AI doesn't run into problems with technology. It runs into problems with implementation.
Our interpretation of the 2025 studies by McKinsey and BCG, as applied to the reality of SMEs and financial institutions in French-speaking Switzerland.
The adoption of AI is now widespread. Value creation, however, remains rare. This gap is not a problem of models, computing power, or technological maturity. It is organizational, methodological, and cultural.
This is precisely where we come in.
What the Numbers Really Say
Two landmark studies published in 2025, despite using different methodologies, reach the same conclusion.
McKinsey, *The State of AI in 2025*(nearly 2,000 organizations surveyed):
- 88% of organizations regularly use AI in at least one function.
- Only 39% attribute any impact of AI on their operating income, and for most of them, that impact is less than 5%.
- About 6% of respondents identify as “AI high performers.”
- 21% of organizations using generative AI have actually redesigned at least part of their processes. This factor has the strongest correlation with financial impact.
BCG, Build for the Future 2025, and AI Radar(more than 1,250 companies, 1,800 executives):
- 5% of companies generate value at scale.
- 35% are scaling up without yet seeing a significant impact.
- 60% see little or no value in it.
- 75% of executives list AI among their top three priorities, but only 25% believe they are deriving significant value from it.
The conclusion is clear: the gap is not widening between those who use AI and those who do not. It is widening between those who integrate it into their work processes and those who simply add it to their existing systems.
Seven obstacles, three families
Both firms identify the same breaking points. We have grouped them here according to the way they are actually addressed in practice.
The foundation: what determines reliability
- Data quality and governance.Data scattered across business systems, file-sharing platforms, and email inboxes; unmanaged repositories; and a lack of designated owners. Applying AI to ungoverned data produces answers that are both plausible and false—which is the worst of both worlds.
Our take:Waiting for a flawless data estate before getting started is a mistake in sequencing. Data should be handled as a parallel effort, scope by scope, starting with those that support the priority use cases.
- Existing systems:Legacy business applications with limited integration capabilities, sometimes maintained by a single vendor. They don’t hinder experimentation; they hinder scaling.
Our take:The issue isn't about modernizing the entire IT system, but rather identifying the two or three integration points that are critical to scaling up, and addressing them first.
- Governance, Compliance, and Risk.nLPD, professional secrecy, FINMA requirements for asset managers and financial institutions, and the European AI Act for organizations operating in the EU market. In addition, there are requirements for traceability of data processing, handling of hallucinations, and control over data location.
Our take:In French-speaking Switzerland, and particularly in the financial sector, compliance is not a barrier to adoption. It is, in fact, a prerequisite. A clear framework, established early on, accelerates projects rather than slowing them down, because it eliminates roadblocks down the line.
Organization: What Influences Adoption
- A lack of strategy and prioritization.A proliferation of proof-of-concept projects without oversight, without a business owner, and without success metrics. Energy is wasted, budgets are depleted, and value fails to materialize.
Our take:It’s better to have three well-thought-out, carefully evaluated, and properly executed use cases than a dozen initiatives running in parallel. The discipline of prioritization is the primary driver of profitability.
- Skills.While there is a real shortage of technical talent, that isn’t the real issue for a medium-sized organization. The real issue is building the skills of the business teams—the ones who will be using the tool every day.
Our take:Critical thinking isn't about knowing how to use a tool. It's about knowing when to trust it and when not to. It can be learned, guided, and documented.
- Organizational change.Resistance, skepticism about the results produced, ingrained habits, and concerns about roles. These reactions are legitimate and cannot be addressed through top-down communication.
Our take:Adoption is achieved through daily practices, with key users, visible short-term benefits, and the flexibility to make adjustments—not by rolling out a license company-wide.
Execution: What Determines Value
- The transition from pilot testing to measurable impact.This is the most common sticking point. Pilots are successful, but then nothing happens—due to a lack of an owner, a budget for full-scale implementation, or metrics defined before launch.
Our take:A use case without metrics defined in advance can never be justified to the executive committee. The question isn’t “Does it work?” but “What does it change, for whom, and by how much?”
On the Path to Maturity
The data organization models described by the two firms—ranging from centralized systems to decentralized architectures such as data mesh—outline a relevant path forward for large corporations.
We would like to make a useful caveat here: for an SME in French-speaking Switzerland or an independent wealth manager, aiming for the final stage of this continuum makes no sense. A well-managed centralized system, with a clearly identified owner and simple governance, generates more value than a sophisticated architecture without the resources to sustain it.
Useful maturity is the level of maturity that the organization is capable of sustaining over the long term.
What We've Learned
None of the seven obstacles is technological. All relate to methodology, governance, and human support.
This conviction underpins our approach:humans enhanced by rigorous judgment, not the other way around. AI expands our teams’ capacity for analysis and execution. It does not replace judgment, accountability, or industry expertise.
Our IAgérance support offering is based on this approach and on a short, deliberately low-key process:
Assessment → Workshops → Roadmap → Deployment → Optimization
Short stages, limited scopes, and verifiable results at each iteration. That’s what sets apart the 5% of companies that create value from the 60% that see none.
The Next Step
If you recognize yourself in any of the seven obstacles, the first step is rarely a project. It’s a structured conversation about your actual situation: your data, your regulatory constraints, your business priorities, and the maturity level of your teams.
We offer this framework in the form of a brief assessment, at the end of which you’ll have a clear understanding of your priority use cases and the conditions required to implement them.