AI literacy gives everyone in a company a shared baseline of understanding about artificial intelligence. That baseline covers three things, what AI can do, what it cannot do, and what is allowed with company data. A literacy programme therefore does not start with a tool, nor with technical training. It starts with a sponsor on the executive committee, rules of use written down in plain language, and two or three operational pain points that teams want gone.
Tandem supports more than 40 SMEs and mid-caps across more than 60 AI projects, from Doctolib to InfraVia and on to PayFit, Nomination and KparK. The same pattern shows up on every engagement. The organisations that stall are not short of tools, they are short of a baseline. This guide sets out the path Tandem applies, the formats that genuinely work, the metrics to track, and the mistakes that make a programme collapse after six weeks.
What AI literacy actually means in a company
AI literacy is a collective, cross-functional effort aimed at shared understanding rather than technical mastery. It removes two symmetrical blockers, the fear of being replaced and the naive belief that AI will do everything for you. It answers the why and sets the boundaries. Training comes next and answers the how, job by job.
| Stage | Goal | Audience | Cadence |
|---|---|---|---|
| AI literacy | Understand AI, its limits, the rules of use | The whole organisation | Short sequences, then rituals |
| AI training | Acquire concrete practice on your own tasks | By job family or team | Repeated hands-on sessions |
| Deployment | Ship copilots and automations to production | Pilot teams, then wider rollout | Projects of a few weeks |
Confusing the three is the most common mistake. A company that buys 300 AI assistant licences without literacy work gets a usage curve that rises for two weeks and then collapses. The opposite creates frustration. A company that runs literacy work and never deploys leaves its teams convinced of the value, with tools that have not moved an inch.
Why AI literacy has become urgent
France's lag is measured. According to Insee (Insee Première no. 2120, published on 21 July 2026), 18% of French companies with 10 or more employees used at least one AI technology in 2025, against 10% in 2024 and 6% in 2023. The rate has tripled in two years, and it hides a wide maturity gap. Companies of 10 to 49 employees stall at 15%, those of 50 to 249 reach 31%, and those above 250 climb to 58%.
Among smaller structures, the France Num barometer presented by the French Directorate General for Enterprise (6th edition, published on 15 September 2025, 11,021 companies surveyed) shows that 26% of micro-companies and SMEs use at least one AI solution, double the 2024 figure. The sector gap is striking, with 41% in IT and telecoms against 9% in agriculture.
The second signal is harsher. Adoption does not mechanically produce results. In its study « The state of AI » published in November 2025, McKinsey found that more than 80% of the organisations surveyed see no measurable effect of AI on their operating profit. The factor most correlated with success is not the choice of tool, it is redesigning how work flows. A culture and organisation problem, then, not a licensing one.
The third signal is often invisible to leadership. Employees have taught themselves. The Anthropic Economic Index, in its May 2026 snapshot, ranks France 10th out of 121 countries for Claude usage intensity relative to its working-age population, with an index of 3.97, close to four times what population alone would predict. In that same snapshot, 48% of French conversations are personal use against 34% work use, and 14% concern learning. These figures describe observed usage, not jobs or occupations. They tell leaders one thing above all. Your teams already use AI, just outside the company framework.
Where to start, the first four weeks
An AI literacy programme does not start with a training catalogue. It starts with a short framing phase. Here is the sequence Tandem applies most often in SMEs and mid-caps.
The order matters. A literacy effort that starts with operational teams without going through leadership hits the first scheduling trade-off, because nobody frees up half a day for a subject their management has not prioritised.
The standard path in three stages
On the programmes Tandem runs, the path is structured in three stages and six topics, understand, activate, sustain. It is the backbone of our AI literacy offering, and it transposes as is to a programme run in-house.
That third stage is the one companies cut most often, and it is the one that decides the outcome. AI platforms ship major features every three to six months, so a programme designed as a one-off event is obsolete before the year is out.
A session inspires. A ritual installs.
Which formats actually work?
Four formats come up consistently, and a fifth costs almost nothing. The common rule fits in one sentence. Participants work on their real files, never on textbook cases.
For self-directed learning, point teams to a structured path rather than a list of links. Our guide to putting AI to work right now is often the starting point in the programmes Tandem runs.
The five mistakes that make a programme collapse
These five traps come back engagement after engagement. We detailed them, with figures, in our analysis of companies that see no return on their AI investments.
How to measure whether AI literacy worked
Successful AI literacy shows up in metrics, not in an end-of-session feeling. Five measures are enough to steer a programme.
- The share of weekly active users on the deployed assistant, department by department.
- The number of use cases documented and reused by other teams.
- Reported time saved on the targeted tasks, measured before and after.
- The share of employees able to state the rule on sensitive data.
- The number of active internal champions, meaning those who actually run sessions or unblock colleagues.
At InfraVia, a European infrastructure fund managing 16 billion euros, bringing the teams on board came before deploying around ten Microsoft Copilot agents in their M365 environment. The details are in the InfraVia case study.
How long does it take and what does it cost?
Duration depends on size and ambition. Expect a few weeks for a micro-company whose owner drives the subject, two to three months for an SME that wants a diagnosis and workshops by job family, six months or more for a multi-site mid-cap. Cadence matters more than volume, and four short sequences spread over a quarter embed usage better than one dense day forgotten within a fortnight.
On budget, literacy is the lightest line in the chain. The recurring cost that really weighs is AI assistant licences, in the range of 30 to 40 euros per user per month on the professional platforms we observed during our engagements in April 2026. At that price, one hour saved per user per month already pays for the licence. So the subject hinges on real usage rather than on price negotiation.
If the scope is still unclear, the next step is not a training catalogue but a framing exercise. Our AI audit method for companies explains how to map processes and rank use cases by return on investment before committing to anything.
Start with the baseline, not the tool
AI literacy is not a box to tick before the real projects, it is what makes the projects possible. Three decisions are enough to start this week. Appoint a sponsor on the executive committee. Write one page of rules of use. Ask three teams what makes them lose time every week.
Everything else follows from those three decisions. And if your organisation already has dormant licences and teams using AI outside the framework, the subject is no longer adoption, it is regaining control. Our AI framing and audit offerings start exactly from that situation.
Tools give you the capability. Culture creates the usage.
