CODLAB / Blog · AI & integration

Introducing AI in an SME: Purpose, Role, Data Paths and Checks under the EU AI Act

18 Nov 2026 CODLAB Team 5 min

As of November 2026 · CODLAB editorial team

Short answer: Introduce AI in four steps: clarify purpose and role, map the data paths, plan the checks and collect the open legal questions. The EU AI Act sets its requirements according to use and role, so this description is the basis for any further review by your data protection and legal experts.

1. Start with the question, not with the model

Many AI projects start with a tool: a chat assistant, a language model, a demo. For introducing AI in a company, the order works better the other way round. Three questions carry further than any choice of model: what should the AI do, which data does it use, and who checks the result? Whoever answers these questions in writing has already done the most important part of the preparation.

2. Clarify purpose and role

Regulation (EU) 2024/1689, the AI Act, distinguishes among others between providers, who develop an AI system or place it on the market under their own name, and deployers, who use an AI system under their own authority (Article 3). Most companies using a ready-made service are deployers. Anyone who builds a system themselves or substantially modifies it, however, may also end up in the role of a provider. So record what the system is used for and who is responsible for it.

3. Classify the use

The AI Act follows a risk-based approach. Certain practices are prohibited (Article 5). Special requirements apply to high-risk systems; which systems these are is set out in Article 6 and Annex III, for example in areas such as employment or access to essential services. Systems that interact directly with people are subject to transparency obligations (Article 50). Many everyday applications fall into none of these groups. Whether that is the case depends on purpose and functioning, however, not on whether the system is only used internally: an internal assistant that interacts directly with employees should also be checked against the transparency obligations. The result should be recorded with reasons, not merely assumed. An automatic classification does not replace that reasoning.

4. Map the data paths

For each use, write down on one page which data the system receives, where it is processed, which services are involved and how long anything is stored. Alongside the AI Act, the GDPR continues to apply unchanged. Where processing is likely to result in a high risk to the rights and freedoms of people, Article 35 GDPR requires a data protection impact assessment. A model name alone does not describe a data path: logs, interfaces and maintenance access belong to it as well.

5. Plan checks and assign responsibilities

  • A trial before going live: define typical and difficult cases, ideally with synthetic or approved data, and decide in advance how success will be measured.
  • Human review: name who checks the results and who may decide to stop the system.
  • AI literacy: Article 4 of the AI Act concerns AI literacy; take the exact scope of the obligation from the regulation in its current version. Independently of that, we recommend a short, documented briefing of the people involved: what the system can do, where it can be wrong and when it is stopped.
  • Open legal questions: collect everything you cannot answer yourself for your data protection and legal experts.

6. A worksheet rather than a certificate

Our guide to the EU AI Act describes exactly this working method: clarify purpose and role, map data paths, plan checks and collect open legal questions. You receive it as a PDF by email after your request. It is a working aid, not a certificate of conformity; it does not replace an individual legal review and contains no automatic classification of your project. Current legal texts and EU guidance remain the reference, such as the text of the regulation on EUR-Lex. If you want to go further: in the architecture and AI audit we record the agreed use, data paths and responsibilities, and you receive a written report with open questions for your data protection and legal review.

Frequently Asked Questions (FAQ)

Am I a provider or a deployer under the AI Act?

Anyone using a ready-made AI system under their own authority is usually a deployer. Anyone who develops a system or places it on the market under their own name is a provider. The terms are defined in Article 3 of Regulation (EU) 2024/1689; if in doubt, clarify the role with your legal experts.

Do I need a conformity assessment for every AI application?

No. Special requirements apply above all to high-risk systems under Article 6 and Annex III; systems that interact with people are subject to transparency obligations under Article 50. Whether a use is affected should be recorded with reasons.

Does the GDPR still apply alongside the AI Act?

Yes. The AI Act does not replace the GDPR. Where processing is likely to result in a high risk to people’s rights, Article 35 GDPR requires a data protection impact assessment.

Share this article

Comments

No comments yet. Be the first to comment!

Leave a comment

Will not be published

Related Articles

Project on your mind?

We develop tailored software and AI solutions. Let's start your next project together.

Free initial consultation