AI has never attracted so much interest. Yet behind the hype lies a more bitter reality: 95 % of enterprise AI projects fail. François Madjlessi, currently Director of Digital at Savoie Mont Blanc University and a member of the Information Governance and AI programme at ESSEC, shares his views on the reasons for these failures, the conditions for success and the steps involved in building a genuine corporate AI strategy.
Why do so many AI projects fail?
Every year, ESSEC surveys major French companies, particularly those in the CAC 40 index. The findings are always the same: incredible enthusiasm for AI, but few concrete successes at organisational level.
AI brings individual power, but not yet collective power. We're talking mainly about personal use - an augmented employee who does a search faster or writes a report more efficiently - but we're seeing very little transformation on a corporate scale.
And so, 95 % of AI transformation projects fail (source: MIT, July 2025 & ESSEC AI Governance Observatory - September 2025). The paradox is striking: AI is everywhere, but its overall impact on performance drivers remains weak.
The main difficulty lies in the scaling up collectively, In other words, the ability to turn AI into an engine for transforming the entire organisation.
When I was digital director at Paris Dauphine, we launched a French version of ChatGPT for teachers and students, to personalise learning. The project worked technically, but it came up against resistance to change. Teachers saw it as a challenge to their methods. Only 5 %s were enthusiastic.
Lessons learned All too often, the starting point is technology, without considering the impact on processes and businesses.
What are the causes of failure?
There are four main causes.
1. A disconnect with the business
An AI project cannot succeed without a thorough understanding of internal processes. Understanding how the business actually works is an essential prerequisite for a company's AI maturity.
2. Unclear objectives
Many managers say they want to «do AI». But AI is not an end in itself. The objective must be clear: what value does the company want to create? What vision does it want to achieve?
3. Poor data quality
AI is only effective if it is based on reliable and well-governed data. Without this, algorithms produce inconsistent or biased results.
4. Shadow IT
How can you have a coherent corporate approach when 95 % of employees are using non-referenced tools? This fragmentation slows down any structured initiative.
What do you need to make it work?
1. Choosing the right use cases
Success depends on selection of relevant pilot projects, where AI brings real value. You have to start small, quickly demonstrate a concrete impact, and then generalise.
The most effective AI is that which simplifies employees' lives without changing their habits. For example, automating the transcription of voice into reports is a simple, high-yield use case.
2. Establish clear governance
Successful AI projects are validated by the company chairmanship and supported by the Comex. This sponsorship is essential to get all directions on board.
3. Assessing the company's maturity
Maturity varies greatly between organisations. A large company like TotalEnergies does not have the same digital culture as an SME. The approach therefore needs to be adapted to the level of internal preparation.
4. Relying on external partners
The success stories we have seen are often those of companies that have worked with external specialists capable of understanding their business and rapidly industrialising solutions.
5. Keeping people at the centre
AI must never become a black box. You have to retain control, interpret the results and never pass the buck to the machine. AI is a facilitator, not a manager.
How do you involve management and the business lines?
The starting point is the conviction of the general management that AI can create value. Next, we need to :
- Commissioning a multi-disciplinary team to identify eligible professions;
- Appointing AI ambassadors in each direction;
- Quick, practical demonstrations to create membership.
In a compliance laboratory, we saw an AI agent comparing old and new standards: a month's work saved! This kind of concrete success changes perceptions.
Thanks to these ambassadors, the company can break down the barriers between its departments and create a collective dynamic.
An inspiring example of an AI project: the case of the SICAME group
The SICAME energy group asked a simple question: «Can AI transform a business like ours?»
The response was a structured action plan: creation of’AI ambassadors, development of a some sixty specialised agents, and setting up a IA Innovation Committee.
Each business unit has its own agents, who can be pooled across entities. From community managers IA promote best practices internally via a Ghost platform, by driving innovation marketing. Every three months, a review of use cases is shared with Comex.
As a result, AI has become a strategic focus business information and a a lever for industrial transformation.
What advice do you have for a company just starting out?
I distinguish five levels of AI maturity :
- Level 1 Use of AI on an individual basis (personal productivity).
- Level 2 Use of RAG (Retrieval Augmented Generation) with internal documents, including confidential documents.
- Level 3 creation of specialised agents.
- Level 4 agents integrated into the information system.
- Level 5 The «automatic agentique», capable of processing millions of e-mails.
AI is particularly effective for repetitive tasks: technical support, compliance, document management, etc.
Another essential tip for your AI project :
Involve an external partner from the outset. There are too many parameters to master to succeed alone. Once the skills have been acquired, you can gradually bring them in-house.
What about data sovereignty?
As an expert from the academic world, I insist on digital sovereignty.
The data must stay with you. We can use American technologies, but the servers and processing must be hosted locally.
It's a question of RGPD compliance, but also trust. It recommends building sovereign infrastructures, for example by relying on European hosting providers such as OVH and local start-ups capable of developing secure RAG solutions.
Who are your training courses aimed at?
In my seminar « Innovating and transforming your business using data and AI (traditional or generative)».», I have three main types of audience:
- from IT professionals, who need to understand the new technical and strategic challenges;
- from business line managers, They are looking for concrete examples of how they can transform their practices;
- and regulatory experts, and the ethical and legal challenges of AI.
What I give them are real-life use cases, successes and failures, to help them understand how to tackle the transformation. It's also a situation-based approach, starting with their problems and tackling all the business and technological issues at stake.
We explore the market players (Mistral, Microsoft, Google, Meta...), the organisational and regulatory aspects (AI Act, Data Act), as well as the implementation of AI and its evolution. All illustrated by numerous examples.
What do you need to remember for a successful AI project?
A successful enterprise AI project is first and foremost about a history of governance and maturity.
The 5 % of successful projects do not seek to using AI, but to creating value through AI, This is based on the conviction of management, collaboration between business and technology, and a progressive and sovereign approach.
AI does not magically transform a company. It is the men and women who support it, understand it and adapt it to their reality that make the difference.
