ServicesAI chatbot development
AI chatbots that answer customers and staff faster, from your own data
Customers and staff get answers to everyday questions at any hour, and your service team spends its time on the cases that need a person. We build AI chatbots and assistants that answer from your documents and systems, carry out approved actions such as checking an order, and pass a conversation to a person with its context when they should. We build them for customer service, sales support and internal help desks across Europe. Our engineers work from Germany and Poland, in English, German or Polish.
How a conversation is handled
- 1
The question is matched to sources and tools the user may access.
- 2
The answer cites the document or record it comes from.
- 3
Actions such as a return wait for confirmation or follow fixed limits.
- 4
Anything uncertain goes to a person, with the conversation attached.
What changes for your business
Routine questions get answered straight away, and your team handles the ones that need judgement. Questions about delivery times, returns, order status, opening hours or internal IT and HR rules come back every day. A chatbot that answers them from approved sources, and hands everything else over with the conversation attached, shortens the wait for customers and takes the repetition off your staff.
- Answers at any hour to routine questions, from your approved sources
- Fewer repeated questions to your support team and internal experts
- New staff productive sooner, with an internal assistant that knows your processes
- Simple actions done in the chat, such as checking an order or starting a return, after confirmation
- A person always within reach, with the conversation passed on so nobody has to ask again
More detail
Customer contact is where companies use AI most. In Bitkom's 2026 survey, 72% of companies that use AI use it in customer contact. How a whole service team works with it, from ticket classification to the AI Act notice, is on our customer service use case page.
AI chatbots we build
We build on large language models, not on decision trees: a search finds the relevant passages in your sources, and the model answers from them. Where a fixed menu serves users better, for example to choose a language or a department, we keep it.
More detail
- Customer service chatbots that answer questions about orders, products and policies, and hand over to your support team
- Internal help desk assistants for IT, HR or operations questions, answering from your own documentation
- Sales and pre-sales assistants that find product details, availability and terms
- Assistants inside your product, so your customers get help where they work
- Voice assistants for phone lines, where the speed of each response matters
When an off-the-shelf chatbot is enough
Many help desk and CRM products now include an AI chatbot. If your answers live in one knowledge base and the chatbot only has to answer, such a product is often the faster choice, and we will tell you so in the first call.
More detail
A custom chatbot is worth building when answers depend on several systems, when the chatbot has to act and not only answer, when data must stay in your environment, or when you need to test and control its behaviour in detail.
Would you like to talk this through for your own case? The first call is free and takes 30 minutes.
Book a 30-minute callAnswers you can check
The main risk of a business chatbot is a confident wrong answer. We reduce it in three ways: every answer cites its source, the chatbot is set up to decline when its sources don't cover a question, and a test set of real questions, including ones it should refuse, runs before every change. No chatbot is right every time; the test set shows how often yours is right on your questions.
More detail
Permissions apply to every answer. A customer sees only their own orders, and an employee gets answers only from documents their role may open. Limits on what the chatbot may do are enforced in the systems behind it, not only in its instructions, because a message or a document can contain text written to steer the model.
Handover to your team
A chatbot that can't hand over frustrates customers. We agree with your support team which topics, customers and situations go to a person, and the chatbot passes the conversation with its context so nobody has to ask again. Your team also sees which questions the chatbot could not answer, which shows where the documentation needs work.
More detail
Many customers still want to be able to reach a person. In a study by the Capgemini Research Institute published in March 2025, more than 70% of the German consumers surveyed preferred a human.
A hypothetical customer service chatbot
An online retailer's support team answers the same questions about delivery times, returns and order status every day. The chatbot answers policy questions from the returns and shipping pages, looks up order status in the shop system for logged-in customers, and starts a return only after the customer confirms it. Complaints, damaged goods and any refund above an agreed amount go straight to a person.
More detail
The team tracks the share of conversations resolved without handover, the reasons for handovers and customer ratings, next to the cost per conversation.
From pilot to a live chatbot
Every chatbot we launch comes with what your team needs to keep it right after we leave:
More detail
- A test set built from past tickets or chat logs, with expected answers and sources
- A limited launch on one channel or customer group first
- Monitoring of answer quality, handovers and cost per conversation
- A runbook for adding sources, changing what the chatbot may do and investigating a wrong answer
Questions
What does an AI chatbot cost, and what drives the cost?
The cost depends on how many sources and systems it uses, whether it only answers or also acts, how many languages and channels it covers, where it runs and how strict the testing must be. Model usage adds a running cost per conversation, which we measure from the start. Each step has a fixed price, agreed in writing after the first call.
From what company size is an AI chatbot worthwhile?
Size matters less than volume and repetition. If your team answers many similar questions every week from documents and systems, a chatbot can take a share of them. We take a sample of past tickets or chats, check how many the chatbot answers correctly from your sources, and set the cost per conversation against the time your team spends on them today. If an off-the-shelf product in your help desk does the job, we tell you.
Must we tell users that our chatbot is an AI?
In most cases, yes. Article 50 of the AI Act, which applies from 2 August 2026, requires that people are told when they interact with an AI system, unless that is obvious. We build the notice into the chat window and the handover. Whether an exception applies in your case is for your lawyers to decide.
How long does it take to build an AI chatbot?
It depends on how many sources and systems it needs and how strict the review is. You get an estimate after the first call, once we know the use case and the systems involved.
Which language models do you use for chatbots?
The one that does best on your questions at an acceptable cost, from commercial providers your data owners approve or as open-weight models on your own infrastructure. No vendor pays us for choosing theirs.
Can the chatbot work in our website, app or support tool?
Yes. We connect it to the channels you already use, such as your website, app, help desk or internal chat, and to the systems it needs to read from or act in.
Can an AI chatbot meet GDPR requirements?
We build it so your data protection officer can assess it: which data it sees, where it runs, how long conversations are stored and who can read them. Whether it meets GDPR is your advisers' decision, and we prepare what they need for it.
How does a chatbot project start?
With a free 30-minute call about the questions your customers or staff ask most. We need no access to your systems or data for it.
Related services
Data integration for AI
Answers from file shares, SharePoint, ERP and tickets in seconds, for assistants and agents, with permissions carried through and tested.
Knowledge assistants (RAG)
Staff find answers in internal documents in seconds, with sources cited, permissions kept and tests on your own questions.
AI agents and process automation
Agents that take on the routine steps of complex cases and pass decisions to people, with a cost per handled case.
System access for agents (MCP)
AI agents that look up and update data in your systems, with narrow permissions, human approval for changes and tests.
Custom model training
A small open-weight model trained for one task and compared with your current model on cost and accuracy.
On-premise AI
AI on offers, contracts and customer files, with the data staying in your company: own servers or own cloud account, tested before go-live.
AI system testing
Before go-live, acceptance tests on your own cases tell you how often your chatbot, assistant or agent is right, in terms your QA team already uses.
From the blog
RAG citation quality
How to test whether the passages an assistant cites actually support each claim it makes, with a claim-level check, a worked example and acceptance criteria.
Designing agent escalations
When a process agent should hand a case to a person, what the handoff must contain and how to run the queue that receives it.
RAG document permissions
How access rules travel from source systems into a retrieval index, where they get lost, and which negative tests show that a user cannot retrieve what they may not open.
Talk to an engineer
Bring a few questions that customers or staff ask every day, and we will look at which of them a chatbot could take off your team. The call takes 30 minutes and needs no access to your systems or your data.
Booking runs on Google Calendar.
When you book a call, you speak to Mickey Graf
He founded Cloudsail and has put machine learning into production since 2017, including AI engineering at trivago and SAP. The call can be in English, German or Polish.
Prefer to write first? Leave your email and a few lines. Mickey reads every message himself and replies by email with times for a call.
Look for the draft in your email program
Send it from there, and Mickey replies in person with times for a 30-minute call. If no draft has opened, write to us at miki@cloudsail.com.