- → A chatbot or simple assistant: €3,000 to €8,000 excl. VAT.
- → A targeted agent or copilot with integrations: €10,000 to €25,000 excl. VAT.
- → A custom business application (first version): €30,000 to €150,000 excl. VAT.
- → Price depends mostly on scope, data and the level of customisation, not on “AI power”.
- → The real variable is adoption: a tool that gets used pays for itself, a tool that gets ignored is 100% cost.
Price ranges by project type
An “AI project” means nothing until the deliverable is defined. Here are the benchmarks observed in the French market (public price lists of specialised agencies and IT firms, 2025–2026):
| Project type | What it is | Range (€ excl. VAT) |
|---|---|---|
| Chatbot / simple assistant | A precise use case, no heavy integration | €3,000–8,000 |
| Targeted agent / copilot | One team, one use case, with integrations (CRM, ERP…) | €10,000–25,000 |
| Custom business application | First version of a complete product, in production | €30,000–150,000 |
| Recurring cost | Hosting, tokens, maintenance, support (per month) | €80–600/mo |
For a first project, the sweet spot is the targeted agent or copilot (€10,000 to €25,000): serious enough to create measurable value, tight enough to stay manageable. At very small scale (one person, one simple need), a first building block can start lower, relying on subscriptions and light configuration.
What really moves the budget
For the same deliverable, two quotes can vary threefold. The real factors:
- Scope. A precise use case costs a fraction of a vague, sprawling “AI project”. It's the first lever for savings.
- Data quality and access. Clean, accessible data = fast project. Scattered, dirty or hard-to-integrate data = the line item that inflates the bill most.
- The level of customisation. Assembling existing building blocks costs less than developing a fully bespoke tool.
- Integrations with your systems (CRM, ERP, business tools) and security / compliance requirements (GDPR, hosting).
- Adoption. A tool nobody uses is 100% cost and 0% return. Training and support are part of the real budget.
The costs people forget
Development is only the visible part. Also plan for:
- Data preparation and integration (often 20 to 40% of the effort).
- Change management: training, support, measuring real usage.
- Maintenance and iterations: an AI tool lives, adjusts and improves after going to production.
Estimate your project's budget and gain in 2 minutes
Describe one of your teams: our calculator estimates the time AI could hand back to it every year, valued in euros. Conservative assumption, instant result.
Open the AI ROI calculatorHow to estimate your project's budget
The right method isn't to ask “how much does AI cost”, but to start from value:
- Pick one team and one precise use case (the opposite of a sprawling project).
- Estimate the time currently absorbed by repetitive tasks, and the share AI can realistically hand back (prudently, never 100%).
- Set that annual gain against the cost of a first targeted tool to get the payback.
Our AI ROI calculator does the first half of the work in two minutes: your team's annual time handed back, valued in euros, on a deliberately conservative assumption. The cost is quoted after scoping.
Frequently asked questions
How much does a first AI project cost for an SME?
A first targeted project most often lands between €10,000 and €25,000 excl. VAT. A simple chatbot or assistant starts at €3,000 to €8,000; a custom business application runs from €30,000 to €150,000. The price depends mostly on scope, data quality and the level of customisation.
What does a custom AI agent or copilot cost?
Between €10,000 and €25,000 excl. VAT for a targeted scope (one team, one use case), with integrations, scoping, development and training included, plus a recurring cost of around €80 to €600 per month. In a small organisation, a first building block can start lower via subscriptions and light configuration.
What are the hidden costs of an AI project?
Beyond development: data preparation and integration, systems integration, security and compliance, adoption and training, then maintenance and iterations. It's often adoption, not the technology, that determines profitability.
How do I estimate my AI project budget?
Start from one team and one precise use case, estimate the time absorbed by repetitive tasks and the share that can prudently be recovered, then compare it with the cost of a first targeted tool. The AI ROI calculator estimates that gain in two minutes.
Sources and method. There is no standardised price for custom AI development: the ranges above aggregate the public price lists of specialised French agencies and IT firms (2025–2026), notably La Fabrique du Net (a panel of 71 budgets), itSystèmes, Digital Unicorn, Nerolia and Sedigitaliser; hence ranges rather than a single price. Abandonment rate of roughly one project in three after the pilot: Gartner, 2024 (generative AI). Median payback of around 6 to 7 months: McKinsey, 2025. Productivity gains: controlled studies by NBER (Brynjolfsson et al.), BCG × Harvard (758 consultants), GitHub (Copilot).