AI TRANSFORMATION FOR SWISS SMEs
AI transformation takes effect differently, depending on whether your bottleneck is operational, structural or cultural.
Most Swiss mid-sized companies have an AI pilot running or behind them. Few can state clearly, at executive level, why the pilot works or why it peters out, and which concrete steps move the company forward in AI adoption. What helps with AI implementation is not more initiative and more experiments, but a shared understanding of what AI means for the company and a strategic order with which the decision holds in the boardroom.
Bernhard Nitz · transformind GmbH · Kilchberg, Zurich, · May· 2026
THE CENTRAL QUESTION ON AI ADOPTION
Where executive teams stall before they roll out AI.
One question comes up in almost every meeting on AI strategy, often only after the first disappointing pilots. It is in the room, but rarely answered clearly. Here it is, with its answer. The reasoning follows, in the order in which it holds in the boardroom.
Why do AI initiatives fail in the Swiss mid-market?
Partly on the technology, or on expectations of the technology that are set too high, partly on an underestimation of the complexity.
Mostly because the AI tool is deployed at a bottleneck where it cannot take effect, or even makes the situation worse.
Three bottleneck types must be distinguished: operational, structural, cultural. The effect of AI depends on which type you are dealing with. Clarifying this before the rollout prevents pilots that peter out and investments that cannot be explained.
"What an AI initiative can achieve is not decided in the choice of tool, but at the bottleneck it meets."
What the board has not yet decided.
When AI is on the agenda in a board meeting, it is rarely about tools. It is about whether the executive team and the board share the same idea of what AI is meant to mean for this company. A shared target picture, a shared expectation of pace, value contribution and risk appetite.
Three questions remain open in many Swiss mid-sized companies.
-
First, the competitive question: are we losing touch right now, or is a market bubble forming that we should not rush to invest in.
-
Second, the duty-of-care question: what responsibility does the board carry personally when AI systems prepare or make decisions in the company, and how does this sharpen with the EU AI Act from August 2026.
-
Third, the justification question: when the innovation or transformation budget is released, by what logic was it prioritised, and why precisely so.
Beneath these three questions lies the pattern many executive teams know. In day-to-day business, AI initiatives have long since started. Pilots are running, individual tools are in use, individual employees are driving the topic. But the translation of this movement into a language with which the board can meet its strategic responsibility is missing. What is hard to clarify in the board does not begin with the question of which pilot comes next. It begins with the question of what the next twelve months are actually meant to be about.
Three different bottlenecks, three different AI effects.
In every organisation, at any given moment, there is one place where value creation is slowest. This place sets the pace of the whole. It can have three different characters.
TYPE 1
Operational bottleneck
Data, operational processes, coordination, routine. Repetitive tasks too slow, data maintenance fragmented, manual processing dominant.
TYPE 2
Structural bottleneck
Strategy, decisions, information flow, clarity of responsibility, technical systems, knowledge, skills. Too many parties involved, unclear conditions or responsibilities, long approval paths.
TYPE 3
Cultural bottleneck
Trust, readiness, acceptance. Scepticism towards new tools, fear of losing tasks, missing ownership.
At operational bottlenecks, AI works as expected. Where routine is too slow, it becomes faster. Where data is scattered, it becomes structured. Where texts are laborious, they become efficient. This effect is well documented, it appears early, and it is tangible in everyday work.
At structural bottlenecks, AI works differently. It produces more data, more reports, more suggestions. It does not replace the decision that is missing. It accelerates the production of information at a point where the organisation already has information. What is missing is the clarified responsibility.
Here a sharpening is added that, for many SMEs in 2026, becomes the real problem: the auditability of AI output. AI suggestions sound plausible at first glance, but their derivation is barely traceable. Whoever delegates a suggestion takes on responsibility for a result whose reasoning they cannot verify. At a structural bottleneck, decisions do not become faster. They become less certain, because a new layer of hard-to-audit recommendations enters an already overloaded decision space. Further information is also available at the SwissBoardForum.
Academic context
This observation is not new. Ronald Coase and Oliver Williamson showed decades ago that organisations weigh coordination costs against verification costs.
AI agents lower coordination costs and raise verification costs. What looks like acceleration is in truth a shift in the information asymmetry between principal and agent. With classic delegation, an executive can ask an employee for the reasoning behind a recommendation. With simple AI delegation, that reasoning exists in a form that is difficult to question. Moreover, an AI is not directly responsible for its output; that responsibility is carried by the people who train it, who task it, or who accept its results.
"AI lowers coordination costs and at the same time raises verification costs. What looks like acceleration is a shift of responsibility."
At cultural bottlenecks, AI works differently again. Here the bottleneck lies not in processes or in the decision architecture, but in the organisation's readiness to work with the new technology at all. The reasons can be manifold: a lack of knowledge, a lack of time, expectation pressure that is too high or too low, a missing or too narrow frame, scepticism, concern about one's own role, mistrust of what one cannot follow oneself, right down to the question of how safely one can voice an uncomfortable opinion, a difficult question or a perceived weakness.
At a cultural bottleneck, a technical rollout produces resistance instead of adoption. The tools are ready, but go unused, or are used only at the surface. The usual reaction, to organise more training, runs empty here. What helps is not more tooling, but a different form of introduction. One that asks the people involved for understanding and ownership before the tool finds its place.
The most expensive place in many Swiss mid-sized companies does not lie in execution. It lies either in the decision itself or in the readiness to make serious use of the technical possibilities. Whoever deploys AI without having found the real bottleneck does not accelerate value creation. They move the problem to a place where it becomes more expensive.
How can I know what the next right step is for AI adoption in my company?
The most important thing is to listen, gather information, and form assumptions about the situation before you act. Analyse the situation together with representatives of your organisation, including the board. Invite people from different areas, with different levels of seniority, to a dialogue or to complete a structured survey.
Bernhard Nitz has developed a practice-oriented diagnostic and steering model for transformind.
Ambiflow SME diagnosis and steering
Ambiflow is a diagnostic and steering model for transformations that captures bottleneck clarity, value-creation flow, ambidexterity , information flow, social viability and leadership-system maturity across six dimensions. For the AI question, Ambiflow sorts the application perspective into three building blocks.
Where AI takes effect, when the order is right.
Ambiflow Digital sorts the AI question into three building blocks. The order is not arbitrary: whoever tackles level 3 before level 1 automates the wrong place.
LEVEL 1
Diagnostics
Ambiflow makes diagnostically visible where your bottleneck lies, before it shows itself as a symptom. It aggregates Pulse-Check data, detects discrepancies between the leadership and employee perspectives, and quantifies early indicators. The executive team carries this level.
LEVEL 2
Steering
The Ambiflow analysis also serves as a cockpit for the executive team.
-
Decision preparation
-
Early warning
-
Impact tracking
The question shifts from "what do the numbers say" to "which decision is due". The executive team carries this level together.
LEVEL 3
Optimisation
AI deployed at the bottleneck identified by Ambiflow, not everywhere at once. Here things are automated, accelerated, scaled, but in a targeted way and with measurable impact. The functional leads carry this level.
Most SMEs start with optimisation at level 3, because they cannot yet locate AI clearly enough at the strategic level, and because the functional leads, in application, are where the fastest effect is expected. That is precisely what produces the pattern you know: fast pilots, slow impact, hard-to-explain benefit, an overburdened organisation and finally resistance. The order decides the relationship between effort and return.
"Whoever begins with optimisation automates the wrong place. The order decides the relationship between effort and return."
What this looks like in typical AI diagnoses.
Three patterns that frequently appear in diagnostic work, one for each bottleneck type. The details vary, the structure is typical.
PATTERN 1 · OPERATIONAL PILOT MEETS STRUCTURAL BOTTLENECK
Industrial company with an organisation-wide Copilot rollout
A manufacturing mid-sized company has rolled out Microsoft Copilot across the organisation. After a few months, usage is uneven and the benefit hard to measure. The diagnosis shows: the real bottleneck lies in order release, not in the efficiency of individual tasks. There, too many people decide with too little clarity of responsibility.
The Copilot rollout is not merely ineffective at this point. It makes the bottleneck worse. AI output sounds plausible at first glance, but its derivation is barely traceable. Whoever delegates a suggestion takes on responsibility for a result whose reasoning they cannot verify. Order release does not become faster as a result, it becomes less certain.
The recommendation: first clarify the decision architecture. Who decides what, with which information, in what timeframe. Only then deploy AI in a targeted way, where the output remains verifiable.
PATTERN 2 · STRUCTURAL BOTTLENECK IN KNOWLEDGE
Professional services firm with its own GPT instance
A consulting and project organisation sets up its own GPT instance. The executive team's expectation: faster proposal creation, better use of knowledge for presentations and concepts. After a few months, adoption is high among two power users, low across the rest of the organisation. The diagnosis shows: the bottleneck lies in the responsibility for the company's knowledge. No one maintains it systematically and no one feels responsible for it, so the AI finds no data foundation. The recommendation: clarify responsibility for knowledge maintenance before the AI is rolled out broadly.
PATTERN 3 · CULTURAL BOTTLENECK
Family business with experienced specialists
A grown family business introduces an AI-supported tool for case processing, with the aim of increasing efficiency. The technical solution is ready, the training is done. After three months it becomes clear: usage is low, above all among the experienced employees. A few test it, the long-serving carry on as before.
The diagnosis shows: the bottleneck lies neither in the tool nor in the decision architecture. It lies in a legitimate concern among experienced employees that their professional judgement will be devalued if an AI makes suggestions they cannot follow. And worse: they fear being replaced by the AI over the medium term. More training sharpens the problem instead of solving it. What helps here is clarity on the medium-term perspective. What will the job profile probably look like in one to two years, is a reduction planned? It also takes clarity on where experience remains decisive and where the tool relieves the load. The recommendation: together with the experienced employees, define what the AI introduction means for them, what AI may do and what it may not.
Frequently asked questions.
What is Ambiflow?
Ambiflow is a diagnostic and steering model for transformations, developed by Bernhard Nitz. It captures six dimensions of an organisation: bottleneck clarity, value-creation flow, ambidexterity balance, information flow and decision-making, social viability and leadership-system maturity. Ambiflow works fully without AI as a classic diagnostic method. With AI it accelerates diagnostics, steering and optimisation. It is used above all in Swiss and DACH mid-sized companies.
Why does Ambiflow work even without AI?
Ambiflow is a diagnostic and steering logic, not a software product. The method works fully with classic formats: Pulse-Check, qualitative interviews, situational picture. AI accelerates the three levels of diagnostics, steering and optimisation, without being a precondition for them. The method holds for organisations with and without AI maturity.
What distinguishes Ambiflow from classic AI consulting?
Classic AI consulting begins with tool evaluation, use-case workshops or platform selection. Ambiflow begins with the bottleneck question. Only once it is clear where your bottleneck lies and what character it has is a decision made about AI tools. This order prevents pilots that peter out, because the tool is set in the right place. Ambiflow is not an AI implementation provider and not a platform.
We already have an AI pilot. Is it too late for this?
On the contrary: a running or completed pilot is the best starting point. The rapid diagnosis asks what the pilot reveals about your bottleneck. That answer is missing from most reviews and decides whether the next investment hits home.
How does Ambiflow handle the auditability of AI output?
Auditability is not a technical but an organisational question. Whoever is accountable for an AI suggestion must be able to verify the reasoning. Ambiflow therefore separates clearly the points at which AI delivers delegated recommendations (level 3, optimisation) from the points at which human decision-making cannot be replaced (levels 1 and 2, diagnostics and steering). This separation is made explicit in the diagnosis and is a precondition for any AI initiative that is to remain defensible in the boardroom.
When is the right time for a rapid diagnosis?
The diagnosis is not a crisis tool. It works best when a concrete decision is pending: to scale or stop a pilot, to release or hold a budget, to start a new initiative or wait. The more concrete the pending decision, the more precise the diagnosis.
What does the rapid diagnosis cost?
The fee is set as a fixed price after the first conversation, once the context is sufficiently clear. The rapid diagnosis has a clearly defined scope and a clearly defined outcome. On that basis a flat fee is agreed that is fixed before the work begins. The first conversation itself is free of charge.
In which language does transformind work?
Swiss German, standard German and English. The diagnosis runs in the language in which your leadership team works internally. Multilingual teams are served in parallel, in order to make language-related discrepancies visible in the analysis.
Who carries out the diagnosis?
Bernhard Nitz himself. No subcontracting, no junior consultants. This is the boutique principle of transformind: a mandate means working with the person with whom the first conversation was held.
DEEPER LINKS
MODEL
Ambiflow as a model
The six dimensions, four logics of thought and ten principles that underlie the AI application perspective.
ADVISORY
Get to know the adviprocess
How a collaboration beyond the rapid diagnosis is typically structured, from kick-off to anchoring.