How much does an AI solution cost — and where budgets get burned

We get the question “what will it cost” on every first call, and rightly so. The trouble is that the answer “it depends” is both true and useless. So let us break it into numbers you can actually plan a budget with.

The ranges below are indicative, exclude VAT and apply to small and mid-sized companies in the Slovak market. We give an exact figure only after the analysis — never off the cuff on a first call.

A budget has four parts

Most misunderstandings come from comparing things that are not comparable. A quote for “an AI chatbot” can mean anything from €2,000 to €40,000 depending on what is inside it.

1. Audit and analysis — €900 to €5,000

Mapping processes, assessing data, ranking use cases by return. For a small company, 1–2 weeks. It looks like the item you can skip. You cannot — or rather you can, but you pay for it later and more.

2. Pilot — €3,500 to €8,000

A solution running on your real data, 2–4 weeks. The goal is not a finished product but a number: how many cases it handles, where it fails, whether it is worth continuing.

3. Production rollout — from €5,000, commonly €9,000 to €20,000

Integrations, permissions, monitoring, evaluations, failure states, training. This tends to be the largest item and also the one cheap quotes leave out.

4. Operations — €290 to €600 per month plus tokens

Oversight, updates, reacting to provider changes. Plus model usage — for a typical corporate deployment, in the range of tens to low hundreds of euros a month, depending on volume.

Why quotes differ so much

If you get three quotes that differ fivefold, it is not because two of them are lying. Almost always they include different things. Ask specifically about:

• Evaluations. Does the price include a test set and accuracy measurement? Without it you do not know what you bought.
• Integrations. Connecting to an ERP or CRM is work. If it is not in the quote, you will pay for it later.
• Permissions and audit. For a knowledge base these are a necessity, not a premium extra.
• Training and documentation. Without them you will call the vendor for every change.
• Who owns the code and the data. This belongs in the contract in black and white.

Five places where budgets get burned

1. Starting with three processes at once. It looks more efficient; it is the opposite. Three parallel pilots mean three times the unknowns and usually none of them finishes. One process taken to completion delivers more than three left half-done.

2. There is no “before” number. If you do not know how many hours the process costs today, you can never prove the investment paid back. That is the cheapest measurable step you can take — and usually a week of note-taking is enough.

3. Paying for an integration that already exists. Connectors for common tools are ready-made. Developing a Slack or Google Drive integration should not be a budget line item today.

4. Jumping straight to a custom model. Training is the most expensive route and in most cases unnecessary. A better prompt and access to your data solve more than people expect.

5. Forgetting about operations. The project is deployed, the budget is spent, nothing is left for oversight. Six months later it stops working because an API or a model changed — and nobody was watching.

How to calculate the return

A simple model we use on a first call:

Monthly saving = (hours × hourly rate × share of automated cases)
               + the cost of errors that disappear

Payback = (audit + pilot + rollout) / monthly saving

An example that is easy to compute: two people spend two hours a day each retyping orders. That is roughly 80 hours a month. If an agent handles 70% of cases, you save about 56 hours. At an internal rate of €20 that is roughly €1,100 a month, plus fewer dispatch errors.

With a €12,000 rollout, payback is about 11 months. That is a reasonable number. If yours comes out above two years, the project is probably not the right first step.

What can be done cheaply

Not everything has to cost tens of thousands. Things with a good cost-to-benefit ratio:

• A knowledge base over a single data source. Narrower scope, faster rollout, immediately visible benefit.
• The audit on its own. You are buying a decision, not a solution. And the output works with another vendor too.
• Connecting to ready-made MCP servers. If your tools have connectors, this is configuration, not development.

Summary

A realistic budget for a first AI project in a small or mid-sized company is €12,000 to €25,000 including audit, pilot and rollout, plus €300 to €600 a month for operations. If someone promises you a production solution for €3,000, they are either leaving something important out or you will pay for it later.

And if your calculation shows payback beyond two years, the best advice is not to do that project yet. We will tell you so on the first call — a no-obligation consultation can be booked with no strings attached.