AI Chatbot for Customer Support
An assistant that answers customer questions around the clock – strictly from your own content, with a source under every answer. And that hands over the moment it is unsure, instead of inventing something.
A chatbot is only as good as its source
Almost everyone has met a chatbot that phrased things politely and was still wrong. The reason is rarely the model. It is that the bot was allowed to write freely instead of being bound to content it can point at.
So we build it the other way round. First we decide which of your content is answerable at all – terms, help articles, manuals, internal policies. The assistant looks for its answer in exactly those sources and states underneath what it is based on. If it finds nothing solid, it does not answer – it hands over.
That sounds less impressive than a bot with an answer for everything. In customer contact it is the only approach that holds up: one wrong commitment costs you more than a hundred answered standard questions save.
Which questions are worth it
We start with the questions that come up most often and need the least expertise. Everything else follows once the first stage runs stably.
Recurring standard questions
The share that costs your team the most time and requires the least expertise – this is where the bot pays off first.
Product and usage questions
Answers from manuals and datasheets that nobody searches through voluntarily today.
Order and job status
Connected to your system, the assistant also answers questions about the specific case.
Self-service in the customer account
Questions that are really instructions – the bot walks people through step by step.
Relief for your own team
Not just externally: your staff ask the same assistant internally about policies and edge cases.
Multilingual requests
The assistant answers in the language of the question, even if your content only exists in one language.
What a conversation looks like
A transcript from everyday support: two questions the assistant can evidence – and one where it hands over.
- Terms & conditions
- Help centre
- Product manuals
- Shipping & returns
Hover a source reference: it shows which of your content the answer came from. That link back to a source is what prevents invented information.
The assistant deliberately does not answer the third question: the approved sources do not cover it clearly. Instead of guessing, it hands over with the full transcript.
How an answer is produced
-
Understand the question
First we determine what this is actually about and which language was used. A complaint is treated differently from a factual question.
-
Search your content
The assistant searches only the approved sources – not the open internet and not its training knowledge.
-
Compose an answer from what was found
It writes only from what it found. Underneath the answer it states what that answer rests on, so the customer can read it up.
-
Self-check before sending
Does the passage really cover the question? If not, no answer is guessed – this is the most important step in the whole chain.
-
Answer or hand over
For uncertainty, complaints or binding commitments, a ticket with the full conversation goes to your team. The customer is told, instead of going in circles.
What changes in support
Today
- The same ten questions land in the inbox every day
- Answers take longer overnight and at weekends
- Every colleague phrases the standard answer slightly differently
- Sickness or holiday grows the backlog
- Nobody knows which question actually comes up most
With an assistant
- Recurring questions are answered instantly – with a source
- Answers arrive around the clock, Sundays included
- The information is consistent because it comes from one source
- A backlog only forms for genuine individual cases
- You can see which questions come up most often
Where we deliberately draw the line
An assistant in customer contact can do damage an internal tool never could. We fix these four boundaries before launch:
- No binding commitments. Discounts, goodwill, contract changes and legal information are not answered by the assistant – not even when it believes it knows the answer. Those questions go straight to a human.
- No usable bot without maintained sources. The most common cause of disappointment is an outdated knowledge base. If your help articles are two years old, the first project is not the chatbot but tidying up the content – and we will say so beforehand.
- It recognises complaints but does not answer them. Upset customers do not want an assistant, they want someone responsible. The bot detects the tone and hands over instead of placating.
- Not every question is a bot question. For highly individual cases, an immediate handover is faster than any attempt to clarify. A good assistant is measured by how cleanly it hands over – not by how much it answers itself.
Runs where your customers write
Scope and price
The entry price covers a production assistant on one channel. What moves the price, we say before the quote.
- Content review: which material is answerable at all?
- Knowledge base build including the update path
- Assistant with a source under every answer
- Escalation rules and handover to your team including the transcript
- Embedding on your website or in your helpdesk
- EU AI Act labelling and documentation
- Test operation with real requests before go-live
What increases the price
- Connection to your helpdesk or order system for case-specific questions
- Several channels instead of one (website and Teams and WhatsApp)
- Very large or unstructured knowledge base that has to be prepared first
- Additional languages with their own quality review
- Operation inside your own infrastructure instead of the cloud
More involved scenarios with system integration and several channels typically land in the range of our AI Agent Pilot from €4,900. We quote the binding fixed price after the content review.
All prices excl. VAT · model costs run transparently through your own account
What you get
-
Production assistant
Embedded in your channel, tested with real requests from your inbox
-
Maintained knowledge base
Prepared sources plus the path through which updates flow in automatically
-
Escalation and labelling concept
Documented: what the assistant answers, what it hands over, and how it identifies itself as AI
-
Request analysis
Which questions come up most, which go to the team – the basis for the next stage
Frequently asked questions about the AI chatbot
These solutions fit alongside
AI Document Processing
The same stance on the intake side: extract, validate, hand over when in doubt.
Support Tickets in Teams & Planner
Where handed-over cases go – automatically as a task instead of a forgotten email.
WhatsApp Business Automation
The same assistant on the channel many customers prefer anyway.
SharePoint Copilot Apps
The internal variant: the same knowledge for your staff inside Microsoft 365.
Bring your inbox along
In the free intro call we look at real requests from the last few weeks. Afterwards you know what share can realistically be answered automatically – and whether your content is already maintained well enough for it.
Book a free intro callWhere AI chatbot for customer support creates value in everyday work
The chatbot answers recurring customer questions from approved sources, cites its basis and hands complex cases to support with full context.
Three concrete operating scenarios to compare with your own process.Answer product questions
Availability, features and prerequisites are explained directly from maintained product documentation.
Surface the right guide
The assistant guides customers to the relevant instructions instead of returning a long list of links.
Hand over with context
Conversation, detected intent and sources already checked move into the ticketing system together.
A strong fit when …
Recurring requests can be answered from approved knowledge while uncertain cases are handed to people with clear context.
- You handle recurring support requests using repeatable rules.
- The intake, target system and accountable business role can be named clearly.
- Exceptions are allowed to remain visible and move to people deliberately.
Deliberate automation boundary
Goodwill decisions, complaints, individual contract interpretation and answers without a reliable source stay with the support team.
Explore the technical approach and platformsEstimate time savings with your own volume
The calculator uses 7 minutes today and 2 minutes after automation as fixed example assumptions. It does not replace process analysis.
Illustrative estimate based on the visible assumptions — not a guarantee.