While we are inundated with announcements about AI enabling instant revolution and massive productivity gains, few reports take a sober view, and this is precisely the case with the 2025 edition of ServiceNow’s Enterprise AI Maturity Index.
While most studies seek to highlight the speed of adoption, the number of use cases, or productivity gains based solely on self-reported data, this report focuses on an interesting indicator: maturity. And this maturity, according to data collected from nearly 4,500 businesses worldwide, is not only low, but in sharp decline compared to last year. This finding could be cause for concern, but in my opinion it is actually a healthy return to reality.
In my view, this decline should not be seen as a failure but rather as a symptom of a rather salutary awakening. Businesses are discovering, sometimes brutally, that adopting AI is not enough to transform themselves, and that between enthusiastic experimentation and sustainable transformation, there is an organizational, managerial, and cultural chasm.
It is this gap that the report highlights without attempting to sugarcoat it, and this is precisely what makes it useful reading for those who want to go beyond the promise and think about the real conditions for implementing artificial intelligence at the business level.
One last point before getting to the heart of the matter. I often say that you only need to read the name of the author or sponsor of a study to guess what it says, and this one is no exception to the rule. ServiceNow has long been the undisputed leader in Robotic Process Automation, so it’s no surprise to read that the future lies in agentic AI. But for me, this is obvious, and at least ServiceNow’s expertise and perspective on the subject is a welcome change from the charlatans who try to wrap anything and everything in AI and believe thatthey are going to change the world with flashy marketing. I believe that a peak was reached at the last Vivatech, with many visitors admitting to being disappointed by the number of projects that seemed opportunistic, led by founders with little expertise or with unproven economic viability. I would also add that at a time when it is becoming clear that many publishers are trying to present their figures in a favorable light so as not to show that “it’s not selling that well” and that revenues are far from covering costs, ServiceNow seems to be the most honest and transparent player on the subject, even finding favor, which is rare, with the highly critical Ed Zitron (The Hater’s Guide To The AI Bubble).
In brief:
- The 2025 report from ServiceNow highlights a decline in the average AI maturity score (from 44 to 35 out of 100), interpreted as an awareness of the real challenges rather than a failure of AI.
- Businesses are realizing that adopting AI is not enough: structures, governance, skills, and processes must be fundamentally transformed to reap lasting benefits.
- Agentic AI, capable of orchestration and autonomous execution, is identified as the next step, but only a few businesses have begun concrete projects and are ready to effectively manage it.
- A minority of businesses (“pace setters”) stand out for their integrated vision and implementation capacity, but their example does not guarantee widespread transformation of the economic fabric.
- It could be suggested that the report calls for slowing down in order to structure, emphasizing that the value of AI depends above all on how it is conceived, framed, and integrated into organizations.
A sharp but welcome decline
The figure may be surprising or even worrying, but the fact is that the average AI maturity score has fallen from 44 to 35 out of 100, a relative decline of more than 20%. This could be interpreted as a wake-up call, but I prefer to see it as a sign of maturity in the sense of an awareness of the real issues at stake. Admittedly, this awareness has come late, but we can hope that it will be beneficial, and this setback, which may seem worrying at first glance, is perhaps the healthiest thing that has happened since the emergence of the generative AI wave.
We are finally ceasing to confuse adoption with understanding, piloting with deployment and, above all, transformative potential with actual transformation. We are returning to a level of lucidity that forces us to ask the right questions without getting carried away by the initial euphoria that masked the real issues.
Unlike other reports, which are often overly enthusiastic and written to reassure or attract customers and investors, this one takes on an almost disenchanted tone, not because AI does not work, but because the vast majority of organizations have underestimated the complexity of its integration. Maturity is not measured by the number of use cases or the number of prompts generated, but by the ability to align strategy, governance, skills, and operating models in a systemic and sustainable approach. And this is precisely where many businesses are lagging behind, which, strangely enough, they did not anticipate.
A race for innovation that transcends structures
The study in no way questions the quality of recent technological advances. On the contrary, it highlights the rapid emergence of what is known as agentic AI, a form of artificial intelligence embodied by autonomous agents capable of acting proactively to achieve defined objectives. This takes us beyond the realm of text and image generation and into that of orchestration, automation, and execution.
But while the technology is evolving rapidly, businesses are not keeping pace in terms of their ability to take advantage of it. Only 33% of respondents have started concrete projects involving agentic AI, and among the 40% who say they want to launch such projects in the next 12 months, a majority admit that they do not have the necessary safeguards in place, whether in terms of governance, security, or simply clear evaluation criteria.
This is a classic pattern that experience should allow us to anticipate, namely that innovation is advancing faster than the ability of the structures adopting it to absorb it. It is not so much a lack of will as a lack of vision, the ability to prioritize, to provide a framework, and to translate a technological intention into organizational transformation. This phenomenon is not new, as it has accompanied every wave of digital transformation for decades, but it is taking on a scale rarely seen before, due to the complexity of AI, its non-deterministic nature (Businesses are deterministic, generative AI is not, and that’s a real problem.), and the methodological vagueness that still too often surrounds how to measure its value.
It’s not AI that needs improvement, it’s the organization
The main message of the study is that AI, in its various forms, works, but what doesn’t work, or doesn’t work yet, are businesses that seek to deploy it without reviewing how they think, structure, and manage their operations. This is in line with a recent BCG study on the subject (Generative AI: your organization is worth more than your tools).
The ServiceNow study shows this very clearly: less than 30% of the businesses surveyed have a clear vision of the skills needed to leverage AI at scale. Most have not defined an impact model, do not have shared operational indicators, and have not engaged in any meaningful training efforts or adoption policies. And when training does exist, it is often limited to tool usage, without any real acculturation to the underlying issues.
In many cases, we therefore remain in an experimental, opportunistic, and sometimes disordered mindset, which generates both skepticism and fatigue. If this fatigue sets in, it risks undermining the very credibility of AI initiatives in the coming months.
As Frédéric Cavazza points out in several of his posts ([FR] We don’t need better AI, we need a better understanding of AI and [FR]We don’t need better models, we need better products), the central issue is not the intrinsic performance of models, but their integration into a work environment that is understandable, manageable, and evaluable. What we need is not better AI, but better products, better processes and, above all, an understanding of what we are really dealing with.
The Pacers: trailblazers, not the norm
Of the 4,500 businesses surveyed, only 18% demonstrated a higher level of maturity and tangible results. These are the “Pacesetters” that the report defines as businesses that combine strategic vision, strong governance, a platform-oriented approach, a skills development policy, and the ability to implement at scale, and which are effectively coming out on top.
But should they be held up as a model? I doubt it. The history of technology in business shows that this vanguard has always existed, and that it has never alone guaranteed widespread transformation.
The real issue is diffusion: how quickly will the rest of the business world adopt the same levers, and to what extent? If, in five or ten years’ time, only these 18% have managed to transform their businesses, then we will have to admit that AI, far from being a shared revolution, will have served mainly to widen the performance gap between those who already knew how to transform and those who did not.
AI agents: the next big thing or the next mirage?
What makes this report more credible than many others is that ServiceNow is not an opportunistic player in generative AI, but an automation specialist whose DNA is rooted in workflows, processes, and streamlining information flows. When they talk about agentic AI, they do so not to impress, but to maintain a consistent trajectory.
ServiceNow does not question generative AI, but through the importance it places on agentic AI as the next step in industrializing uses and managing complex systems, we understand that generative AI, while useful for initiating momentum, cannot alone support sustainable transformation. This is also, even more clearly, the conviction that I share: generative AI, however spectacular it may be, will remain a transitional stage, whose limitations already call for something else. It will need to be combined with more structured and structuring forms of AI that are goal-oriented, capable of making decisions, coordinating actions, and optimizing systems. It is this type of AI that is likely to bring real operational differentiation, as demonstrated by Moderna’s recent initiative (HR and IT merger: Moderna redesigns its organization for and with AI).
But this requires prerequisites such as a reliable data architecture, solid governance, competent human supervision, and above all, the ability to rethink work beyond the tool logic.
The risk would be to expect agentic AI to fix the mistakes of generative AI, even though it is part of a higher level of complexity.
Bottom line: slow down to transform better
The decline in the average AI maturity score is not a failure but a step forward. It marks the end of a phase of illusion and the beginning of a more lucid, more structured, and undoubtedly more effective cycle in the long term. Businesses are not lacking in technology, but they are lacking structure and method.
What this report tells us is not to do things faster, or even to do more, but to do things better. To do this, we must accept to slow down, clarify, and lay the foundations. AI is not the solution in itself; it is only a means, whose value depends entirely on how we use it.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
Related articles:
Generative AI in the workplace: revolution or illusion?
Why enterprise AI can’t keep up with consumer AI: beyond ChatGPT, a more complex reality
Will AI replace juniors? The false debate that’s only the tip of the iceberg
You don’t need to trust AI, you need to trust the people who train it.
AI heading for an economic dead end?
Generative AI: your organization is worth more than your tools
