Yannick H.,
Too Long; Didn't Read
AI initiatives don't fail because of technology - they fail because they depend on a single enthusiastic person who then goes on vacation. A pragmatic AI competency team of 3-5 people, each investing 2 hours per week, permanently embeds AI expertise in the organization. The model doesn't scale with headcount, but with clarity around roles, mandate, and decision criteria.
The problem: AI depends on one person
Do you know the feeling?
You have someone in the company who has truly caught the AI bug. Who reads every article, tests ChatGPT prompts, explores Copilot, and evangelizes internally. Without them, there would not be any AI initiative at all.
And then they go on vacation for two weeks.
And everything comes to a standstill.
This is not an exception - it is the typical pattern we keep seeing in SMEs. AI enthusiasm exists, but as an individual initiative, not as organizational capability. As soon as the "AI person" is busy, sick, or away, everything fizzles out.
The problem is not a lack of motivation. It is a lack of structural foundation.
That is exactly what an AI competence team is for.
What an AI competence team is not
Before we get to building one: let us briefly talk about what you do not need.
Large companies build "AI Centers of Excellence" with 20, 30, sometimes 50 people. Dedicated department, dedicated budget, dedicated roadmap cycles, governance committees with six signatures. That is enterprise thinking for enterprise problems.
As an SME with 80 or 300 employees, you do not need that. And honestly, it would also be counterproductive - too much structure invites bureaucracy and slows down exactly the agility you have as an SME.
What you need is something smaller and more effective.
3 to 5 people. No full-time positions. A clear mandate. And a regular meeting that still takes place even when the AI enthusiast is on vacation.
Think about it like a working group, not a department.
Figure: The difference between a classic enterprise AI Center of Excellence and the pragmatic SME model, same function, a fraction of the effort.
The five roles - and what they really mean
Here is the most important clarification: These roles are not job titles and not full-time positions. They are functions that someone in your organization takes on - alongside their regular work.
AI Champion (Executive Management)
This is the person who grants the mandate and represents it upward. Not operationally involved, but present. If executive management stands behind the competence team, decisions are made faster and budgets are approved. If not, the team will quietly fall asleep again after three months.
The role of the AI Champion is not to lead AI pilots themselves. It is to clear the path.
AI Coordinator (Operational)
This is the person who keeps the competence team running. Scheduling, maintaining the backlog, documenting decisions, tracking follow-ups. In an SME, this is often someone from IT or project management - someone who likes structure and does not let things go dormant.
The AI Coordinator is the team’s quiet engine.
Domain Experts (2-3 people from the business units)
This may be the most underestimated role. You need people who know how the work really functions - in production, in customer service, in accounting. Without them, you evaluate AI use cases in a vacuum.
(We have seen pilots that were technically flawless and still failed - because no one asked whether the process even works the way the IT colleague thought it did.)
Tech Lead (IT/Data)
Someone who can assess the technical side. Not necessarily an AI specialist - but someone who knows what data is available, which integrations are realistic, and which vendor promises should be treated with caution.
Compliance Representative
Especially in Switzerland, especially with the EU AI Act, NIS2, and GDPR in the background: An SME that introduces AI without someone with a compliance perspective on board is creating risks. This does not have to be a full-time compliance person - often someone from legal or risk who reviews things regularly is enough.
What the team actually does
Now comes the practical part. Because without a concrete task framework, a "competence team" quickly becomes just another meeting without impact.
Here are the five core tasks:
Use-case assessment. The team collects AI ideas from all areas and evaluates them systematically. Not every idea is worth a pilot. Good assessment criteria: How high is the data potential? How clear is the business benefit? How much effort will implementation take? What happens if it does not work?
Pilot support. When a use case goes into implementation, the team is not the implementer - but it is the critical companion. It ensures that success criteria have been defined before the pilot starts.
Governance assurance. Which AI tools may be used? Which data may be fed into the model? Who approves new AI applications? These questions do not have to be answered in a 40-page policy document. But they do have to be answered - ideally before someone types confidential customer data into a public LLM.
(We wrote more about this in our post on AI governance for SMEs - if you do not yet have a foundation there, that would be a good starting point.)
Knowledge transfer. The competence team is the internal learning unit. What worked in the pilot? What did not? Which tools proved effective? This knowledge must be actively shared - in short updates, internal demos, and brief training formats.
Vendor evaluation. The AI market is very noisy right now. Every provider promises transformation in 30 days. The competence team creates the filter: Which vendors are tested? According to which criteria? Who makes the decision?
How you work together
The design of the workflow is almost more important than the distribution of roles.
Two hours per week, all together. That sounds little - and it is, intentionally. More than that encourages people to opt out because it is "too much effort." Two hours is manageable.
A shared backlog. All use cases, ongoing pilots, and open questions go into one place. Not into five different email threads. A simple Trello board or a shared Notion document is enough.
Clear decision criteria. The team does not decide everything - but what falls within its decision-making scope needs clear rules. Which use cases move into a pilot? What costs too much? What is excluded for compliance reasons? These rules should fit on one page.
No long minutes. Decisions are documented, not discussions. What was decided? Who does what by when?
Figure: The workflow rhythm of a pragmatic AI competence team, minimal coordination effort, maximum clarity about next steps.
Building it in six months
Many teams fizzle out because they start too abstractly. Here is a concrete roadmap - not as a rigid rulebook, but as guidance.
Months 1-2: Assemble the team and clarify the mandate
Identify the five roles in your organization. Approach the people individually - explain what will be expected of them (2h/week, no new full-time task). Get executive approval. Put the mandate in writing in two paragraphs: What is the team’s task? What does it decide? What does it not decide?
Start in month 2 with the first use-case inventory. What AI ideas are there in the company? Do not start with the big projects - start with what people are already doing or already considering.
Months 3-4: Launch the first pilot, set governance basics
Choose a use case with clear business value and manageable complexity. Define success criteria before you begin. Smaller is better - a pilot that delivers results in six weeks is more valuable than a large project that is still in planning after three months.
In parallel: Write an initial AI policy. Not a doctoral thesis. Which tools are allowed? Which data may be used? Who approves new applications?
(We have written a separate article on AI training and awareness - a good supporting measure for the pilot.)
Months 5-6: Measure, learn, roll out
Evaluate the first pilot. What worked? What did not? What would you do differently? Share the learnings internally.
Launch the second pilot - ideally in a different area of the business this time. Roll out initial training measures so the knowledge does not remain only within the competence team.
After six months, you will have: one completed pilot with measurable results, a governance foundation, and a team that knows how it works.
That is more than most SMEs can show after a year of AI initiatives.
What regularly goes wrong
We have supported SMEs in building these structures. Here are the mistakes we have seen most often:
Too formal too quickly. A kickoff workshop with a consultant, a 10-page charter document, a four-part governance committee - and then the first meeting comes and half the people do not have time. Start small. Grow with the team.
Too IT-centric. If the competence team consists mainly of IT people, you evaluate use cases from a perspective that is not representative. The person from customer service or production brings context that no technician has.
No executive sponsorship. This is the most common death blow. Without someone at management level who supports the team and clears the path when resources or decisions are in question, the competence team sinks into the operational burden of everyday work.
Launching use cases without success criteria. "Let us see what comes out of it" is not a pilot. It is an experiment without a learning effect. Before every pilot starts: What is success? How do we measure it? By when do we decide?
Making the team too large. Beyond five people, coordination effort increases sharply and decision speed drops. Better five people with clear roles than eight people without them.
A look at the AI readiness check
Before or while you build the competence team, it is worth taking an honest look at the maturity of your AI readiness. Not as a brake, but as orientation: Which use cases are realistic with the data you have? Where is the foundation still missing?
We have developed an AI readiness check that covers exactly that. It helps the team prioritize realistically from the start, instead of launching projects that fail because of data quality or missing infrastructure.
You can also read more in our article on AI in SMEs between ChatGPT chaos and real business value.
Figure: The pragmatic implementation plan, two months of foundation, two months of first pilot, two months of learning and rollout.
The three insights
AI competence is an organizational issue, not an individual issue. As long as knowledge and energy are concentrated in one person, every initiative is fragile at the same time. An AI competence team distributes responsibility and makes AI adoption more resilient to absences, resignations, and operational pressure situations.
Small and clear beats large and vague. Three people with a clear mandate and a regular rhythm deliver more than a ten-person committee without a decision-making framework. The temptation to overdesign the structure is real - resist it actively.
The first six months determine the team’s character. What you build during this time - how you make decisions, how you handle failures, how you share knowledge - shapes how the team will function in two years. Build habits, not structures.
If you would like to build an AI competence team and are not sure where to start, we would be happy to talk with you. Not to sell you a big initiative - but to find out together what makes sense for your organization. Learn more on our AI consulting page.
