In short: in France, a chatbot connected to your documents usually costs between €8,000 and €25,000, a business assistant connected to your tools between €25,000 and €70,000, and an autonomous agent able to act across several systems between €60,000 and €150,000 or more. On top of this build budget comes a monthly usage cost, proportional to volume, ranging from a few tens to several thousand euros per month.
The term "AI agent" now covers very different things, from simple automation to a system that makes decisions on its own. That is the first source of confusion in quotes. This article helps you work out what you actually need, and what it really costs.
Chatbot, assistant or agent: what are we talking about?
- A document chatbot answers questions from your content: knowledge base, documentation, terms and conditions. It reads; it does nothing.
- A business assistant is connected to your tools (CRM, ERP, helpdesk, database). It looks up current information, drafts answers, fills in records. A human approves.
- An autonomous agent chains several steps to reach a goal: it decides which actions to take, calls several systems, handles errors, and only escalates to a human when needed.
- AI-augmented automation is a fixed workflow (for example in n8n or Make) in which an AI model performs one step: classifying an email, extracting data from an invoice, summarising a document.
The more autonomy, the higher the cost, not because of the AI itself, but because of everything that has to be built around it to make it reliable.
Price ranges
| Type of solution | Examples | Build budget | Timeline |
|---|---|---|---|
| AI-augmented automation | Sorting incoming emails, invoice extraction, meeting-note summaries | €2,000 – €10,000 | 1 to 3 weeks |
| Document chatbot | First-line customer support, internal FAQ, HR assistant | €8,000 – €25,000 | 3 to 6 weeks |
| Connected business assistant | Sales assistant linked to the CRM, quote-writing help, copilot inside your SaaS | €25,000 – €70,000 | 6 to 12 weeks |
| Autonomous agent | End-to-end handling of support requests, lead qualification and follow-up, case management | €60,000 – €150,000 and more | 3 to 6 months |
These amounts assume an experienced team in France or Europe. They include scoping, development, integrations, building test sets and going to production.
What drives the budget
Integrations
This is almost always the biggest item. Reading a PDF costs nothing; connecting cleanly to a legacy ERP, managing access rights and syncing data takes weeks. Before asking for a quote, list the tools the agent needs to access, for reading and for writing.
Data quality
An assistant built on outdated, contradictory or scattered documentation will give outdated, contradictory or incomplete answers. Cleaning and structuring content is often an underestimated part of the project.
Level of autonomy
An assistant that suggests while a human approves is simple and safe. An agent that emails a customer itself, changes an order or refunds a purchase needs guardrails, limits, full logging and much more thorough testing.
Reliability requirements
An internal tool can be wrong now and then. An agent that talks to your customers, or works in a regulated field, must be evaluated systematically: reference question sets, measured accuracy rate, regression tests every time the model or the instructions change.
Confidentiality
Personal data, trade secrets, health data: depending on the case, you will need providers that do not train on your data, hosting in Europe, or even a self-hosted model. Each level of requirement has a cost.
Usage cost: the line people forget
Once in production, every exchange with an AI model is billed by the provider (OpenAI, Anthropic, Google, Mistral) based on the volume of text processed. The unit cost is low, but it multiplies with the number of users.
| Monthly item | Order of magnitude |
|---|---|
| AI model calls | From a few cents to a few tens of cents per conversation depending on the model and length |
| Hosting and vector database | €50 – €500 |
| Monitoring, logging and evaluation | €0 – €300 |
| Maintenance and continuous improvement | 15 to 25% of the build budget per year |
Example: a support assistant handling 1,000 conversations a day, at an average of 2 to 10 cents per conversation, amounts to roughly €600 to €3,000 per month in model usage. That is often far less than the human time saved, but it must be part of the business case from day one.
The levers to keep this bill under control are well known: use a lighter model for simple tasks, cache frequent answers, limit the amount of context sent with each request, and set usage caps.
Three costed examples
A support assistant for an online retailer
It answers questions about products, delivery and returns from the documentation and order tracking, and hands sensitive cases over to a human. Indicative budget: €20,000 to €35,000, over five to eight weeks, then a few hundred euros per month in usage.
A sales copilot linked to the CRM
It prepares meetings by summarising customer history, writes meeting notes and suggests follow-ups. The salesperson approves every action. Indicative budget: €35,000 to €60,000, over two to three months.
An agent handling incoming requests
It reads requests received by email and web form, classifies them, retrieves information from three internal tools, handles simple cases on its own and prepares a complete file for the others. Indicative budget: €70,000 to €120,000, over three to five months.
These examples are orders of magnitude to help you place a project, not quotes.
The pitfalls that blow up the budget
- Starting with the tool instead of the problem. "We want an AI agent" is not a need. "We spend 40 hours a week sorting requests" is one, and it can be costed.
- Aiming for full autonomy in the first version. Start with an assistant that suggests and a human who approves. Increase autonomy when the metrics justify it.
- Not measuring quality. Without a reference test set, you cannot tell whether a change improves or degrades the answers. It is the number one reason projects stall.
- Forgetting usage costs in the business model, especially if the AI is built into a product you sell to your own customers.
- Depending on a single model provider. Models change every few months. An architecture that lets you switch protects you on both price and quality.
- Mistaking an impressive prototype for a product. A demo that works on ten hand-picked questions says nothing about the next thousand. We cover this in vibe coding: from prototype to production.
How to get started without taking risks
- Pick a single use case that is frequent, measurable, and where mistakes are not serious.
- Measure the current situation: time spent, volume, response time, error rate.
- Build a pilot in four to six weeks, with systematic human approval.
- Evaluate on real cases and compare with the starting metrics.
- Gradually extend the scope and the autonomy.
To identify the use cases that create real value, see also our article on integrating AI into a SaaS.
Frequently asked questions
How much does an AI chatbot cost for a business?
For a chatbot that answers from your documents, expect between €8,000 and €25,000 to build, then a few tens to a few hundred euros per month in usage depending on conversation volume.
What is the difference between a chatbot and an AI agent?
A chatbot answers questions. An AI agent acts: it looks up your tools, chains several steps and performs actions, such as updating a customer record or processing a request, with more or less autonomy.
How much does it cost to use a model like ChatGPT or Claude in production?
Providers bill based on the volume of text processed. Depending on the model and the length of the exchanges, a conversation costs from a few cents to a few tens of cents. For 1,000 conversations a day, that often means between a few hundred and a few thousand euros per month.
Can you build an AI agent with a no-code tool?
Yes for simple automations, with tools like n8n or Make, for a budget of a few thousand euros. As soon as the agent needs to access sensitive data, act across several systems or talk to customers, custom development becomes necessary to guarantee reliability and security.
How long does it take to deploy an AI agent?
Four to six weeks for a focused pilot with human approval, two to three months for an assistant connected to several tools, and three to six months for an autonomous agent in production.
👉 Have a use case in mind? Explore our AI chatbot and assistant and AI agent automation services, or book a call to get a realistic estimate.