In a manufacturing plant somewhere in Europe, a predictive maintenance system is running with 95% accuracy. By any technical measure, it works. The team that’s supposed to use it doesn’t. Schedules stay manual. Downtime stays where it was. The value the system was supposed to deliver disappears at the point of non-adoption.
That is the failure mode of most AI initiatives — and the central thread of the first episode of Transformation Compass, the podcast of ISC Paris’s DBA program. In it, Jean-Christophe Lessie, Partner at Boston Consulting Group, former CIO of Veolia (CIO of the Year 2019), and ISC Paris alumnus, sits down with Nabil Ghantous, director of the DBA program, to trace a pattern he has seen across two decades of transformation work in industry.
The pattern has very little to do with the algorithm.
Lessie opens with a frame that should be obvious and is almost never operationalized.
« You can build the smartest AI tool in the world. If people don’t use it, it’s useless. »
The example he walks through is the one that quietly haunts every digital transformation program. A predictive maintenance system, 95% accurate, technically excellent. If the maintenance team doesn’t trust or understand the tool, they don’t use it. If they don’t use it, the value disappears.
The conclusion he draws is short and exact:
« People embrace AI when they trust AI. Trust is really key. »
In most organizations, the conversation about AI adoption skips this layer. It assumes trust will follow performance. It rarely does. Trust is a behavioral outcome, not a technical one. It requires a different kind of effort, with different leadership skills, on a different timeline — and most AI roadmaps don’t budget for it.
The single most underestimated sentence in any AI rollout is the one most executives never hear in the room where the rollout is decided.
« You can’t change anything if employees don’t understand the why. ‘What’s in it for me?’ This is a clear question everybody’s asking when you start to change. »
Lessie’s Veolia case is the demonstration. Veolia is a 160-year-old company — the very definition of legacy. When Lessie joined as CIO, the company launched what he calls the Zero Data Center strategy: turning off 80 data centers worldwide and going full cloud, in 2015. The technical shift was manageable. The cultural one was the work.
« We created an internal marketplace where all 80+ business units worldwide could freely select cloud solutions — we empowered the local teams, leaving the choice to embrace it or not. »
The choice mattered. Imposed adoption tends to produce malicious compliance. Voluntary adoption — especially when it is visible and supported — produces real traction. At Veolia, adoption climbed above 60% within the early phase.
The same principle held on the operational side. Lessie describes the AI rollout in plants and waste sites with precision:
« Our success was not just technology. It was about convincing workers in the water plants and waste sites to use AI — and to convince them that AI is their partner, not their replacement. We trained the people, we celebrated quick wins, with leadership visibly celebrating each success. Over time, AI became really a coworker for those people. »
The Veolia approach reveals something many executives miss when they design adoption strategies. The real lever is not the announcement. It is the visible, repeated reinforcement of small successes.
Three elements ran in parallel.
Training was not framed as a remediation exercise. It was a positioning exercise — situating AI as a partner, not a threat.
Quick wins were not measured solely for ROI. They were measured for narrative value: did they give frontline workers a concrete reason to keep using the tool?
Leadership visibility was the multiplier. Executives showed up at each milestone, attributed credit publicly, and made the adoption itself — not the technology — the focal point of the story.
Most AI rollouts do one of those three reasonably well. Few do all three together, sustained over time. The result, in Lessie’s framing, is that AI never actually becomes part of how work gets done. It stays a feature. It doesn’t become a habit.
« AI became really a coworker for those people. » That sentence is the goal. Everything else is operational scaffolding around it.
Late in the episode, Lessie offers what is arguably the most original framing of the conversation, and one Nabil suggests he should trademark on the spot.
« I call it organizational acupuncture. You identify the pain points and you press on it until it doesn’t hurt anymore. You finger-point your outdated systems, your skill gaps, your silos — and you solve them step by step. »
The image works because it correctly describes what executive transformation actually is — not a top-down program, not a culture deck, but a sequence of focused interventions on specific organizational friction points.
The discipline is dual: locate the friction precisely (which most transformations fail at, because they treat the symptom, not the source), and apply pressure long enough for the system to reorganize around the resolution.
This is a leadership practice, not a campaign. It is repeated, surgical, and patient. It is also where the human factor and the strategic factor finally meet.
The full episode goes further — into the strategic distinction between productivity and transformation, the difference between legacy companies and digital natives, and the focus discipline of the 25% of organizations that actually extract significant value from AI. We unpack that strategic angle in the companion piece, why 70% of AI transformations fall short.
But the through-line is consistent. The AI transformations that work are the ones where the human factor is not an afterthought to the rollout, but the central object of executive attention. Trust, communication, visible leadership, and surgical focus are not soft skills layered on top of a technical project. They are the project. It is also why, increasingly, senior leaders are rebuilding their own research skills — the capability to ask the right question is what decides whether the human side of a transformation is read correctly at all.
