“We don't use AI at our company.” Are you sure?
Employees may already use AI before your company introduces it. Five questions to understand those habits and identify the guidance and training your team needs.
Picture a small business on a Monday morning. A salesperson asks ChatGPT to rewrite a proposal. The office manager requests a summary of a long supplier document. Another colleague turns on AI note-taking for a video call. Meanwhile, at a business meeting, the owner says: “We haven't introduced AI yet. We'll look into it later.”
The employees wanted to finish their work sooner. They did not launch an implementation project, and they may not even have requested a subscription. A few separate experiments nevertheless became everyday habits. The business started using AI before anyone made a shared decision about it.
As an owner or manager, it is worth finding out whether something similar is happening in your company. A short conversation can reveal where AI helps, which questions need answers and what your team would benefit from learning.
AI enters the business one task at a time
A difficult customer email or a document in another language can prompt the first attempt. If the answer looks useful, the colleague returns to the same tool next time. Later, they may provide an entire email thread or spreadsheet, hoping that more context will produce a better result.
AI features can also appear inside software the company already uses, as writing suggestions or meeting summaries. A list of company subscriptions therefore gives only part of the picture. You need to understand which features people actually use and what they do with them.
There is evidence behind this scenario. In the University of Melbourne and KPMG's 2025 international study, 58 per cent of surveyed employees reported intentionally using AI at work. The study covered 47 countries; it does not establish the rate of use in Hungarian SMEs. It does give managers a reason to ask about the habits of their own teams.
What the finished document cannot tell you
A manager usually sees the finished proposal or summary. That document does not reveal whether it was prepared in a company or personal account, what information was entered, or whether anyone checked the AI's claims. AI use that an organisation does not know about or has not approved is often called shadow AI.
These details matter because they call for different decisions. Rewriting a public product description mainly raises questions about accuracy. With a customer complaint or an internal pricing spreadsheet, you first need to establish whether that information may be entered into the chosen tool. The product's name alone cannot settle this: the account type, settings, service terms and company rules all matter.
The conversation can also uncover methods worth sharing. One colleague may already have a good way to prepare a weekly report that others know nothing about. Understanding their process lets the team decide where it can help and where the results need checking.
Where EU AI Act Article 4 fits in
AI literacy is relevant to ordinary office work too. The European Commission's guidance explicitly discusses companies whose employees use ChatGPT for tasks such as advertising copy or translation. It says those employees should be informed about the specific risks, including hallucinations.
The absence of a formal AI project is therefore not enough to determine whether the business has obligations. Its actual use of AI and organisational role need to be examined. Article 4 concerns measures supporting the development of AI literacy; it does not prescribe the same course for everyone. Our guide to EU AI Act Article 4 explains the requirements and practical record-keeping considerations.
Five questions to ask your team
Start with a short discussion in a team that handles plenty of correspondence or documents. Explain that you want to understand useful methods and resolve uncertainty. Colleagues can describe their workflow without showing the history of their personal accounts. Use fictional or appropriately prepared material for examples.
- Which work tasks have you tried using AI for?
Ask for a concrete example from the most recent occasion. “I use it for customer emails” becomes more useful when you learn whether the colleague adjusts the tone, generates a complete response or summarises a long exchange. - Which tool and account do you use?
Include personal subscriptions and AI features in existing software. This shows where shared access exists and where someone still needs to decide whether a particular setup is suitable for work. - What information do you give the tool?
Discuss the kind of data: public text, customer details, internal documents or financial information. Removing names can still leave details that identify a customer or disclose a business situation. - How do you decide whether the answer is usable?
Ask what they check, which source they use and who approves the final work. Discuss what happens when a convincing answer changes a figure or a condition. - Where would clearer guidance or practice help?
Ask about tasks someone avoided because they were unsure. They may need better access, an answer about data handling or help checking the result they receive.
Keep a short record of the answers: task, tool and account, information used, checking method and open question. Assign someone to follow up each unresolved issue. Colleagues should know who will get back to them and when.
Use the answers to shape the training
The situations you collect show what management needs to arrange and what the team needs to practise. Providing suitable access is a separate task. Uncertainty about a tool's terms needs resolving before people rely on it. Training can then build on those decisions.
| What you discover | Management decision | Useful practice |
|---|---|---|
| Customer emails are drafted in personal accounts. | Agree which tools and accounts may be used, and with what data. | Prepare an email so that only permitted information is entered into the tool. |
| AI summaries are read without the original document. | Decide which claims need checking against the source and who checks them. | Find an omitted condition or a change in meaning in a summary. |
| One colleague has a useful method others do not know. | Establish which tasks the method suits and what access colleagues need. | After a demonstration, let everyone try the method on a fresh example. |
After an exercise, ask participants to explain how they spotted a problem and corrected it. Their answers can reveal where another example or a clearer internal guide would help. Practise with company-approved tools and permitted data so that colleagues can use the same method afterwards.
Someone may be good at writing prompts but have little experience checking sources. Another person may need help getting started. Actual examples from work help you decide which topics everyone should cover together and where separate practice will be more useful.
Start with one team and one recurring task
Choose a regular piece of work, such as preparing customer emails. Discuss the five questions, resolve open issues about access and data, and try a workflow you have agreed together. You can build on that experience when you speak to the next team.
QvAI's practical corporate AI training is adapted to participants' tasks and the tools their organisation uses. If you have already collected a few situations where colleagues need help, those provide a useful starting point for a training plan.