Over the past few years, AI has become the unavoidable topic for every organization. It is everywhere: in keynotes, in leadership meetings, in strategic plans. Budgets are following, and so are ambitions.
And yet, a massive paradox is taking shape.
- 91% of French companies consider AI a priority (Les Echos / Viva Tech, 2025).
- 88% say they already use it in at least one business function (McKinsey, 2025).
Organizations have never invested so much in AI, and they have never struggled so much to get concrete value out of it. Because behind the announcements and the promises, the reality is harsh:
- Only 1% of organizations consider themselves “mature” in their AI rollout (McKinsey, 2025).
- And 95% of generative AI projects in companies fail (MIT, 2025).
The question is no longer whether AI matters. Everyone knows it does. The problem is that almost nobody knows yet how to make it work at scale.
The good news? AI at work isn’t a tooling problem. It is a front-line problem: the AI that transforms a company is the AI anchored in the teams’ real work.
1. The state of AI adoption at work: why are so many licenses gathering dust?
AI is no longer a bet. It is already here, in the licenses, the policies and the leadership speeches. And yet, between the stated ambitions and the reality on the ground, the gap is striking.
Leaders expect strong adoption: 81% of them plan to bring AI agents into their organization moderately or extensively (Microsoft, Work Trend Index 2025). But on the ground, the licenses bought aren’t being used. Very often, the teams have the tool but don’t know what to do with it in their daily work.
This gap isn’t marginal, it is the heart of the problem. Most companies measure their AI like an IT project, with metrics disconnected from the real work. They count licenses, not uses. They celebrate pilots, not concrete gains. And when the budget comes up for renewal, nobody can say what AI has really brought.
→ Go further: Why do your Copilot licenses go unused? The full analysis of the causes and the levers.
2. Identify the barriers to AI adoption (and how to remove them)
When you dig into what is really blocking, technology is rarely the cause. The real barriers are organizational: the way licenses are rolled out, the way use cases are chosen (or ignored), the way skills are passed on, or aren’t.
A telling signal: 49% of employees use AI tools that their employer hasn’t approved (BlackFog study, 2026). The use already exists, but it happens off the radar. It is both a risk and an opportunity: the desire is there; what is missing is the framework.
→ Go further: What are the 4 real barriers to AI adoption, and how do you remove them?
3. Know who you are dealing with: the 5 profiles facing AI
Organizational barriers are only half of the story. The other half plays out in the individual attitude of each employee. In every organization, the same five profiles come up, each with its own levers and pace:
- The enthusiast: your future ambassador.
- The cautious explorer: the biggest adoption potential.
- The pragmatist: only moves on hard numbers.
- The skeptic: comes around when the results speak.
- The resister: don’t force them during the launch phase.
The rule is simple: trying to convince everyone at once is the best way to convince no one. The challenge is to recognize each profile, activate the ones who are ready first, and let time do its work for the others.
→ Go further: How do you manage each profile individually?
4. Remove the fear before rolling out: the key to AI literacy
Before talking about tools or method, there is a prerequisite that many AI programs forget: fear. 44% of French employees fear that their job will be replaced by AI or a robot (according to an Odoxa study).
This figure doesn’t tell a story of resistance to change. It tells a story of missing AI literacy. AI literacy is the process through which employees acquire the bearings and the confidence they need to understand what AI does, what it doesn’t do, and how it fits into their work. It isn’t technical training. It is building a shared culture, and that work happens before the rollout, not after.
→ Go further: How do you defuse the fear of AI and build a real culture of experimentation?
5. Train teams on real problems, not on tools
Training is the most actionable lever for going from licenses to use. The gap is clear: trained employees use AI twice as often as untrained ones (Ipsos BVA study for Google France, March 2026). But not all training is equal. The training that works starts from the teams’ business pain points, not from a tool’s catalog of features. Buy-in can’t be decreed; it is built around a real need that people have expressed.
→ Go further: How do you find the right AI training topics in your teams?
6. 5 good practices to turn AI into a performance lever
Employees often don’t know how to talk to AI in their daily work. Left to themselves, many try it once, get a generic answer and conclude too quickly: “AI isn’t much use in our line of work”.
The difference between gadget AI and AI that creates value doesn’t come from the tool: it comes from a way of talking to AI that can be learned. Nothing sophisticated, but nothing intuitive either. These reflexes are passed on, shared between peers and anchored in routines, and, repeated over time, they turn AI from an individual tool into a lever of collective performance.
→ Go further: 5 steps to turn AI into a real performance lever in your teams
7. Metrics and AI ROI: how to prove the business impact
Without measurement, AI remains a promise. And that is precisely what holds most programs back: 88% of organizations already use AI, but only 39% see a measurable impact on their operating income (McKinsey, November 2025). Companies count licenses rather than uses, and put forward projections instead of proof. When it is time to take stock, most leadership teams have nothing to show.
Steering AI doesn’t take a battery of complex KPIs. It takes a change of angle: stop counting licenses and start measuring real use, quarter after quarter. A few well-chosen metrics, tracked regularly, do more than an exhaustive dashboard abandoned after two months. It is the combination of these metrics, and the discipline of tracking them, that separates the companies that prove their ROI from the companies that can’t.
→ Go further: What are the 4 metrics that really prove the business impact of AI?
Conclusion: the AI that transforms starts from the front line
The 6 workstreams described in this article have one thing in common. They talk neither about tools, nor models, nor technology. They talk about method, framework, and the trust placed in employees.
AI moves forward where teams take the lead. Not before.
FAQ: your questions about AI at work
Why isn’t AI generating any ROI?
If AI isn’t generating ROI in your company, the problem is usually neither the tool nor the budget. It is organizational: 95% of generative AI projects fail to prove a measurable return, and only 39% of organizations see a tangible business impact. There are four causes: licenses rolled out without targeting specific jobs, no concrete use cases, shadow AI that nobody builds on, and employees left alone with the tool and no method. ROI appears when you stop counting licenses and start measuring real use.
How do I roll out AI in my company?
Rolling out AI in a company isn’t about buying licenses in bulk, but about a method in 6 workstreams: target 2 or 3 priority jobs rather than rolling out to everyone, identify use cases bottom-up (the pain points the teams live with every day), remove the fear through AI literacy before the rollout, train on real business problems rather than on tools, equip people with good prompting practices, and steer with real usage metrics, quarter after quarter. AI moves forward where teams take the lead, not before.
Where do I start with AI at work?
To start with AI at work, the first step isn’t choosing a tool, it is assessing the maturity of your organization. Find out where you stand on four axes: real adoption (how many employees use AI every week), use cases identified on the ground, your teams’ level of AI literacy, and whether you have metrics to steer by. Then target one priority job with repetitive pain points, activate the enthusiasts as ambassadors, and train on concrete cases. Teamstarter’s AI maturity quiz gives you a free diagnosis in 2 minutes.
How do I get my teams to use AI?
Getting your teams to use AI takes a different approach for each of the 5 employee profiles (enthusiasts, cautious explorers, pragmatists, skeptics, resisters). Activate the enthusiasts with an official ambassador role. Bring the cautious explorers on board with demos by their peers on concrete cases. Convince the pragmatists with hard evidence of time saved and ROI. Let the skeptics and the resisters watch the first successes. Trained employees use AI twice as often as untrained ones: training anchored in their real business cases remains lever number 1.
