Chatbots for Business: Costs and Examples 2026

What does a chatbot for business cost in Austria?

An AI assistant for a company website starts at around €2,900 one-off plus about €79 per month for operation and model costs; a simple rule-based bot sits below that, and a system connected to inventory or booking software sits well above. Those are my own package prices, and they are the reason this article can name figures at all: in my market survey of 24 August 2026, exactly one of 24 Carinthian providers reviewed publishes prices for AI services.

The rest work with non-binding first calls. For a business trying to find out whether the topic is worth pursuing, that is the least useful answer available: you have to sit through three meetings to discover you were thinking in the wrong range. This article therefore names the cost range, the running items and the places where such projects get expensive.

How do a rule-based bot and an AI assistant differ?

A rule-based bot follows a decision tree; an AI assistant reads your content and formulates an answer from it. The distinction sounds technical, but it decides the cost, the maintenance effort and what happens when somebody asks a question nobody anticipated.

For a business with five always identical requests, the rule-based bot is the more honest choice: cheaper, predictable, no model costs. As soon as customers ask in their own words, the calculation flips. A decision tree meant to cover every phrasing turns into a maintenance project that is never finished.

Which running costs actually apply?

Three items that quotes often merge into one figure: operation, model costs and maintaining the knowledge base. Operation is the server share and usually the smallest part. Model costs depend on the number of conversations and the length of the answers; they are the only item that grows with success. Maintaining the knowledge base is the item most often missing: when prices, opening hours or services change, somebody has to feed that in, or the assistant answers wrongly for twelve months.

With me, the monthly fee covers all three: €79 per month at the entry tier, €149 per month with a system integration. Like every monthly fee here, it runs for at least twelve months and is cancellable monthly after that. Anyone comparing two quotes should check this point first: a low one-off price with open per-conversation billing is often more expensive in the second year than a higher fixed price with a capped monthly fee.

What does an assistant cost over three years?

Between roughly €5,700 and just under €10,000, depending on the tier. The calculation is deliberately boring: the one-off price plus 36 monthly fees, with no discounts and no assumed savings.

All amounts are net as a small business, with no VAT. Whether it pays off hangs on a single number, and you know it better than any provider: how many similar inquiries per week somebody currently answers by hand. From around ten recurring cases a week it usually pays off, below that rarely.

What does such an assistant look like in practice?

The most honest demonstration is a live one, and it sits on this website. The assistant here knows the services, the packages and the price table; when it names an indicative price, it calculates it with the same function that sits behind the price calculator on the pricing page. It does not invent a figure, it reads the same source as the visible table.

Just as important is what it does not do. It does not enter into a contract, it does not name a binding final price, and it claims nothing about your existing website without having looked at it. If it finds no solid answer, it hands over to the contact form instead of inventing a plausible-sounding one. That handover point is exactly what separates a usable assistant from one that quietly produces errors; it belongs in every quote you collect.

How does this stand with GDPR and data protection?

Inputs run through a data processor under a contract per Art 28 GDPR and are not used to train third-party models. The service in use belongs in your privacy policy by name, together with its legal basis and retention period, not as a vague note about “AI technologies”. An assistant that touches health, legal or HR data belongs on infrastructure inside the EU, and that point is settled before the start, not after the first incident.

Two details are regularly overlooked. First: a chat window is a data collection even when nobody fills in a form, and the retention period of the conversation logs needs a justification. Second: embedding an assistant from a third-party platform usually pulls in its scripts and connections too, and those need consent. How to solve that cleanly is covered in the article AI chatbot and GDPR.

Why do chatbot projects fail?

Because of a missing use case, almost never because of the technology. A project meant to improve “customer communication” has no checkable endpoint and therefore no fixed price. A project meant to answer appointment questions outside opening hours has both. The second common reason is a thin knowledge base: an assistant can only answer what is written down somewhere. If prices, procedures and limits are nowhere in writing, that becomes visible during the project, and writing that content is then the actual work.

The third reason is a quote that talks more about models than about your process. If architectures come up before the question of which inquiry the assistant will take over, it gets expensive without anyone being clear what for. The difference between an embedded assistant and the free language model in a browser tab is covered in the article Chatbot or ChatGPT.

What is the next step?

Keep a tally for one week of which question comes in how often. That list is worth more than any consulting call, because it delivers the use case and the case volume at the same time. Once it exists, it takes minutes to say whether a rule-based bot is enough, an assistant makes sense, or neither does.

The full scope with all three packages is at AI assistant for your website, and chained processes at agent systems with a database. If you first want to know which tasks pay off in a Carinthian business at all, the overview is at AI automation Carinthia; the national view is at AI agency Austria.