A new tool is installed, the licences are paid for, the supplier has delivered a clean demo. Three months later, no one is using it. This is the most common picture in failed AI projects at SMEs, and the reason almost never lies in the software. The technology works. What does not work is its introduction to the people who are supposed to work with it. Anyone who wants to introduce AI and keep their staff on board must first solve acceptance, not architecture.
Why AI projects fail on acceptance, not technology
The widely cited MIT study “The GenAI Divide: State of AI in Business 2025” examined 150 leadership interviews, a survey of 350 employees and 300 public AI deployments. The result: only around 5 per cent of the generative AI pilot projects examined measurably accelerated revenue, while the large majority showed no discernible contribution to results. The researchers’ reasoning is the decisive point: the core of the problem is not the quality of the models but a learning gap in tools and organisations.
This matches what we see in practice at Vollmer Labs. A language model that drafts an email or pre-checks an invoice is rarely the bottleneck today. The bottleneck is the question of whether the team trusts the result, understands the benefit and does not feel railroaded. The same MIT study explicitly cites as a success factor not only a central AI lab but also enabling line managers to drive the introduction. Acceptance arises where the work is done, not in the meeting room.
The four fears your staff have
Behind the word “resistance” there are usually concrete, understandable concerns. Anyone who does not take them seriously is fighting symptoms. In Swiss figures, these concerns are well documented: according to the EY European AI Barometer 2025 (500 respondents in Switzerland, 4,942 across nine European countries), 43 per cent of employees in Switzerland are worried about the negative effects of AI on their jobs, and 76 per cent expect AI to lead to job cuts. At the same time, 86 per cent already use AI tools.
The four recurring fears:
- Fear for the job. “If the machine can do this, will they still need me?” This concern is the loudest and the strongest blocker.
- Fear of losing competence and being shown up. “I don’t understand this and I’ll look out of date.” It is precisely the experienced, valued members of staff who fear being seen as behind the times.
- Distrust of the results. “Can I rely on this, or is the thing making something up?” This concern is legitimate and should not be talked away.
- The feeling of being bypassed. “No one asked me, and now I’m just supposed to use it.” A top-down decision without involvement almost inevitably produces defiance.
These fears are neither irrational nor “typically Swiss”. They are the normal reaction of people who do not know what a change means for them. As a managing director, your task is not to argue them away but to answer them.
Transparency and “support, not replace” as the foundation
The most effective lever is mundane and yet rarely used: stating honestly how things stand. Practical experience demonstrably reduces fears; with an 86 per cent usage rate, Swiss employees are among the European front-runners according to EY, and it is precisely this proximity that turns scepticism into routine. In concrete terms, transparency means:
- Saying what the AI is meant to do and what it is not. “It drafts the reply; you decide and send” is a different message from “It does your job”.
- Saying what happens to the data. Which information goes where, who has access, what stays in-house. According to the Swiss surveys, data protection concerns are among the most frequently cited reservations.
- Saying what you do not yet know. An honest “We are testing this, and if it is no good we will drop it” builds more trust than any polished demo.
The second pillar is the stance of “support, not replace”. This is not a marketing line but a design decision. An AI that prepares routine work and leaves the decision to the human addresses the fear for jobs in a way no reassuring promise can ever achieve. Our productive building blocks are built exactly this way: the CPA accounting agents that run daily in a US fiduciary practice prepare the bookings; the human gives approval. The RFQ tool rfqbuddy.com structures enquiries but does not replace the decision on whether and how to quote. Anyone who introduces AI this way does not have to promise that no job will disappear, because the task visibly remains with the human.
Quick wins: start small, where it hurts
The second mistake after a lack of transparency is the all-out assault: AI is meant to overhaul the core business straight away. This overwhelms people, produces maximum resistance and maximum risk. The better path is a quick win, a small, visible improvement to a task that no one likes doing anyway.
Good candidates for a first quick win:
- A tiresome, repetitive task, not a sensitive core process.
- A task whose result is easy to check, so that trust can build.
- A task where a mistake is not costly.
- A task whose relief the staff feel themselves, not just the accounts department.
When the burden falls away and the job remains, the mood shifts. “This takes something away from me” becomes “This takes something off my plate”. It is important to involve the people who do the work every day: research and practice on change processes show that acceptance rises when staff are allowed to test new tools, give feedback and help shape them, rather than implement a top-down directive. In some sectors we have productive cases; in others the building blocks are running, but not yet in that exact sector, for example in hospitality, where our tool jeffri.ch, aimed at kitchen studios, is in a Swiss pilot. That honesty is part of it.
A realistic roadmap in four steps
No big bang, but a sober process that builds trust before it picks up pace:
- Listen and clarify the goal. Where do people lose time every day? What should the AI solve, and how will you measure success? Clear success criteria defined in advance significantly increase the hit rate, according to the analyses of AI projects.
- Choose a quick win and announce it transparently. One task, a small team, a clear framework for what the AI does and what the human decides.
- Run a pilot, gather feedback, adjust. Learn on a small scale before rolling out widely. The staff in the pilot become either ambassadors or brakes, depending on how seriously their feedback is taken.
- Expand where it has proven itself, and stop where it has not. An abandoned attempt is not a failure but an inexpensive lesson. Communicating this openly protects your credibility for the next step.
A word on the legal framework, so that transparency is not merely goodwill: on 12 February 2025 the Federal Council decided to ratify the Council of Europe’s AI Convention and to prepare a consultation draft by the end of 2026 covering, among other things, transparency, data protection, non-discrimination and oversight. Even today, the duties of data protection law apply when personal data is processed. Transparency towards the team is therefore not only sensible but moves in the direction of what is expected anyway. Anyone who proceeds this way introduces AI without losing their staff, and it is precisely this that decides between success and standstill, not the choice of model.