It is often the first question, and the hardest to get a clear answer to: how much does custom AI actually cost for a small business? Between providers who quote a single flat fee and those who refuse to give a number before a three-week audit, it is fair to want concrete benchmarks.

The price of custom AI depends mostly on how complex it is to connect to your existing data, far more than on the artificial intelligence itself, which today represents a small share of a project's total cost.

What really drives the price

Four factors weigh far more heavily than the simple fact of using AI:

  • The number of systems to connect: a chatbot that answers from a fixed text costs less than an assistant that needs to query your CRM, your stock, and your calendar in real time.
  • The volume and cleanliness of your data: already structured data (a clean customer file, an up-to-date catalogue) noticeably reduces setup time.
  • The level of autonomy expected: a tool that only answers questions costs less than an agent capable of acting on its own (sending an email, changing an appointment, placing an order).
  • Ongoing maintenance: an AI tool is never fully fixed once delivered; it keeps evolving with your needs and the underlying models.

Real ballpark figures

Without replacing a proper quote (which always depends on your specific case), here are the ranges generally observed for a small business:

  • A simple, targeted tool (a chatbot connected to a catalogue, an automated diagnostic form): typically a few thousand euros, development and deployment included.
  • An assistant connected to several tools (CRM, calendar, inbox) with a first level of autonomy: a noticeably higher budget, generally spread over several weeks of development.
  • An advanced AI agent capable of acting across several systems and making decisions within a defined scope: an investment closer to a full software project, comparable to an ambitious custom website build.

These figures remain benchmarks: the best way to know your real budget is still a free diagnosis that precisely identifies what needs to be built.

The hidden costs rarely mentioned upfront

Beyond initial development, three items come up systematically and are rarely mentioned at the first meeting:

  • API costs: AI models (Claude, GPT, or others) charge per volume of text processed. For most small businesses, this cost stays modest (often a few tens of euros a month), but it grows with actual usage of the tool.
  • Hosting: a tool connected continuously has to run somewhere, generally for a low monthly cost if the architecture is well designed from the start.
  • Maintenance: security updates, adjustments following a model change, evolutions tied to your business. A tool without maintenance becomes outdated, exactly like a website.

Comparing with the real cost of a generic SaaS

A generic AI subscription always looks cheaper at first glance. But added up over several years, and multiplied by the number of disconnected tools a small business ends up accumulating (see our article on choosing between custom AI and SaaS), the maths often balances in favour of a single, integrated tool that you own outright.

Frequently asked questions

Can you start with a small budget? Yes, it is even recommended: starting with a tool targeted at a single task lets you validate real usage before investing in a larger system.

Does the price include maintenance? Not always by default: it is a point to clarify explicitly with your provider, since an AI tool without follow-up degrades over time, unlike traditional software that stays stable.

Does a free quote commit you to anything? No. A proper diagnosis should help you understand your real need before talking about price, with no obligation to proceed.

Want a concrete quote for your project? Request a free diagnosis: together we identify what is genuinely worth building, and what can wait.