Why AI initiatives always fail for the same reasons

The failure of artificial intelligence projects is often presented as the result of a problem of maturity, skills, or even governance (Without governance, the gains from AI are virtual). The causes vary, but when we look at initiatives that are halted or run out of steam, a recurring theme emerges.

These failures often go unnoticed and are more often expressed by a slowdown and a gradual shift towards more modest uses, which are safer because they are less engaging. But it would be wrong to conclude that AI is being rejected. I would rather say that it is contained where it benefits individuals without overly questioning the organization (What works today with AI without any particular effort).

In short:

• The failure of AI projects often has less to do with technology than with a lack of governance, sustained engagement, and clearly defined accountability within organizations.
• Many initiatives are launched without a clear leader or purpose, leading to a gradual loss of momentum and a reduction in initial ambitions.
• AI frequently comes into conflict with existing structures, roles, and decision-making processes, causing resistance or silent neutralization.
• Inter-team coordination, which is necessary for real impact at the business level, is often avoided because it reveals latent organizational tensions.
• The gap between the fast evolution of AI models and the pace of decision-making or measurement of results in businesses slows down the effective integration of AI projects.

Projects without real leadership

Many AI initiatives are launched with real support, often at the highest level, but this support often remains purely symbolic. It is expressed at launch in the frenzy to do “things with AI” at all costs, but rarely lasts over time and even less so in terms of setting an example. However, uses that involve the organization require decisions that are sometimes uncomfortable in terms of intentions and responsibilities (Taking back control of enterprise design: intention before tools and AI First is not AI Only: clarify your intentions before transforming your business).

In the absence of governance and decisions, the project continues in practice but is emptied of its substance. It becomes a simple tool isolated from operational reality, carried by a few individuals who keep it going as long as they can, then gradually run out of energy. The initiative is not stopped but becomes secondary.

But even when there is a designated leader, there is a risk that a lack of understanding of the real issues means that they are not the right person for the job (Who is handling your artificial intelligence projects? Probably not the right people.).

A responsibility that is never truly assumed

Another recurring theme concerns responsibility. As long as AI remains confined to an assistant role, the issue is secondary, but as soon as its use influences a decision that engages the business, this ambiguity becomes a problem.

Many projects fail at this turning point because no one is prepared to fully assume the consequences of a partially automated decision. In this context, the simplest solution is often to reduce the scope of AI until it no longer poses a political problem, because this is indeed a political issue.

Uses that clash with the existing organization

AI initiatives are often thought of as additions, rarely designed as elements that could challenge the existing balance. And, logically, when they begin to produce a tangible effect, they inevitably clash with the organization as it already functions.

But this is not a question of technology, but rather of scope, roles, decision-making processes, and, more broadly, the (re)design of work. Who decides what, who arbitrates between conflicting objectives, who accepts to see their expertise partially challenged? When these questions remain unanswered, AI logically becomes an irritant, and the organization reacts as it knows how to do very well, or even as it was designed to protect itself from any external body that, like a virus, disrupts its functioning. It absorbs, neutralizes, and sometimes rejects.

Perhaps even worse, AI, like other technologies before it, can lead the business in an unintended direction simply through organizational passivity (If your business isn’t designed for AI, it will end up being designed by AI).

Coordination that no one really wants

Many initiatives fail when explicit coordination between teams, functions, or entities becomes necessary, because it is important to remember that for individual gains to translate into collective gains that are tangible at the business level, the approach must be based not on individuals or even teams, but on end-to-end workflows (Collective appropriation of AI: the only condition for tangible impact). As long as its use remains local, the issue can be avoided, but once it becomes cross-functional, the question of coordination becomes inevitable.

However, this is never neutral. It highlights disagreements and differences in priorities, not to mention power games. Many AI projects fail because they require a level of coordination that the organization usually only mobilizes in crisis situations.

Without coordination, or at least effective coordination, the project often reverts to a scale that is more tolerable for the organization.

Temporality issues

AI introduces a relatively new temporality, with models that evolve quickly on the one hand and benefits that are slow to materialize on the other ([FR]OpenAI identifies a growing gap between AI model capabilities and real-world uses). This temporality is at odds with structured organizations that act and make decisions over the long term while measuring results over the short term (quarterly) (“If the rate of change on the outside exceeds the rate of change on the inside, the end is near.” – Jack Welch and When the pace of management is slower than the pace of business…).

Many projects fail because they simply cannot find their place in these cycles. They are too slow to demonstrate their value quickly, while at the same time, models evolve almost every month.

However, for a business to achieve its goals or, at least, get what it expects from its project, it must go through the necessary stages: adoption, then productivity, then revenue. But this takes an almost unavoidable amount of time, and if that time is not granted, the initiative will be halted midway (Technologies sell productivity, but businesses want revenue and AI from productivity to P&L: nothing happens by chance).

Here again, the most common solution is to reduce the ambition, even if it means losing the initial interest.

Conclusion

If AI initiatives often fail for the same reasons, it is not because of deficient technology but because they come up against deep-rooted structures, reflexes, and sometimes even corporate cultures. They put established operating methods under stress without the organization having explicitly decided to change them.

In this context, failure is not an accident but the logical result of a mismatch between what AI makes possible and what the business is prepared to take on if it has not thought through its project and its possible consequences at length beforehand. Projects disappear not necessarily because they fail, but because they go beyond what the system can absorb.

This observation does not call for an immediate solution, but simply helps us understand why, behind the apparent diversity of situations, the same scenarios repeat themselves and why any discussion of the AI First business cannot avoid reflecting on these fault lines.

To answer your questions…

Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)

Related articles:

Adoption and impact of AI: lessons (and limitations) from the latest McKinsey and BCG studies AI First: lots of debate and few answers What works today with AI without any particular effort Two interpretations of AI in the workplace: agentic uses and capability overhang AI adoption does not replace productive appropriation AI from productivity to P&L: nothing happens by chance