, ERP and shipping
The agent reads order status, delivery data and invoices through the interfaces of shop, merchandise management and carriers. What it may change, you decide.
-
Mailbox and phone system
The customer service mailbox is connected, and replies are created as drafts. For the phone system we set up the handover to your team.
-
Knowledge sources
We read FAQs, manuals and earlier replies where they are: in folders, a knowledge base or the helpdesk. You do not have to move anything.
How we introduce AI customer care
-
Step 1: Your message
You tell us which enquiries cost your team the most time. We get in touch with you personally.
What you do: write to us. There is nothing you need to prepare.
-
Step 2: Looking at enquiries and knowledge
We look at a selection of your past enquiries and your existing knowledge and sort the topics: what the agent can answer on its own and what a person must decide. After that you receive a written proposal.
What you do: show examples.
-
Step 3: Build and test
We build the agent and test it on your past tickets before any customer sees it.
What you do: look at the test results.
-
Step 4: Draft mode with your team
The agent writes draft replies, your team reviews, corrects and sends them. Your customers never receive anything unchecked.
What you do: correct and give feedback.
-
Step 5: Releasing step by step
The agent answers simple topics such as delivery status or opening hours itself once it has proven itself in testing. More topics follow when the figures are right. You decide how far the automation goes.
What you do: decide what should run automatically.
Approval levels: what the agent does alone and what it does not
- Level 1: draft. The agent writes, your team checks, corrects and sends. Every topic starts here.
- Level 2: on its own with spot checks. Once a topic has proved itself in testing, such as delivery status, the agent answers by itself. Your team keeps checking a sample and sees the measurements.
- Level 3: always a person. Refunds above your limit, complaints, cancellations, legal questions and anything the agent cannot answer with confidence go to your team.
Data protection and the AI Act.
- Operation in the EU. Wherever possible, the agent, data and models run on servers in the EU, ours at AWS in Frankfurt. You receive a data processing agreement. More under AI and data protection.
- Recognisable as AI. The agent identifies itself as AI at the start of the conversation. The AI Act has required this since August 2026 (Art. 50). More under transparency obligations.
- Sparing with data. The agent sees only the data it needs for the enquiry. We define retention period and deletion with you, and if there is a works council, we involve it from the start.
Frequently asked questions about AI in customer service
How do you prevent wrong answers in AI customer service?
The agent answers only from your knowledge base and names the source. If it finds no reliable passage, it hands over to a person. Before launch we test it on your past tickets, and in operation we measure the answers continuously and evaluate your team's corrections. Nobody can promise an error-free answer, but errors can be limited, detected and fixed.
How many enquiries can an AI agent resolve on its own?
That depends on your topics. For delivery status and opening hours the share is high, for complaints with exceptions it is low. A figure in advance would be a guess. We measure it on your past enquiries before launch and tell you what is realistic.
Does it work with our ticketing system and shop?
In most cases yes, provided your ticketing system, shop or ERP system has an interface (API). We check this before you decide.
Can the customer ask for a person?
Yes, at any time. The person takes over with a summary of the conversation so far, so nobody has to explain everything again.
Can the agent issue refunds or make changes?
Only within the limits you set. Small changes such as an address or a delivery slot are common. Refunds and exceptions are possible up to a set amount and with approval from a person.
In which languages does the agent reply?
In the language of the enquiry, even if your knowledge exists in another. Which languages you offer we decide together, and we test them on examples before launch.
Is AI in customer service GDPR compliant?
That depends on the set-up. We build so that data stays in the EU, the agent sees only what it needs, storage and deletion are regulated and you have a data processing agreement. This does not replace legal advice.
Do we have to tell customers that they are writing with an AI?
Yes. Anyone who interacts directly with an AI system must be told that it is AI, unless that is obvious (Art. 50 AI Act). We build the notice in from the start.
Does the agent replace our team?
No. It takes the recurring questions off the team so that it has time for cases that call for judgement and empathy. Decisions and exceptions stay with your employees.
Do you also build phone assistants?
Yes, but we advise starting with chat and email. On the phone, speech recognition, speed and handover matter even more. Once chat and email run well, we extend the channel on request.
When does automating customer service make sense?
When the same questions come back every day and the answer comes from existing knowledge or system data: delivery status, invoice copy, opening hours, returns. It does not make sense where empathy, negotiation or exceptions are needed.
How can emails in customer service be automated?
The agent reads incoming emails, sorts them and writes a draft reply with data from your systems. Your team checks and sends. Later, simple topics run by themselves once the measurements are right.
What is AI support?
Customer service in which an AI agent answers questions and prepares tasks while people decide and handle exceptions. What this looks like day to day is described in the sections above.
What does AI in customer service cost?
That depends on topics, channels, connected systems and the number of enquiries. Before we begin, you receive a written proposal with a clearly described scope. Without your approval, no costs arise. We limit the running costs of model usage with upper limits per conversation.
Which questions does your customer service answer most often?
Send us a few sentences. We will reply personally, free of charge and without obligation.