000
For an SME, the first-year AI budget should be set by use case, not as one global envelope: expect 600 to 1,100 EUR per employee for a structured rollout, split across licences, training, integration and governance, and released in tranches.
Read the full guide
AI in Business 2026: Complete SME Guide (Strategy, ROI, Adoption)
Answer a few questions and get a personalized assessment with recommendations tailored to your industry.
Assess my AI maturityRelated articles
Dive deeper with these complementary articles.
AI in business means handing software systems the tasks that used to require skilled human time: writing, analysis, sorting, customer replies. In 2026, the question for an SME is no longer whether to experiment, but which two or three uses to pick, measure and sustain.
For a 30-person SME, a first-year AI envelope of 17,000 to 34,000 EUR excl. VAT is realistic, which works out to 570 to 1,130 EUR per employee. But the right reflex is not to approve that amount in one go: break it down by cost line, then release it in tranches as each use case proves its value.
This article does not repeat the price detail we covered in How much does AI cost an SME in 2026. It answers the next question: once you know the orders of magnitude, how do you build the envelope, split it, defend it in front of leadership and manage it over twelve months? For the full picture, see our guide to AI in business for SMEs.
An AI budget is built by adding up costed use cases, never by applying a percentage of revenue. We have not found any such rule backed by a serious source for French SMEs, and the ratios in circulation are almost always averages from large groups.
The method fits in three columns for each use case: the time or money spent today, the expected gain, and the cost to get there. An example: sorting inbound requests takes 6 hours a week at the sales front desk. At 35 EUR of loaded hourly cost, that is roughly 10,900 EUR a year. If automation recovers half of it, the gain is 5,400 EUR, which supports an integration budget of 2,000 to 4,500 EUR. The calculation is the same for every line, and it is one your finance lead can read. For the detailed gain method, see our article on calculating AI ROI in SMEs and the glossary entry on AI ROI.
This approach has one advantage: it stops you from buying technology without a target. According to the Insee business ICT survey, among French companies using no AI, the first obstacle cited is the lack of perceived usefulness (71%), ahead of cost (32%). A budget built by use case answers that obstacle directly.
For an SME getting started, a reasonable starting key is about 25% licences, 27% training, 35% integration and 12% governance. These are not standards; they are the proportions you get from the median scenario of a 30-person SME, to be adjusted to your situation.
| Cost line | Indicative share of year 1 | Amount for a 30-person SME (indicative) |
|---|---|---|
| Licences | 25% | 4,000 to 9,000 EUR |
| Training | 27% | 5,000 to 9,000 EUR |
| Integration and automation | 35% | 6,000 to 12,000 EUR |
| Governance and compliance | 12% | 2,000 to 4,000 EUR |
| Total | 100% | 17,000 to 34,000 EUR excl. VAT |
These amounts are orders of magnitude drawn from our own quotes and vendors' public price lists. Licences are the only line you can compare online, which leads many managers to spend their negotiating energy there. Yet that is rarely where the envelope is won or lost: an unused licence costs less than an integration project started on unusable data.
If your company has fewer than 10 employees, the logic is the same at a smaller scale, with an envelope of 2,000 to 5,000 EUR. Our article on AI for micro-businesses on a limited budget covers that case.
Split the envelope into three tranches tied to evidence: about 30% for a 60 to 90 day pilot, 40% for the rollout, 30% held in reserve. The principle is simple: the next money is released only if the previous tranche's indicators are met.
This setup is cheap to run and protects against a common scenario: a budget approved in January, fully spent by March, with nobody able to say what it produced. To build the pilot calendar, our 90-day AI roadmap provides a week-by-week outline.
A sound AI budget states from the start when you stop. Without that, a disappointing project keeps consuming resources because nobody dares say it is not working.
Three simple criteria are enough for most SMEs:
These thresholds are management benchmarks we recommend, not standards. What matters is that they are written down before the start and signed off by leadership. The indicators to track are detailed in our article on AI KPIs in business, and in the glossary entry on KPIs.
If you must trade off between two lines, protect training: it determines the return on licences and integration. Equipping twenty people with an assistant at 25 EUR a month costs 6,000 EUR a year; if half never open it, half of that line is lost.
Training has two other budget features. It can be funded: a Qualiopi-certified provider opens the door to OPCO support, with terms depending on your sector and headcount. The rules are described in our article on funding AI training through your OPCO. It has also become an obligation for organisations deploying AI systems: Article 4 of Regulation (EU) 2024/1689 (the AI Act) requires a sufficient level of AI literacy for the staff concerned, as explained in our article on the Article 4 obligation.
In your budget table, therefore, separate the gross training amount from the estimated out-of-pocket cost after funding. The first measures the real effort, the second what leaves your cash.
A 45-minute review each quarter beats an annual budget left untouched. Tools, prices and usage change too fast for a January forecast to hold until December.
The review asks four questions: which use cases kept their promise, which should be stopped, how much of the reserve has been used, and which new expenses appeared (subscriptions taken by teams without approval, for example). That last question is the most useful in practice: undeclared subscriptions are common, and they are the entry point to the off-policy usage covered in our article on AI governance in SMEs.
Think about year 2 as well. Training, integration and governance are mostly entry costs; recurring spend concentrates on licences and maintenance. For the 30-person SME in our scenario it is around 5,000 to 12,000 EUR against 17,000 to 34,000 EUR in year 1. If leadership discovers that gap mid-year, it will wrongly read it as a saving or as a drift.
What AI budget should a 20-person SME plan for? Expect 12,000 to 24,000 EUR excl. VAT in the first year for a structured rollout, or 600 to 1,200 EUR per employee, before OPCO funding. These figures are orders of magnitude from our own quotes; have them validated by a quote tailored to your processes.
Should AI be budgeted as capital or operating expense? Licences and usage-based access are operating expenses. Training and integration may fall under either, depending on the case: ask your accountant before building the budget.
How much reserve should be planned? We recommend about 30% of the envelope as a reserve tranche in year 1. If your data is scattered or poorly structured, go closer to 40%, because integration is the line most prone to overruns.
How do you convince a reluctant leadership team to approve this budget? By proposing only the first tranche, with a bounded pilot, indicators and a stop criterion. It is easier to approve 30% of an envelope with exit conditions than 100% on a promise.
Can free tools replace a licence line? For testing, yes. For professional use, free offers generally do not give guarantees on how your data is used. A professional plan is most often a compliance prerequisite, not a comfort.
Does an upfront audit change the budget? Yes, because it identifies the two or three use cases feasible on your existing data and avoids an oversized project. See our article on the AI audit for SMEs and the glossary entry on AI audit.
Which budget mistakes do you see most often? Equipping everyone in month one, budgeting nothing for training, and setting no stop criterion. We list more in our article on AI deployment mistakes in SMEs.
We mainly work on two budget lines: training your teams and framing usage. Our AI Awareness for Business course lays the common foundation for the pilot tranche, and Operational AI for SMEs builds business use cases with your teams during the rollout. GrowthPerf is Qualiopi-certified, which makes these courses eligible for OPCO support depending on your situation.
If you are preparing a budget for the new term or next financial year, a 30-minute call is enough to say which lines make sense for your company and which can wait. Request a tailored quote: we price line by line, with the estimated out-of-pocket cost after funding. And to step back to the big picture, our guide to AI in business for SMEs places this budget in a complete approach.