.svg)
“How much does an AI project cost?” That’s the first question we’re asked in meetings, often before we’ve even understood the need. And the honest answer is that the question is poorly phrased: the term “AI” encompasses at least four very different levels of projects, with budgets ranging from €5 per month per user to over €150,000 for the first version of a business application.
In short: an AI agent costs anywhere from €5 per user per month (for managing consumer AI usage) to over €150,000 for a comprehensive enterprise software solution. In between, a custom-built agent connected to your data starts at around €5,000 for a targeted module. The determining factor isn’t the size of your company; it’s the nature of your needs.
Asking how much “an AI project” costs is like asking how much “a vehicle” costs: the answer makes no sense unless you specify whether you’re looking for a bicycle, a car, or a truck. In this article, we provide our four actual price ranges, along with the specific scenarios they’re based on. All figures come from our operations at IT Systèmes. These are indicative ranges—the final price always depends on the exact scope, which we calculate in-house—but these orders of magnitude already give you an idea of whether you’re talking about a €5,000 project or a €150,000 project. And that’s often what helps break the deadlock.
2. Our pricing: the 4 levels of IT Systems support
Level 1 — AI Proxy: Host Shadow AI starting at €5/month
Your employees are already using ChatGPT, Claude, Gemini, or Copilot—on their work computers, on their personal phones, and sometimes by pasting snippets of contracts or customer emails without even thinking about it. This is “Shadow AI ”: AI that’s present throughout the company, without any oversight or control.
You have two bad options and one good one. Blocking access: people will just switch to 4G on their phones and carry on. Opening everything up without oversight: you’re taking a major legal risk, with no way to track what’s coming in and going out. The right option is managed access: a single interface to AI services on the market, hosted in France, that anonymizes sensitive data before sending it and logs all requests.
Pricing: Starting at €5 (excluding tax) per user per month, with a setup fee of €1,800 (excluding tax). At this price point, the question isn’t “Can I afford it?” but “Can I afford not to?” A well-defined AI usage policy costs less than a single data breach incident.
Level 2 — Copilot Audit: €1,500 to find out if you're ready
You want to deploy Microsoft 365 Copilot. First, you should know this: Copilot doesn’t create new access points; it reveals the ones that already exist. If your SharePoint contains folders shared with “Everyone,” or if anonymous sharing links are scattered across HR or sales files, Copilot will bring all of that to the surface and make it searchable by any employee. Many companies deploy Copilot, discover the problem in production, and then spend the following months putting out fires. An audit beforehand prevents this.
Price: €1,500 (excluding tax), flat rate per day. A one-day non-intrusive, read-only analysis covering 5 key areas and 40 control rules (based on Microsoft Zero Trust and CIS M365 Benchmark standards). You’ll receive three deliverables: a detailed report, a prioritized action plan, and an executive summary. Additional costs to consider: the cost of Microsoft 365 Copilot licenses. (Copilot pricing has changed since Copilot was integrated into Business plans on July 1, 2026—check the current rate with your reseller.) This is a necessary step before launching Copilot on a solid foundation, and a good starting point for securing your AI project.
Level 3 — Custom AI Agent: starting at €5,000 for a targeted module
You have a time-consuming business process: an overwhelmed Level 1 help desk, repetitive accounting requests, RFPs to draft, and a customer service team that can’t keep up. You want to automate, but neither a generic Copilot nor a basic chatbot will do: you need something that understands your data and business rules and can operate within your information system.
This is where custom AI agents come into play: a combination of large language models (LLMs), data-driven search (RAG), your business rules, and integration with your IT systems. This is the level where, in the vast majority of cases, value is realized most quickly for small and medium-sized businesses.
Price: starting at €5,000 (excluding tax) for a targeted module. A module refers to a specific use case: a helpdesk agent who handles Level 1 tickets, an agent who qualifies incoming leads on your website, or an agent who drafts responses to requests for proposals based on your database. The price then increases based on the number of modules, the complexity of the system to be integrated, and the quality of the initial knowledge base.
Two concrete examples, with figures
Helpy — our AI helpdesk agent. We tested it on our internal IT operations before offering it to our clients. Before: 270 hours of Level 1 support per month, with 15 technicians overwhelmed by repetitive questions. After 11 months in production: 141 hours per month, saving 129 hours each month. 65% of tickets resolved autonomously. The cost per ticket dropped to 26 cents, compared to €15–30 in the industry. Zero layoffs: 9–10 technicians redeployed to handle actual incidents and client projects.
Our POC on competitive bids. It used to take us four days to prepare the technical proposal for a bid. With an agent leveraging our successful bids, we’ve cut that time to two hours. The direct result: a 40% increase in proposals submitted and a 22% increase in revenue in the relevant market. No layoffs—just a more efficient use of sales time.
In both cases, the return on investment is measured in months, not years. That’s what makes Level 3 so profitable: the ratio between its cost and the time it takes to pay for itself.
Level 4 — Advanced Development: €30,000 to €150,000 for your business software
You have a core business function that no standard SaaS solution adequately addresses. You’re cobbling together solutions in Excel, or using a generic tool that’s ill-suited to your needs and wastes your time every day. This is your unique selling point—the 20% that truly sets you apart from your competitors—and it’s precisely where off-the-shelf tools fall short.
Historically, having custom software developed for that 20% cost between €150,000 and €500,000 over 12 to 18 months. This was out of reach for most small and medium-sized businesses. AI has broken down this barrier: today, we can build the same software for €30,000 to €150,000 in 30 to 90 days, thanks to a method that captures your business processes directly from your existing software and files, eliminating the specification workshops that used to consume the bulk of the budget and time.
Price: €30,000 to €150,000 (excluding tax) for the initial version, depending on the complexity of the business and the scope of the project. Once the initial version is delivered, you own the code, you manage the hosting, and future updates are handled independently: you are no longer dependent on a SaaS provider.
A real-world example: Obeevi
Obeevi is a video production agency we worked with. Before: tracking in Excel, scattered information, forgotten follow-ups, and standard CRMs ill-suited to the video industry. After our custom development: a bespoke CRM integrating the stages of video production, delivered in 22 days from concept to production. Obeevi owns the code, manages the hosting, and can update the system independently.
3. Expected expenses
Regardless of the level, the cost of an AI agent isn't limited to implementation. Several budget items are often underestimated during the scoping phase.
Development and integration. The most visible component: architecture, API connectors, workflows, and training the agent on your data. This typically accounts for 40 to 60% of the initial budget. Connecting an agent to legacy systems (outdated ERPs, undocumented databases, proprietary APIs) can drive up integration costs—which is why an architectural audit is essential before any cost estimates are made.
Infrastructure and hosting. Cloud servers (Azure, AWS, GCP, OVHcloud), vector databases, log storage. Expect to pay between €50 and €300 per month for moderate cloud volumes. On-premises hosting, required in certain regulated sectors (banking, healthcare, defense), involves a higher initial investment but provides full control over data.
LLM token consumption. Each request consumes tokens billed by the provider (Anthropic, OpenAI, Mistral, Google). API rates in 2026 range from approximately $1 to $5 per million input tokens and $5 to $25 per million output tokens for state-of-the-art models, with lightweight models (such as “Flash” or “Haiku”) costing fractions of a dollar. In practice, an agent often handles a request for just a few cents (on HelpyBot, 26 cents per resolved ticket, all-inclusive). Prompt caching reduces the cost of repetitive requests by up to 90%.
Ongoing maintenance and optimization. An AI agent is not a “one-off” project: analyzing failed conversations, expanding the knowledge base, refining prompts, and upgrading models. Allocate 10 to 20% of the initial budget annually. It is this optimization that boosts an agent’s autonomous resolution rate from 60% at Month 0 to around 80% at Month 6.
Team training. Often overlooked: your teams need to learn how to work with the agent. Change management, internal documentation, onboarding.
4. Market pricing models
Understanding these models is essential for estimating your costs and avoiding unpleasant surprises. Understanding these models helps you estimate your costs and avoid unpleasant surprises.
License (SaaS). A fixed monthly subscription that includes a quota of interactions or features. This is the approach taken by packaged solutions (Intercom, Zendesk AI, Crisp). Predictable budget, limited customization.
Pay-as-you-go. You pay for actual usage: per conversation, per action, or per credit. Salesforce Agentforce, for example, launched a $2-per-conversation model, which has since been supplemented with credits per action and user licenses. A key concern raised by many CIOs: billing per conversation means paying for failures as well. With a 60% resolution rate, a $2/conversation model actually amounts to about $3.33 per case actually resolved. “Per-resolution” models (Intercom Fin, Zendesk, HubSpot) only charge for successful resolutions.
Customized (fixed-price or time-and-materials). A fixed-price contract sets the price in advance based on a defined scope of work—this is our preferred approach at IT Systèmes, as it ensures transparency and budget control. A time-and-materials contract charges based on hours worked, which is better suited for exploratory projects.
Hybrid. An upfront development fee combined with a monthly subscription for maintenance, support, and tokens. This is the model we recommend: the best balance between predictability and flexibility.
5. Hidden costs you shouldn't overlook
The technical cost of integration. Connecting an agent to legacy systems (outdated ERP systems, undocumented databases, proprietary APIs) can double or triple the integration budget. An audit of your existing IT architecture is essential before any cost estimates are made.
The token explosion. A poorly optimized agent (with overly long contexts, no caching, or an oversized model for simple tasks) can see its token cost increase fivefold. Optimizing prompts and choosing the right model for each task are key ways to reduce costs.
Regulatory compliance. GDPR, NIS2, ISO 27001, AI Act: depending on your industry, compliance can add 10 to 30% to your budget. Sovereign hosting, encryption, audit trails, security testing—these are non-negotiable in regulated industries, yet often missing from quotes provided by less scrupulous service providers. We cover this topic in detail in our guide on how to secure an AI agent project.
The cost of "doing nothing." The most insidious mistake: failing to automate. An employee who spends four hours a day on tasks that an agent could handle in a matter of seconds represents a significant opportunity cost in terms of time, responsiveness, and competitiveness.
At what point does an AI agent become profitable?
The answer depends on your specific use case, the volume of interactions, and the current cost of manual processing. For a custom agent (Level 3), the return on investment is calculated in months, not years—our internal cases prove it: HelpyBot reduced our support hours from 270 to 141 per month, and the RFP POC generated a 40% increase in proposals and a 22% increase in revenue. For a comprehensive business software solution (Level 4), ROI depends on the tool’s criticality, viewed as a long-term asset you own.
A key point: ROI improves over time. An agent who resolves 60% of cases at Month 0 can reach 80% by Month 6 through continuous optimization. Every percentage point gained reduces the unit cost.
The AI agent doesn't replace your teams—it frees them from repetitive tasks so they can focus on what requires empathy, judgment, and creativity. To learn more, see our productivity gains by industry.
Why We Display Our Prices
Many digital agencies don’t publish their rates so they can tailor quotes to each client’s perceived budget. We prefer to provide clear price ranges upfront, so you know we’re on the same page even before we meet. If you have a budget of €8,000, you’ll know right away that you’re looking at our targeted Level 3 plan, not a full-scale development project. And if your needs can be met with a simple Level 1 solution at €5 per user per month, we’ll tell you—that’s what we do in 4 out of 10 cases, saying no to custom solutions when a simpler answer exists.
How to identify your needs in 4 questions
- Want to help your teams make the most of ChatGPT and other consumer-grade AI tools? Level 1. Starting at €5 per user per month; get started in just a few days.
- Want to deploy Microsoft 365 Copilot without putting your data at risk? Level 2. €1,500 for the audit, plus the cost of Copilot licenses.
- Are you looking to automate a specific business process (support, sales, HR, accounting)? Level 3. Starting at €5,000 for a targeted module, 4 to 12 weeks.
- Want to replace existing business software or build a solution that no one else offers? Level 4. €30,000 to €150,000 for the first version, 30 to 90 days.
If you're torn between two levels, the right choice is probably the lower of the two. You can always move up to a more advanced level later, as you gain confidence in the material.
Frequently asked questions
How much does an AI agent cost for an SME?
It all depends on what you mean by “AI agent.” For general-purpose AI tools like ChatGPT: starting at €5 per user per month. For a truly custom-built agent connected to your data that performs actions within your IT system: starting at €5,000 for a targeted module. For a complete business software solution that includes AI: €30,000 to €150,000 for the initial release. The determining factor isn’t the size of your company; it’s the nature of your needs.
Why is there such a big difference between €5,000 and €150,000?
Because we’re not talking about the same thing. For €5,000, you’re buying a module: an agent that performs a specific task (responding to simple tickets, qualifying leads), integrated with your IT system. For €150,000, you build a complete, custom-made business application: your core business digitized, with your own code, your own hosting, and your own autonomous updates. That’s the difference between buying a feature and building an asset.
Is the AI agent free?
Some solutions offer limited free versions (OpenAI’s Custom GPT, Google’s NotebookLM). But for professional use—with IT integration, data security, and guaranteed performance—an investment is unavoidable. Free solutions are meant for testing a concept, not for deployment in production.
Are there any recurring fees in addition to the setup fee?
Yes, there are two main cost components. The use of AI models (Claude, GPT, Gemini, Mistral), billed on a pay-as-you-go basis: a marginal but real cost, always included in the initial quote. And hosting, monitoring, and maintenance, depending on whether you manage it yourself or entrust the operation to us. On HelpyBot, the cost per resolved ticket is 26 cents, all-inclusive.
How long does it take to see a return on investment?
For a custom agent (Level 3), ROI is calculated in months. Our internal cases: HelpyBot reduced our support hours from 270 to 141 hours/month; the RFP POC generated a 40% increase in proposals and a 22% increase in revenue within the scope. For a comprehensive business software solution (Level 4), ROI depends on the criticality of the tool, viewed as a long-term asset that you own.
Can we start small and expand later?
This is what we recommend in almost all cases. Start with a module focused on a single use case, measure its value over a few months, and then scale up once you’ve established confidence. Starting out too ambitiously without having proven the value is the surest way to derail a project.
How can the cost of an AI agent be reduced?
Start with a targeted POC, use the right LLM model for each task (a lightweight model for sorting, a powerful model for reasoning), leverage prompt caching (up to 90% savings on repetitive queries), share infrastructure across multiple agents, and choose a partner who clearly defines the project scope from the outset.
How much does an AI automation agency service cost?
Prices vary widely. Large digital agencies often charge by the hour, with quotes that can range from one-third to three times the price for identical projects. Specialized firms generally offer more transparent flat-rate pricing, with a commitment to results. For a reliable estimate, a preliminary assessment is essential.
How do I know if I don't need an AI project at all?
We’ll be honest with you. In 4 out of 10 meetings, we advise against custom solutions: either an off-the-shelf tool is sufficient, the need isn’t fully developed yet, or the organization isn’t ready to embrace the change. AI amplifies your strengths but exposes your weaknesses: if you’re not ready to make changes to the organization, it’s better to postpone the project.
How much does an AI project cost? Between €5 per month and €150,000 for the first version, depending on the scope. These four price ranges cover the vast majority of cases we encounter.
What these prices really show is that the economic barrier to custom solutions has been broken. In the past, specialized software or an AI agent was out of reach for most small and medium-sized businesses. Today, a targeted module starts at €5,000, and a full-featured software package costs less than €150,000 for the initial version. The question is no longer “can I afford it,” but “where are my real bottlenecks, and how much does inaction cost me.”
The actual figure, however, always depends on your specific situation—that’s something we calculate together during a workshop, not in a blog post. Learn more about our AI agents for businesses and our real-world use cases in the workplace.
Request a preliminary assessment—get a personalized estimate within 48 hours.
.svg)