What does an AI chatbot cost?
There is no list price, because “AI chatbot” can mean quite different things: a widget for your website, an internal knowledge assistant, a customer service bot, or a fully self-hosted installation. The price is almost always made up of the same building blocks: a base fee per month or per user, a usage-dependent share, and depending on the hosting model, hardware and setup on top. What decides the bill is less the vendor than how many people use the assistant, how intensively, and where it runs.
Why quotes differ so widely
The biggest difference lies in the billing model. Some vendors charge per user per month – predictable, but expensive once the assistant is rolled out across the organisation. Others charge by consumption, in messages or tokens: cheap to start with, harder to forecast. A third option is a flat rate with an included allowance, where only usage beyond it is billed.
The second large item is the hosting model. A cloud service has no acquisition cost, but the fee runs indefinitely. On-premise shifts that: hardware and setup are one-off, while running costs drop – and beyond a certain level of use the maths reverses. Where exactly that point sits depends on the number of users and on what hardware you already have.
The third block is routinely underestimated: setup and upkeep. An assistant is only as good as the documents it answers from. They have to be connected, kept current and given access rights. It is rarely a large sum, but it is ongoing effort – and it appears in no price comparison.
What makes up the price
A monthly floor, charged per organisation or per user depending on the vendor. Flat-rate models often already include a usage allowance.
Billed by messages or tokens. The range is wide: an assistant that works through long documents consumes a multiple of what a short FAQ bot needs.
Cloud, EU data centre or on-premise. Running it yourself brings hardware and operating costs, but lowers the recurring fee and keeps all data in-house.
One-off: opening up documents, connecting sources, defining roles and access rights. The effort grows with the number of sources and with how disorganised they are.
Ongoing: replacing outdated documents, closing gaps, reviewing analytics. Easily forgotten, and yet it decides the quality of the answers.
From self-service to a named contact. For public bodies and regulated industries, usually the item that decides whether people accept the tool.
Comparing prices across vendors misleads as long as you are not comparing the same thing. A website widget for visitor questions and an internal assistant working across all company knowledge are different products, even though both are called “chatbot”. So do not weigh monthly prices against each other – weigh the expected total cost over a year, including usage, operation and upkeep, for the use case you actually have.
Frequently asked questions
What does an AI chatbot cost per month?
That depends on the use case. An embeddable website chatbot sits well below an internal knowledge assistant for a large workforce. The actual tariffs for KOSMO and NOVA are on the pricing page – where you can also see what usage allowance each one includes.
Is per-user or per-message billing cheaper?
Rule of thumb: few people using it intensively favours per-user billing, many people using it occasionally favours consumption-based. If you cannot estimate either with confidence, a flat rate with an allowance is the calmest choice – the floor is predictable and only the excess varies.
What extra costs come with running it on-premise?
A server with suitable GPUs, power and cooling, plus internal operating effort. In exchange, usage-based fees largely disappear. For organisations with many users or strict data protection requirements it often pays off – for a small use case, usually not.
Is there public funding for the rollout?
Yes. AI and digitalisation projects are widely eligible for funding in Germany and the EU, among others through the research allowance. We have compiled the currently running programmes with their official sources.
Related pages
What this looks like with KOSMO
Theory is one thing – in 30 minutes we show you live how KOSMO does this in your organisation. With your own content.
