Do you know what the first question our prospects ask us is?
We’ve installed ChatGPT, Copilot, we’ve run a few tests… but now, how do we scale this across our organization?
That’s almost word for word what we hear, week after week, ever since generative AI became mainstream. And it highlights a phenomenon that is well known in the IT world: the gap between a consumer-grade tool and an enterprise-wide deployment.
What we observe in the field is confirmed by the data. According to the GenAI Impact Report 20261, nearly two-thirds of generative AI projects in Switzerland are abandoned after the pilot phase.
The problem isn’t the technology
It would be easy to conclude that generative AI has failed to deliver on its promises. It would also be wrong.
The same report shows that companies that successfully move beyond the pilot stage achieve tangible business results. 53% of surveyed organizations report revenue growth of at least 5% thanks to AI. In software development, measured productivity gains average 35%, while one quarter of companies achieve improvements exceeding 50%.
The question, then, is not whether AI can create value. It can. The real question is why two out of three companies fail to get there, and how to become the third that succeeds.
Putting the cart before the horse
Most organizations discover generative AI through an impressive consumer tool demonstration and only then start looking for ways to apply it within their business, instead of starting with their business needs.
The problem is that there’s a world of difference between testing ChatGPT on a personal computer and deploying an AI solution across an entire organization. Enterprise deployment means dealing with proprietary business data, security requirements, access rights, business processes, and governance…areas that consumer AI tools were simply never designed to address.
At Deeplink, we started by mapping real business challenges, industry by industry and role by role. What are the genuine pain points of an SME, a public administration, a service desk, or an association?
This work led us to develop AI solutions built around clearly identified business use cases, not around technological capabilities looking for a problem to solve. As a result, every client can quickly determine whether a particular use case addresses a real operational challenge before we even start discussing tools or platforms.
Only after this discovery phase do we move on to the demonstration. In that order, not the other way around.
Bring teams on board instead of convincing them afterwards
The second recurring mistake is deploying a solution to employees rather than with them. Even an excellent tool will be bypassed if it is imposed from the top down. Resistance to change is rarely about unwillingness, it is about meaning, ownership, and involvement.
Our deployment methodology consists of six phases, designed so that the workload is shared between our teams and the client’s teams. We bring the methodology, the technical expertise, and the infrastructure.
Our clients bring something no AI tool can replace: their knowledge base, documentation, business data, and the expertise accumulated by their people over the years. That is the real asset we leverage and place at the heart of every solution.
The end result is not a generic product with a few customizations. It is a solution built together with the people who will use it, validated by them, and designed to deliver value long after the pilot phase is over.
What we’ve learned
The Swiss generative AI market is entering a new stage of maturity. Organizations that have experimented are now separating lasting business value from impressive demonstrations. That is good news for companies that approach AI deployment with a structured methodology.
The end of naïve experimentation2 marks the beginning of AI that truly delivers.
- ICT – GenAI Impact Report 2026, Adesso Suisse (enquête auprès de 100 cadres dirigeants), juin 2026 ↩︎
- Le Temps, « Finie l’expérimentation naïve », citant la 3ᵉ édition de l’Observatoire de la Data et de l’IA en Suisse (Colombus Consulting, Oracle, HEG-Genève, 23 juin 2026 ↩︎