Artificial intelligence is no longer a technology topic. It has become a matter of deep transformation for the company, and HR plays a central role in it. Here are the levers to pull to move from watching to acting.
The context: why do 95% of AI projects fail?
AI adoption in companies is no longer a question of “if”, but of “how” and “how fast”. 95% of AI projects in companies fail, according to a recent MIT study, but 5% of organizations succeed. And their secret doesn’t lie in the technology.
“Rolling out AI isn’t a technology project. The technology exists and it works. It’s 10% tech, 30% quality data and 60% culture, organization and people.” — Didier Girard, Managing Director at SFEIR & WEnvision
The AI transformation: a matter of skills and organization
AI now touches every white-collar job (finance, marketing, sales, support), and more and more blue-collar jobs too, through conversational interfaces and “digital twins” of factories. It is redefining job descriptions, the skills expected, the organization of work, and even the way people manage.
For HR, five challenges are emerging: redefining jobs, supporting AI literacy, preventing new psychosocial risks, setting a legal framework for uses, and bringing the whole organization on board, employee representatives included.
1. HR strategy: moving past the myth of a purely technical project
The first mistake is to leave AI entirely to the IT teams. This reflex is still deeply rooted in organizations:
“The people we talk to are often tech decision-makers. Our job is to turn that around and bring HR representatives back into the conversation as fast as possible.” — Marie Fontaine, Head of AI Strategy at WEnvision
Some companies understood this very early. Moderna, for example, merged its HR department with its digital department, considering AI to be a human topic first and foremost. A strong signal that the performance of AI projects depends less on the tools than on an organization’s ability to build them into the way it works.
2. Becoming the architects of work
This is undoubtedly the lever that shapes everything else. HR now has to rethink the organization of work in depth: which tasks to give to “human staff”, and which to delegate to “AI staff”?
“We’re moving to a Skill-Based Organization. What we expect from a person today won’t be the same tomorrow. The skills attached to a position have to be updated regularly.” — Didier Girard, Managing Director at SFEIR & WEnvision
In a Skill-Based Organization, work is no longer defined by fixed positions (“jobs”), but by the skills needed to carry out projects.
Three concrete workstreams for HR
This redefinition takes three workstreams: updating job descriptions continuously, spreading a culture of using AI, and investing heavily in training.
It is very important to have a framework in which employees are allowed to make mistakes:
“HR has a role to play in creating a zone of psychological safety: the right to make mistakes, charters of good practice for experimenting. Without that, nobody will really dare to use AI.” — Marie Fontaine, Head of AI Strategy at WEnvision
These three workstreams start with HR leading by example: discover the 10 essential AI use cases for HR, to anchor AI in your own tasks before extending it to others.
3. How do you frame shadow AI and encourage adoption?
A large share of employees already use AI, often in free versions and outside any framework. ChatGPT open in a hidden tab, prompts improvised on the fly, customer data or internal documents copied and pasted without a filter: shadow AI is a reality. And it is a double problem.
On the risk side first: leaks of personal data (hello, GDPR) and exposure of confidential customer information. On the value side next: the best prompts and the clever workflows stay in each person’s head. Nothing is shared, nothing is structured.
Rather than banning it, it is better to frame it.
“Just like when the internet was rolled out in the 90s, you have to give people access, set rules of good use and support them by teaching. Not by banning.” — Didier Girard, Managing Director at SFEIR & WEnvision
And above all, managers have to own their own practice: there is nothing shameful about using AI to write an email or a meeting report. On the contrary, it saves time. Managerial courage also means saying so.
4. Measuring productivity: aim for a x10 impact
Many AI projects fail because they aim too low. A 10% productivity gain is invisible in an organization. Aiming for x2, x3 or even x10 is not.
“If an 8-hour task is finished at 5:30 pm, the employee goes home. That’s nice, but the company doesn’t see it. If the same task takes half the time and they start a new one in the afternoon, then the impact becomes measurable.” — Didier Girard, Managing Director at SFEIR & WEnvision
Case study: the impact of AI on business processes
On the business side, the figures speak for themselves. Where AI is truly built into processes, productivity gains aren’t counted in percentages but in multiples. Cycles get shorter, there are fewer iterations, and teams refocus on what makes their added value. The example of a cosmetics company creating new formulas is telling:
“We took a process that used to take three months down to three weeks, cutting the number of back-and-forths from fifteen to seven. But AI didn’t replace the scientists: they delegated the monitoring, the curation and the literature review to it. They keep the core of the job and the value they bring.” — Marie Fontaine, Head of AI Strategy at WEnvision
It is also a matter of budget. These gains don’t fall from the sky: they have a cost, which has to be planned from the start. Too many companies launch AI projects without measuring their actual consumption, and discover the bill afterwards.
“Tokens are the fuel of AI. You have to budget for that use, otherwise it’s like buying taxis without planning for the gas.” — Didier Girard, Managing Director at SFEIR & WEnvision
5. Governance and risks: anticipate the excesses of AI at work
AI can also go too far. A heavier mental load, more context switching (constantly jumping from one task or tool to another, which fragments attention and costs several minutes of refocusing at every switch), and a risk of burnout for people who are already overstretched.
“AI never gets tired. It never takes a break. The more you feed it, the more it answers.” — Didier Girard, Managing Director at SFEIR & WEnvision
The AI Act and HR compliance: a regulatory framework to build in
On top of this comes a regulatory framework that is getting stricter, and that has to be built in as soon as HR use cases are designed:
“Under the AI Act, several uses, including résumé screening, are classified as high-risk. That doesn’t ban them, but it requires safeguards, traceability and auditability. It’s also a safety net for you.” — Marie Fontaine, Head of AI Strategy at WEnvision
To go further: The AI Act for companies: what the European regulation actually changes
Social dialogue: involve employee representatives in the rollout
One last point to watch: employee representatives. It is best to involve them from the very start of the project, and even to make them full users of AI themselves. Even if it takes a little more time at the beginning, it is the best way to avoid a lasting deadlock later, once the rollout is under way at scale.
Conclusion: HR, the engine of the AI transformation
AI won’t spread through the teams by magic:
“Don’t think that AI will arrive in the teams by magic. There is reluctance, and that’s normal. It comes either from the technology itself, or from human fears about the future of one’s job.” — Marie Fontaine, Head of AI Strategy at WEnvision
“Too often, AI literacy is left to IT. Yet success depends first and foremost on people. It’s up to HR to make itself heard, and to lead by example in using AI.” — Didier Girard, Managing Director at SFEIR & WEnvision
So the question is no longer whether you will roll out AI in your organization, but who will lead the transformation.
FAQ: your questions about HR’s role in rolling out AI
What is HR’s role in rolling out AI?
HR becomes the architect of work: redefining jobs and skills, supporting AI literacy, preventing new psychosocial risks, setting a legal framework for uses and bringing the whole organization on board, employee representatives included. Rolling out AI isn’t a purely technology project: according to Didier Girard (SFEIR & WEnvision), it is 10% tech, 30% quality data and 60% culture, organization and people.
Why do 95% of AI projects fail?
According to a recent MIT study, 95% of AI projects in companies fail, and the 5% of organizations that succeed don’t owe it to the technology. Projects fail in particular when AI is left entirely to the IT teams, when they aim for gains too small to be visible (10% more productivity goes unnoticed, a x2 or x3 factor doesn’t) and when their actual consumption hasn’t been budgeted.
Should shadow AI be banned at work?
No: it is better to frame it than to ban it. Shadow AI, the use of AI outside any framework, often in free versions, exposes the company to leaks of personal data and customer information, and leaves the best prompts in each person’s head. Just like the internet in the 90s, you have to give people access, set rules of good use and support them by teaching, and managers have to own their own use of AI.
What is a Skill-Based Organization?
In a Skill-Based Organization, work is no longer defined by fixed positions, but by the skills needed to carry out projects. With AI, what is expected of a person today won’t be the same tomorrow: HR has to update job descriptions continuously, spread a culture of using AI and invest heavily in training.
