Agentic AI — S1 · 2/12: Why many projects will disappoint
Many agentic AI projects will disappoint in the coming years. And it won’t be because of the technology.
We talk a lot about the revolutionary potential of AI, but I worry that for many organizations, the results will fall short of expectations. My intuition is actually reinforced by the recent Bain survey of 133 life sciences executives: only 20% of organizations are deploying AI consistently at scale and generating measurable value. This is consistent with the feedback I have gathered over time. Even in large groups that were among the first to launch AI initiatives, the measurement of ROI often remains approximate.
For several years, software vendors have been announcing AI integration in their roadmaps. What is available today often remains embryonic. Why? Because there is still a lack of deep understanding of clients’ real needs, the ability to translate those needs into actionable prompts, and the precise identification of process segments that would genuinely benefit from AI.
Take the regulatory field. Many dream of an agentic AI capable of automatically compiling data and drafting marketing authorization dossiers. The reality is quite different. The major bottleneck is not the technology: it is the ability to keep data high-quality, well-structured and accessible to agents, consistently over time and across multiple geographies. No agent, however capable, compensates for data nobody took the trouble to structure.
This is why I believe the real disappointments in the coming years will mainly come from:
- a lack of alignment between leadership and operational teams
- underestimation of the work required on data quality and structure
- the absence of clear governance around use cases and agents
- an overly technological approach that neglects processes and organizational change
The organizations that will succeed are those that treat agentic AI not as just another technology layer, but as a true transformation project: deep process understanding, data cleansing and structuring, team enablement, and appropriate governance.
This is what responsible leadership looks like: having the courage to slow down on marketing promises in order to better prepare the organization, invest in data and process quality, and genuinely support teams — rather than chasing quick wins at the expense of long-term sustainability.
What do you see as the main barrier to successful agentic AI deployment today?