truck8.ai

AI Governance for SMEs: What the Board Owns

AI governance sounds like something for banks with a compliance department. For an SME it comes down to seven decisions by the board, and taken well they make the organisation faster, not slower.

8 min readTom Mekenkamp

What is AI governance, and why is it a board job?

AI governance is how your organisation makes decisions about AI: who decides what gets used and built, which rules apply and who is accountable. It is not a legal file. The law itself is covered in the EU AI Act for SMEs.

Why the board? Because AI cuts across departments. Sales uses a different tool from finance, HR is experimenting with a chatbot and someone in operations has built an agent that reads email. Each department can make its own choices, but nobody sees the whole. And the obligations under the AI Act sit with the organisation, not with the department.

There is another reason. Leave governance to IT and you get governance of technology. In BCG research among 152 CEOs (July 2026), 82% include technology in AI governance and only 30% include HR. Yet the hardest questions are about people and work.

1. Who decides on AI, and with what mandate

Start with one name on the leadership team who owns AI governance. Not automatically the IT manager, but someone who can make decisions across departments. Below that, a point of contact for everyday questions: can I use this tool, can this data go in?

Then agree who decides what. Three levels are enough for most SMEs:

  • Free to use: the approved tools on the list, for internal work, without personal data or confidential customer data.
  • With the owner's approval: anything customers see or receive, anything involving personal data, and any agent that sends something or changes a system on its own.
  • Never: decisions about people (hiring, appraisal, dismissal) without a human decision-maker, and anything the AI Act prohibits.

Put a deadline on every approval

An approval that takes three months is a no in practice. Agree that every request gets an answer within five working days. Then people ask properly instead of quietly sorting it out themselves.

2. An AI register: which AI do you actually use?

For most applications the AI Act does not require a register. But without an overview you cannot answer any other question: not which risk class applies, not who needs training and not where you need to be transparent.

Just ask your people. You will be surprised by what is already running, from free accounts to AI features that were quietly switched on in existing software. Record for each application:

  • Name, supplier and what you use it for.
  • Owner: who maintains it and who decides when it can go.
  • Which data goes in, and whether there is a data processing agreement.
  • Who uses it, and which training goes with it.
  • Whether customers or other people talk to it, or whether it creates content that goes out.
  • The risk class, with a short rationale.

3. Risk classes in one paragraph

The AI Act sorts AI into four classes. Prohibited applications, such as social scoring or emotion recognition in the workplace, are off limits. High-risk applications, such as AI that ranks job applicants or assesses creditworthiness, are allowed, but from 2 December 2027 heavy obligations apply to you as a user too: human oversight, keeping logs, informing employees. Limited-risk applications, such as chatbots and generated content, fall under transparency rules. The rest, by far the largest share of what an SME uses, is minimal risk. Put the class next to every application in your register, and have a lawyer look at anything that comes close to high risk.

4. AI literacy and a skills plan

Article 4 of the AI Act has applied since February 2025. Since the Digital Omnibus it is a duty to take measures to support the AI literacy of your people. You do not have to guarantee a particular level per person, but you must be able to show what you have done. What the law asks exactly is covered in AI literacy.

A skills plan makes that concrete. It needs no more than one page, with for each group what they use, what they need to know and how they learn it:

  • Everyone: what AI can and cannot do, which data must never go in, how to spot a made-up answer.
  • People who work with AI daily: giving good instructions, checking output, knowing when a human has to decide.
  • Builders: agents, integrations, data and security, including misleading input hidden in email or documents (prompt injection).
  • Board and leadership team: what AI means for your strategy, and which decisions sit with you.

Record it, briefly

Keep track of who completed what and when. A simple list is enough. A regulator asks what you did, for whom, and whether that fits the AI you use. With a skills plan and that list, you have the answer.

5. Policy: the rules in plain language

The AI policy is where the decisions above come together: which tools are allowed, which data never, who checks output that goes to customers, who the point of contact is. Two to four pages is enough, in language your employees actually read. A template and the components are in creating an AI policy.

The main difference between a policy that works and one that disappears into a drawer: the first also shows how to do it right. A list of prohibitions alone makes people stop telling you what they do, not stop doing it.

6. Transparency: do people know it's AI?

Since 2 August 2026 the transparency rules in Article 50 apply. If a customer talks to your chatbot, it must be clear that it is AI. If you use AI to make images, audio or video that imitate real people, places or events and look genuine (a deep fake), you must say so. The same goes for AI text you publish on matters of public interest, unless a person has reviewed its substance and someone holds editorial responsibility for it. And providers of generative tools must mark their output in a machine-readable way. For tools already on the market before August, that applies from 2 December 2026.

For governance this means: flag everything in your register that talks to customers or puts content out, and agree who decides when a label is needed. What sits with you and what sits with your supplier is covered in labelling AI content under the AI Act.

7. AI ethics as concrete choices, not a values poster

Almost every organisation has a text about responsible AI somewhere. Almost nobody knows what that text means tomorrow. Ethics only becomes useful when you translate it into choices you can follow and check. A few every board should make:

  • Does AI ever decide about people on its own? My advice: no. A person decides, and that person can explain why.
  • Do we tell customers when AI helps write or think, even where the law does not require it?
  • May an agent send email or change orders on its own, and if so, up to what amount or what risk?
  • Which data never goes to an external AI service, not even with a contract?
  • What do we do with time that is freed up: cut or reinvest? If roles change at scale in the Netherlands, the works council has an advisory right (Dutch Works Councils Act, WOR art. 25).

From choice to rule

Put every choice in the policy or the register, with an owner. Then ethics is no longer a poster but something you can steer on. And when an employee is in doubt, they have an answer instead of a slogan.

The question about freed-up time is the one I most often see left open. That is why a leadership team leaves the boardroom workshop with a decision on freed-up capacity, with a name or a date attached.

Governance that speeds things up instead of slowing them down

The biggest mistake I see is governance as a gatekeeper. Every new application past IT, legal, privacy and the leadership team. The result is predictable: nothing gets through, or everything goes around it.

Turn it round. Agree the guardrails once, so teams can build within them without asking permission: one secure IT foundation for hosting, login, logging and backups; an owner for every application; fixed rules for data and for agents that read and can send email. Then review the exceptions, not everything.

That fits the way AI adoption does scale: not with loose pilots, but with a model line, one value stream fully redesigned by a team with a mandate. Governance gives that team its limits, and within them, its speed. How that works is in leading AI adoption.

AI governance checklist for the board

  • There is one owner of AI governance on the leadership team, and a point of contact for everyday questions.
  • It is agreed what is free to use, what needs approval and what is never allowed. A request gets an answer within five working days.
  • There is an AI register with an owner, the data, the users and the risk class for each application.
  • There is a one-page skills plan, and a list of who completed what.
  • The AI policy is written down, in plain language, and has been discussed with the team.
  • Chatbots and content that goes out are flagged in the register, with agreements on labels.
  • The ethical choices are made and written into the policy, each with an owner.
  • The guardrails are set up so a team can build within them without asking permission per application.
  • There is a date in the diary to review all of this every six months.

Key takeaways

  • AI governance is a board job because AI cuts across departments and the AI Act obligations sit with the organisation.
  • Start with an owner and three decision levels: free, with approval, never. Put a deadline on every approval.
  • An AI register is usually not mandatory, but without one you cannot answer any other governance question.
  • Article 4 asks for measures to support AI literacy. A one-page skills plan and a list of who completed what is the practical answer.
  • Ethics only works as concrete choices with an owner, for example: AI never decides about people on its own.
  • Good governance agrees the guardrails once, so teams can build quickly within them.
TM

Written by

Tom Mekenkamp

AI consultant & founder of truck8.ai

15+ years leading transformations at AB-InBev, Royal BAM and beyond. Now building AI products and helping SMEs implement AI.

AI strategy and governance in one session?

In the Boardroom Workshop (2 or 4 hours) you decide where AI pays off most, who becomes owner and which guardrails apply. Within a week you receive a strategy document with the decisions, the first 90 days and what the AI Act already asks of you.

See the Boardroom Workshop