Companies are rushing to embrace AI, budgets are soaring, and yet tangible benefits remain elusive. Several recent studies point to the same conclusion: a majority of generative AI projects are costly, mobilize entire teams and, for now, fail to deliver any measurable return. Enough to fuel legitimate skepticism among many business leaders.
But if so many projects disappoint, it is not because AI is useless: it is because they start with the technology rather than the problem to be solved. Real value emerges when we stop asking what AI can do and start looking at where it could concretely improve the way work gets done.
At Altesia, we approach this question from several complementary angles. A year ago, we brought together our consultants, internal teams and several experts in multidisciplinary working groups to drive this reflection and transformation within Altesia. Since then, we have been delivering AI training programs for our clients across Finance, Procurement and Human Resources, transforming and optimizing our own internal processes, and building a dedicated offering around automation, agentic AI systems and business process transformation.
At our latest CFO Network Roundtable, we invited expert Cédric Fumière, who shared four principles that now shape our approach.
1. Start with the use case, not the technology
Not every process needs AI. The most common mistake is to ask “where could we add AI?” rather than “where does it create real value?”. Before talking about tools, you need to identify where improvements in speed, quality, cost or decision-making can make a tangible difference to the business.
2. Context is critical
AI is only as valuable as the business knowledge it relies on. This means that processes need to be documented, data must be available and reliable, and responsibilities must be clear before AI agents can be introduced.
3. Treat the LLM as a commodity
OpenAI, Anthropic, Google or tomorrow’s best model should remain interchangeable components, rather than the foundation on which your entire process depends. The model landscape evolves from month to month; locking yourself into a dependency on a single provider means becoming captive to the technology and limiting your room for maneuver in the future. Best practice is to choose the model best suited to each task while retaining control of your architecture, so that one model can be replaced without having to rebuild everything.
4. Integrate AI into the process
Giving teams access to ChatGPT can improve individual productivity. But truly transforming a business requires integrating AI into workflows, systems and decision points, with the appropriate controls and human expertise around it. This is the difference between opportunistic use (useful but fragmented) and a transformation that is sustainable over time and measurable across the organization.
Conclusion
This is also how we see the next phase of AI: less experimentation for experimentation’s sake, and more targeted transformation of real business processes. And we apply these principles to ourselves before recommending them to our clients.
Want to explore where AI could create tangible value in your Finance, Procurement or HR processes? Let’s talk.
