How to implement AI in business: 5 steps and prices
In short. To implement AI in a small business, take five steps: pick one task, check its data and who may see it, run a pilot next to the manual work, measure it on questions with known answers, then scale, change or stop. Prices depend on the step: from 150 € for a first piece of work, from 490 € for enquiries and CRM, from 790 € for an assistant on your documents. Model fees and checking time are extra.
Many owners ask how to implement AI in business and where to begin, and the honest answer is not a tool but a sequence. This guide gives a five-step plan that a business with a few people can follow, shows what each step costs in our own price list, and says when to stop. It is written for owners who want a result they can check, not a demonstration.
How do you implement AI in a business, step by step?
You implement AI by moving through five steps in order, and at each one you decide whether to continue. Skipping a step is how most disappointing projects begin.
| Step | What you do | Output | Stop if |
|---|---|---|---|
| 1. Pick one task | Score candidates on repetition, written rules, cheap mistakes, digital input and quick checking | One task with a score of 4 or 5 | No task scores above 2 |
| 2. Check the data | List where the data lives, who may see it and where it would travel | A one-page data map | The data is sensitive and must stay on your premises, and you have no machine for it |
| 3. Run a pilot | Run the AI version next to the manual work on harmless or fictional cases | Two weeks of side-by-side results | The pilot makes errors nobody can explain |
| 4. Measure | Test on questions with known answers, count errors, count the time spent checking | A table of errors and hours | Errors stay above what you accept after two rounds of fixes |
| 5. Decide | Compare the saving with set-up and running cost | Scale, change or stop | The payback is longer than a year |
For step 1, our five-point score for what to automate first does the choosing. For step 4, a planted-error test shows whether a model that checks other models can be trusted.
What does it cost to implement AI in a small business?
It costs from 150 € for a first piece of work, and it grows with the size of the job and with how much of your data is involved. The table lists our own services as a ladder, ordered roughly from lowest to highest commitment.
| What you get | Price | What is included |
|---|---|---|
| A first piece of work on an agent system | from 150 € | A one-page map, tested rules, confidential files blocked |
| PDFs and scans into checked tables | from 190 € per batch | Agreed document types and fields, source traceability, a list of uncertain values |
| A PC upgraded for local AI | from 390 € plus hardware | Preparation of your own machine so that data stays with you |
| Customer enquiries and CRM automation | from 490 € | One source, one table or CRM, an owner notification and an internal reminder |
| Regular reports from your spreadsheets | from 290 € set-up plus from 90 € a month | One source format, agreed calculations, a trial issue, one monthly issue and human review |
| An assistant that knows your business | from 790 € | One site, one language, approved sources, 20 test questions, two revision rounds |
| A workstation or a team server | by written quote | Sized to your work after we see the tasks |
Three costs are not in these prices, and you should count them yourself. Model usage fees depend on volume. Hosting and support are agreed separately, because they depend on whether the system runs on your machine or elsewhere. Your own time spent checking results is real, and a result nobody checks is a risk, not a saving. For the payback formula, see the worked example in our automation guide.
Be careful with “turnkey”. It should mean a written scope, tested scenarios, instructions for handling errors and a handover. If a supplier promises a result before seeing your data, ask what they would test first.
How many businesses already use AI?
Fewer than you might think, so you are not behind. Eurostat reports that 19.95% of EU enterprises with 10 or more employees used at least one AI technology in 2025, and 17% of those with 10 to 49 employees. Among companies using AI, marketing or sales (34.70%) and business administration (31.05%) were the most common purposes. Start where the others start: with administration and customer messages, where mistakes are cheap and results are easy to read.
Does your team need AI training?
Yes, in a modest form, and the EU treats it as a duty. Article 4 of the AI Act asks providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among their staff and others who use AI on their behalf. The measures should reflect each person’s knowledge, experience and training, and the setting where the system is used. The page states that it has applied since 2 February 2025, and that it does not require a specific level of literacy for any individual.
For a small team, one practical session is enough to start:
- What the tool is good at, and where it makes confident mistakes.
- Which data must never be typed into it.
- Who checks the result before it reaches a customer.
- How to report a wrong answer, and who can stop the system.
Write these four answers on one page and keep it with the system. This is general information, not legal advice.
Which risks should you cover before the pilot?
Cover three before you touch real data.
- Data. Decide what leaves your premises. If customer or health data is involved, read why a business may want private AI.
- Wrong answers. Test with known questions and keep the results. For a visitor-facing assistant, use the 20-question test.
- Decisions about people. Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing that has legal or similarly significant effects, with exceptions and safeguards including human intervention. Keep a person in decisions about refusing a customer, credit or hiring.
When should you not implement AI yet?
Wait if the process itself is unclear, because AI will make confusion faster. Wait if nobody owns the data, if the volume is a handful of cases a week, or if a mistake would harm someone and you cannot check the result. In these cases, fix the process on paper first. That is also cheaper than any tool. A simple test: if you cannot explain the task to a new colleague in five minutes, you cannot yet explain it to a model, and the pilot will only show you that.
How does Castel Studio help businesses implement AI?
We start from one task and one page, not from a platform. AI agents begin from 150 € for a first piece of work, and the other rungs of the ladder are priced above. We test with known examples, keep a person in the loop and log what runs. If you are not sure which step to begin with, our seven questions on AI agents help you choose.
Tell us the task that costs you most time each week and where its data lives. We reply in writing within 48 hours.
Questions
How do you implement AI in a small business?
Pick one task, check where its data lives and who may see it, run a pilot next to the manual work, measure it on questions with known answers, then decide to scale, change or stop. Do not start with a tool.
How much does it cost to implement AI in a small business?
At Castel Studio the ladder starts from 150 € for a first piece of work on an agent system, 190 € per batch for PDFs into checked tables, 490 € for customer enquiries and CRM automation and 790 € for an assistant on your documents. A workstation or a team server is priced by written quote. Model fees and your own checking time are extra.
Can you implement AI turnkey?
Yes, if turnkey means a written scope, tested scenarios, instructions for errors and a handover, with hosting and support agreed separately. Be wary of anyone who promises results before seeing your data.
Do my employees need AI training?
The EU AI Act asks providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among their staff. This has applied since 2 February 2025 and does not require a specific level for each person.