The price of an AI agent, from packaged to program.
The French market has settled into four tiers. The ranges below combine our own engagements (25+ systems in production) with prices published by French providers in 2026.
| Tier | Setup | Recurring | What you get |
|---|---|---|---|
| Packaged solution (SaaS) | €0 to €500 | €29 to €300/month | Generic chatbot, FAQ, hand-off to a human. Not connected to your systems. |
| Well-scoped custom agent | €4,000 to €8,000 | €120 to €600/month | A precise scope executed end to end inside your tools: qualification, document processing, follow-ups. |
| Complex agent | €8,000 to €18,000 | €300 to €1,500/month | Multiple sources, fine-grained decisions, multiple integrations (CRM + ERP + invoicing), reinforced guardrails. |
| Multi-agent program | €20,000 and up | €800 to €2,500/month | Several agents orchestrated across a full value chain, centralized governance and monitoring. |
Two ways to read this table. The wrong one: "packaged is cheaper." The right one: each tier answers a different need. A €79/month SaaS that answers common questions is a good buy if that's what you need. It will never qualify your leads inside your CRM at 2am. The difference between answering and executing is detailed in our comparison AI Agent vs Chatbot.
The 5 factors that take you from €4,000 to €18,000.
Two agents that carry the same name can cost four times as much as each other. Here's what actually moves the quote, in order of impact.
The decision scope
An agent that sorts emails makes simple, reversible decisions. An agent that replies to customers or approves a refund needs careful scoping: what it decides alone, what it escalates, its hard limits. The higher the stakes of the decision, the more guardrail design weighs on the budget. It's factor number one.
The number of sources and integrations
Every connected system (CRM, invoicing, ERP, inbox, internal API) adds design, testing, and maintenance surface. A single-source agent is structurally cheaper than one reconciling five tools. Clean, documented APIs bring the bill down; manual exports push it up.
The reliability bar
An internal agent that pre-fills a spreadsheet can tolerate imperfection. An agent that writes to your customers can't. Reliability is built: a two-week shadow mode, test sets, double-checked outputs, full logging. This invisible work is exactly what separates a prototype from a system in production.
Volume
Volume mostly affects the recurring cost: AI model calls are billed by usage. Processing 200 documents a month costs a few tens of euros; processing 20,000 changes the category and justifies upfront optimization work (lighter models for simple cases, caching, batch processing).
The state of your data
An agent reasons well on clean, accessible data. If your CRM is up to date and your documents are structured, the agent connects fast. If data is scattered across files and inboxes, a structuring phase gets added to the project. It's often the best investment of the lot: it then serves every agent that follows.
The €120 to €2,500/month, line by line.
Recurring cost is the least well explained part of most quotes. Three line items make it up, and each one can be managed.
| Line item | Monthly range | What moves it |
|---|---|---|
| Hosting | €10 to €50 | A European VPS covers most agents. Rises with volume and redundancy needs. |
| AI model calls | €30 to €1,500 | Billed by usage (tokens). Depends on volume processed, document length, and the model chosen. The right instinct: the right model per task, not the biggest one everywhere. |
| Monitoring & ongoing maintenance | €50 to €500 | Tracking success rate, handling exceptions, adapting as your tools or the models evolve. This is what makes an agent last for years. |
A concrete order of magnitude: a document-processing agent handling 3,000 documents a month typically uses less than €150 in model calls. Per document, the marginal cost runs in cents, compared with the minutes of manual entry it replaces.
On model choice (Claude, GPT, Gemini, Mistral) and its impact on the bill, our comparison Which AI to Choose in 2026 breaks down real prices by use case. And for the orchestration layer, self-hosting n8n cuts infrastructure cost significantly: see Self-hosted n8n.
Calculate the return before you sign.
A price alone means nothing. The only question that matters: how much does the agent bring back relative to what it costs. The base formula fits on one line:
Monthly gain = (hours freed × loaded hourly cost) + value of speed + value of 24/7 continuity
A real, anonymized example: a sales-qualification agent for a B2B SMB.
- Situation: two sales reps each spend 1.5 hours a day enriching, scoring, and routing inbound leads, around 60 hours a month total.
- Cost of the agent: €6,500 to set up, €200/month recurring.
- Direct gain: 60h × €35 loaded = €2,100/month, not counting the under-a-minute response time on hot leads, which improves the conversion rate.
- Breakeven: reached before month 4. Over 12 months: around €25,000 of sales time freed up for €8,900 invested.
Over six months, on every AUTOMATE ALL engagement. Below that, we don't recommend the project.
Across our systems in production. High-volume, highly repetitive processes pull the top of the range.
For a well-scoped agent. 5 to 9 months for a complex multi-source agent.
"ROI reached in under 90 days after the audit."
L. Charrière, founder, consulting firm
The instinct to keep: demand the projected gain calculation in the quote. A vendor who prices the setup without pricing the gain is asking you to trust them. A partner prices both and lets you decide on the numbers.
Why quotes range from €1,000 to €50,000.
The same words "AI agent" cover very different realities. To compare quotes that only share a title, check what each one actually includes:
- Design of the decision scope. Who decides what the agent can do on its own? If the answer is "the prompt," the scoping still needs to happen.
- Shadow mode before go-live. An agent tested for two weeks alongside a human before going autonomous. That's the standard for a serious deployment.
- Production monitoring. Success rate, exceptions, monthly cost tracked. An unmonitored agent drifts within a few months.
- Ownership of the deliverable. Code, prompts, workflows, data: who owns them at the end? At AUTOMATE ALL, the client does, 100%, documentation and training included.
- A priced, projected gain. Does the quote state what the agent brings in, or only what it costs?
These five lines explain most of the price gap. A €2,000 quote that includes none of this isn't cheaper: it's a different product. And a €40,000 quote for a scope a well-scoped agent covers for €6,000 deserves the same questions.
Our model: no hourly rate, billing on delivered value. The audit produces an implementation proposal with architecture, roadmap, and projected ROI; the implementation quote is independent of the audit, and the decision is yours. The full method is described on our Process page.
AI agent pricing: what people ask us.
How much does an AI agent cost for an SMB in 2026?
A well-scoped AI agent (one task, well-understood sources, clearly bounded decisions) costs €4,000 to €8,000 to set up. A complex agent (multiple sources, fine-grained decisions, multiple integrations) runs €8,000 to €18,000. On top of that, recurring cost runs €120 to €2,500/month covering hosting, AI model calls, and monitoring. ROI is typically reached in 3 to 5 months for a well-scoped agent.
Why do prices range from €1,000 to over €50,000?
Because the word agent covers very different realities. At €1,000, it's a workflow with a prompt, no guardrails, no monitoring. At €50,000, it's often a multi-agent program for a mid-market company with strong compliance requirements. The five factors that set the price: the decision scope, the number of sources and integrations, the reliability bar, the volume, and the state of your data.
What's the monthly cost of an AI agent?
€120 to €2,500/month depending on complexity. The breakdown: hosting €10 to €50 (a European VPS), AI model calls €30 to €1,500 depending on volume processed, monitoring and ongoing maintenance €50 to €500. A document-processing agent handling 3,000 documents a month typically uses less than €150 in model calls.
Can you build an AI agent for free with ChatGPT?
No. ChatGPT is a conversational assistant: it answers when you write to it, it isn't connected to your tools, and it doesn't work in your absence. A business AI agent needs access to your systems, guardrails, hosting, and monitoring. Open-source building blocks bring license costs to zero, but design, integration, and reliability engineering remain the core of the investment.
Does an AI agent cost less than hiring someone?
For high-volume repetitive tasks, yes, clearly. A well-scoped agent at €6,000 to set up and €200/month recurring costs around €8,400 in year one, when a loaded operational hire costs €35,000 to €45,000 a year. The agent doesn't cover the full range of what a person does: it absorbs the repetitive part and lets you absorb volume growth without hiring.
How do I find out what an AI agent would cost me?
Through an audit. At AUTOMATE ALL, the audit is a fixed fee based on company size: small business €2,500 (2 weeks), SME €7,500 (3 to 4 weeks), mid-market €15,000 (5 to 6 weeks). It maps your processes, identifies the agent with the fastest ROI, and costs out the projected gain. The implementation proposal that follows is an independent quote: you decide afterward, with real numbers.
How long does it take for an AI agent to pay for itself?
A well-scoped agent reaches breakeven in 3 to 5 months on average, a complex agent in 5 to 9 months. The math: hours freed per month multiplied by the loaded hourly cost, plus the value of response speed and 24/7 continuity. Across our engagements, the minimum ROI we target is 20% in six months and the maximum ROI delivered reaches 300%.