An AI chatbot for your business: what it does and what it costs

An ordinary chatbot vs a RAG assistant

A classic chatbot wired straight into ChatGPT sounds clever, but it knows nothing about your company - and when it does not know, it makes something up (a hallucination). In a business setting that is a problem.

The technique is called RAG (retrieval-augmented generation). Put simply: the assistant first finds the answer in your documents (policies, price lists, manuals, emails) and only then answers - adding to that answer a link to the source. It does not answer off the top of its head, it answers from your data.

An assistant that answers only from your documents and cites the source - not generic AI mush.

When it pays off

An AI assistant makes sense when people at your company keep looking up the same information over and over:

• Internal knowledge - "how much holiday do I have left?", "what is the returns procedure?" - employees ask in plain language and get an answer from company documents.
• Customer support - answers to frequent questions from your documentation, 24/7.
• Onboarding - a new hire asks the assistant instead of their colleagues.

I built exactly that kind of assistant - it is called Robin. Employees ask in plain language and get answers from company documents with a link to the source.

What an AI chatbot costs (2026)

The price is set not by the chatbot itself but by the scope around it - how many systems it has to connect (website, email, CRM, bookings) and how cleverly it has to respond. In Czechia, a one-off deployment of a simpler company assistant runs in the region of from 35,000 - 45,000 CZK (roughly €1,400 - €1,800) at agencies; a more complex RAG with a vector database goes higher.

I build AI apps myself, so there is no agency overhead:

On top of the one-off build there is running costs - the model API (Claude, GPT) and hosting are billed by usage. I always say so up front so you know what you are getting into. Everything I can build is on the page Custom AI apps.

How to spot a good supplier

• It asks first about your problem, not about the technology.
• They can explain where their answers come from and how they guard against hallucinations.
• They tell you running costs up front, not on the invoice.
• You get a prototype you can poke at, not just a slide deck.