AI budget for SMEs: where to start

The question ‘how much budget is needed for AI’ is asked as if there is a fair percentage of turnover. It does not exist, and the proposer is selling. The sensible AI budget for an SME is built backwards: you start with the cost of an existing process and decide if it is worth replacing.

This article proposes a method of allocation that works well in companies under one hundred people, where there is no dedicated function and every euro must justify itself.

The Three-Scale Structure

Scaglione 1 — Understand (a few hundred euros). Subscriptions to generic tools for people who can benefit from them, for at least three months. It serves to find out where AI impacts real work. It's the only expense I would even recommend to a company that has decided not to make plans.

Scaglione 2 — Build (1,500 – 8,000 €). One process, the one with the highest annual cost and the smallest exceptions. Not two, not three: one.

Scaglione 3 — Maintain and extend. An annual quota for consumption, maintenance, evolution, plus possibly the second process after verifying the first.

How to choose the process to be funded

The criterion is a ratio of two quantities: annual cost of the process and number of exceptions.

1. Lists the company's repetitive processes.
2. For each calculate: times per month × minutes × hourly cost × 12.
3. For each one the exceptions count: "what happens when it doesn't go as planned?"
4. Sort by decreasing cost and discard those with too many exceptions.
5. The first one that's left is your project.

This exercise lasts an afternoon and changes the quality of the decision more than any advice. The method of calculation is deepened in how much it costs to automate a process.

The recommended breakdown

For an SME facing the first year of AI investments, a breakdown that works:

The two cost items cut to zero first are training and contingency, and this is a predictable mistake: the first determines whether the system will be used, while the second avoids having to stop the project at the first unexpected exception. Read more in hidden costs of AI projects.

Practical examples

Professional studio, 8 people. First year: subscriptions for five people, plus a small project on automation of sorting and recurring responses. Order of total size contained, return measurable in hours of the secretariat. The second year is considered to extend to the preparation of the practices.

Commercial company, 25 people. The most expensive process is the insertion of passive invoices. Project in the mid-range, plus subscriptions for the commercial area. Reserve planned and used for two exceptions emerged in the course of work.

Craftsmanship, 5 people. Here the bill says that no process justifies a project: volumes are low. The correct answer is to stop at echelon 1, which however produces benefit. It is an answer worth accepting instead of forcing an investment.

How to defend an AI budget in the company

Getting the budget is a different problem than calculating it, and in SMEs you play in a short conversation with who decides. There are three ways of setting it, and two work badly.

Doesn't work presenting it as an investment in innovation. It is generic, has no comparison term and invites the question "what if we wait?". Doesn't work even present it as staff savings: it puts on the defensive who will have to use the system, and in the practice of SMEs that saving rarely materializes as staff reduction.

Works present it as a replacement for an already existing and measured cost. "This business costs us 9,000 € per year in hours, automation costs 4,000 one-off plus 600 per year of operation, the tie comes within the first year." It is a verifiable phrase, which can be discussed on the merits and which does not ask for confidence.

To get to that sentence you need the work described above: measure the process first. And it's also why it's better to do it even if you already have the budget — because the starting number is the only thing that will allow you, a year later, to say if it went well.

When using APIs, the quote must distinguish available capacity and expected consumption. The levels OpenAI Build, Launch and Grow They affect the limits of the service; they do not indicate for themselves how much the project will spend.

For a new business it is worth checking the incentives before allocating the expenditure. Claude for Startups provides facilities subject to requirements and deadlines: in the ordinary budget enter only those actually recognized, with a separate forecast for the following period.

The Second Year

Almost all the literature on the AI budget is about the first investment. The second year is the one in which you decide things, and it has a different structure: there is no more to discover, there is to consolidate what works and close what did not return.

The question to ask is only one, on the project of the year before: are today's numbers better than those you had measured before you left? If so, it makes sense to extend to the same or adjacent process, and the second project will cost less because some of the integrations already exist. If not, it is worth figuring out if the problem is the system or adoption, before spending any more.

Most common allocation errors

Beyond the overall figure, the way the budget is distributed determines the outcome more than its size.

The typical misallocation focuses everything on realization and zeros outline and follow. The result is a well-built system that no one has learned how to use and no one maintains: in the second year it is still.

The allocation that works always leaves two small but not zero entries: something for training and adoption, something for variants. They are the two voices that seem to be renounceable in the planning phase and become decisive in the reality phase.

The theme is deepened in hidden costs of AI projects and in how much it costs to maintain an AI system.

Mistakes to Avoid

• Setting a percentage of turnover. It doesn't relate to the value of your processes.
• Funding three projects together. It concentrates risk and prevents learning.
• Cutting training and contingency to zero. It is the small voices that determine the outcome.
• Not planning for ongoing operation. From the second year the system remains maintenance-free.
• Not measuring the starting point. It makes the budget unevaluable.
• Forcing a project when the numbers say no. Stopping at subscriptions is a legitimate choice.

How to apply it in the company

Build the budget in this order and not in the reverse one: first measure the processes, then decide how much to allocate. The figure emerges from the numbers instead of being decided a priori and then filled with content.

Present it internally as a replacement for an existing cost, not as an investment in technology. "This process costs us 9,000 € a year, automation costs 4,000 one-time plus 600 exercise" is a phrase that defends itself. "We invest in AI" is not.

The price framework by project type is in how much AI costs in the company, and the choice between internal resources and supplier in DIY AI or agency.

A final criterion, especially valid for the first year: always prefer a smaller budget spent well to a larger budget spent quickly. The first project has a value that goes beyond the immediate result, because it teaches the company how to do these jobs. A contained, measured and completed project is worth more than an ambitious investment that no one can follow.

Finally, a note on the relationship between budget and timing. An investment spread over two financial years, with one project per year, is almost always more effective than the same amount concentrated at once. Not for financial reasons, but because between the first and second project the company has learned something that changes priorities: almost never the second project is what you would have chosen at the beginning, and it is good to be able to decide it later.

Conclusion

The AI budget of an SME is not decided by looking at turnover: it is built by measuring processes. Subscriptions to understand, a project only to build, a quota to maintain.

The discipline that makes the difference is one: measure the current cost before spending. It is what makes spending defensible before and evaluable after, and in most companies it is also the only thing that is missing.

If you want to understand what processes in your company can be automated with AI, Giallo Studio can help you turn the problem into a real workflow.