Why Your AI Pilots Fail While Your People Are Already Ahead

AI is already at work in your company. Someone in sales has offers drafted for them, accounting is testing a tool for capturing receipts, and perhaps a pilot has already run its course. And yet, when the board asks about the AI strategy, the room goes quiet. Activity is high, progress is not.

The Swiss studies from spring 2026 show this pattern with rare clarity. And they point to a cause that sits on very few executive agendas.

AI use in Swiss SMEs: widespread, but unsteered

Three surveys from recent weeks paint the same picture. The EY AI Survey 2026 of 604 respondents from Swiss companies finds that 89 percent use AI in their working day, while systematic scaling and strategic anchoring remain at an early stage in most organisations. A representative UBS study of around 2,500 companies of all sizes and sectors measures at company level: six in ten Swiss firms deploy AI, and the finding condenses into a single sentence, that many firms use AI and few do so systematically. For SMEs, the same study notes that they use AI less often and less comprehensively, because structured data and the necessary IT infrastructure are frequently missing. And in the SME barometer run by Kalaidos University of Applied Sciences for the NZZ, almost every second executive counts AI among the most important topics of the coming three years, while the same survey records considerable catching-up to do on the strategic use of data: technology decisions are often made from the gut.

The pattern holds internationally. In McKinsey's State of AI from November 2025, 88 percent of companies report regular AI use in at least one business function. Close to two thirds have not yet begun to scale across the enterprise, and only 39 percent see an effect on operating profit. Adoption has arrived, impact has not.

We read this gap as a steering gap. Introducing AI changes processes, roles and decision paths, and it demands the same leadership work as any other transformation.

Why AI projects fail: the bottleneck sits in the leadership system

How this steering gap arises is shown by BCG's study AI at Work 2025, with more than 10,000 employees across eleven countries. 72 percent use AI regularly, carried by leaders and managers. Among employees without a leadership role, regular use stalls at around half, and the study names the reason: only a quarter of these employees experience sufficient support and orientation from their own leadership. Where that support is present, the share of employees with a positive attitude towards AI rises from 15 to 55 percent.

The same study's trend data sharpens the picture: within a single year, regular use among managers rose from 64 to 78 percent, while at the front line it barely moved. The leadership levels acquire the tool for themselves, and in many places that is where the leadership work ends. Using it yourself is not yet steering.

This is the core of our diagnosis: the bottleneck of an AI rollout sits in the leadership system, in the routines with which the executive team sets priorities, makes decisions and holds initiatives over months. We therefore work with leadership first: on decision paths and on the cadence of steering. The technology follows.

Prioritising AI investment in the SME: diagnosis before budget

What a missing diagnosis costs can be read from investment patterns. The UBS study describes selective use without a system, the SME barometer gut decisions on technology questions. In daily practice this means budget flows into what is visible and obvious, a chatbot here, a pilot there. Whether the investment addresses the place that actually slows the company remains untested.

We work with the notion of the bottleneck: the one place in the company that limits the throughput of the whole, whether in order processing, in decision-making or in sales. Change work takes effect when it addresses the bottleneck. For AI investment this shifts the guiding question. Instead of asking which use cases exist, it asks what slows our system most, and whether AI can take effect precisely there.

In the system diagnosis of Ambiflow, the diagnostic and steering framework from transformind, this question is the first step of the work. It subordinates the AI discussion to a bottleneck hypothesis and protects the budget from the most expensive form of failure: twelve months of pilotitis, a sequence of isolated attempts that bind attention and change nothing.

Shadow AI in the company: the raw material of your AI strategy

A second finding from the BCG survey concerns the daily reality of every executive team: more than half of respondents would use AI tools even without IT approval, what the study calls shadow AI. Only 36 percent consider their training sufficient, and 18 percent of regular users have received none at all. The workforce has opened up the technology on its own: without a mandate, without a budget, without guidance.

The common response is a policy that forbids, or a silence that tolerates. Both squander the raw material of an AI strategy: existing exploratory energy. Together with the executive team, we build a guided exploration space from it: clear guardrails on data and confidentiality, a vessel in which experience is shared, and a connection to enterprise steering, so that isolated findings become decisions. In this way shadow use becomes organisational learning, and a compliance topic becomes a piece of strategy work.

Resistance to AI: a rational signal to leadership

That leaves resistance, which many executive teams file under communication problems. One figure from the latest CFO survey by Deloitte Switzerland explains why training sessions and town halls change little: almost half of the CFOs surveyed (46 percent) plan to reduce labour costs through the use of AI over the medium term. Employees read such signals precisely. Whoever responds with scepticism is drawing a rational conclusion from what they observe.

Resistance is therefore one of the most valuable sources of information in a transformation. It shows how much trust the leadership system has built or spent, what experience the organisation has had with earlier initiatives, and where commitments and reality diverge. In our diagnostic work, this social viability takes priority: with instruments such as the Pulse-Check we establish how much change energy, trust and exhaustion are actually present before an initiative is set up. The answer to resistance is work on the leadership system: reliable decisions, commitments kept, priorities that hold up, and a leadership that listens before it persuades.

AI transformation with Ambiflow: the leadership system first

From these findings comes the sequence with which we accompany AI initiatives. The basis is Ambiflow. The framework considers six dimensions of a company: bottleneck clarity, value-creation flow, the balance between day-to-day business and renewal, information flow and decision-making, the social viability of change, and the maturity of the leadership system that carries all the other dimensions.

For an AI rollout this means, concretely: first we clarify whether the organisation can carry another change. Then we work out the bottleneck hypothesis that gives the investment its direction. And across the whole duration we work with the executive team on what ultimately decides the impact: steering routines with a fixed cadence, clear decision roles, support beyond the pilot with review and readjustment after six and twelve months. We work on the patterns and structures that produce behaviour.

Three questions serve as a first self-diagnosis:

1. Can you name and justify where the most effective entry point for AI lies in your value creation?If the answer is a list of use cases, the bottleneck hypothesis is missing, and the budget follows visibility.

2. Do you know who in your company already works with AI today, and with what?  The larger the gap between assumption and reality, the more exploration takes place outside any form of steering.

3. Would your workforce trust the executive team to carry an AI rollout through consistently over twelve months?  The answer says more about your leadership system than about the technical affinity of your employees.

Whoever hesitates here has found the starting point. It lies in the leadership system, and the work on it carries further than any single technology rollout.

AI assessment for executive teams and boards

A realistic entry point is an assessment with limited effort. The rapid diagnosis from transformind delivers a sound bottleneck hypothesis and a one-page situational picture within ten working days, at around two and a half hours of the management's own time. The result shows where your system stands, where AI could take effect, and which conditions in the leadership system need to be created first. More about AI-Transformations

For board members, the question sits one level higher. The swissVR Monitor, the semi-annual survey by swissVR, Deloitte and the Lucerne University of Applied Sciences, has documented two things since its focus edition on generative AI: in many boards the expertise is missing, and regular reporting by the executive team on AI use barely takes place. In the spring 2026 edition, board members name efficiency gains as the greatest opportunity of the coming twelve months. Oversight of the executive management is among the non-transferable duties under Art. 717 of the Swiss Code of Obligations, and under the standard of the Business Judgement Rule, decisions enjoy protection where they were taken on an informed basis, free of conflicts of interest, and through a comprehensible procedure. This calls for less expertise than judgement: the right questions to the executive team and a documented formation of one's own view. This is precisely where Sparring takes effect, a confidential exchange among equals, independent and outside one's own organisation.

The tools get better, the prices fall, the providers change. What remains is the question of whether your leadership system can carry a rollout. That question can be settled before the next budget is approved. Request the rapid diagnosis.