What I learned training 100+ employees on AI in 6 months

The context: why train 100 people on AI

Between September 2025 and February 2026, these engagements covered SMBs and mid-sized companies with 30 to 250 employees, across a range of sectors: food manufacturing, B2B services, specialty retail and software publishing.

What these companies had in common: their marketing, sales and communications teams had started using AI with no coordination at all. Personal ChatGPT accounts here, a shared Midjourney account there, prompts copied from LinkedIn with no connection to the business. Leadership saw the enthusiasm, but also the risk of scattered efforts and the lack of concrete results.

The goal was to build a consistent AI culture within each organization, with uses aligned with business objectives and sound data practices. Learning the tools was only the first step.

The teaching approach: cohorts working on real cases with long-term follow-up

I structured the program in cohorts of 12 to 15 people, grouped by role. Each cohort followed a 6-week program that included:

• A 3-hour introductory session on the fundamentals of generative AI (how LLMs work, their limits, ethical issues)
• 2 hands-on workshops of 2 hours each, focused on use cases specific to the group's role
• A 1-week challenge in which each participant had to apply AI to a real task from their daily work
• A group debrief session to share results and identify best practices

After the first 6 weeks, I kept up monthly support in the form of “AI clinics”: 1-hour sessions where participants brought their real cases to get help with their prompts, workflows or tool choices.

This gradual format was essential. A one-off 2-hour workshop generates enthusiasm. A structured program spread over time generates change.

The resistance phase and how I got through it

Resistance was stronger than I expected. Of the 100 participants, this is the breakdown I observed in the first session:

• 25% were enthusiastic and had already started experimenting
• 45% were curious but skeptical about the concrete value for their job
• 30% were openly reluctant, voicing fears about being replaced or about the quality of AI output

The most common fears had to do with identity. “If AI writes my emails, what justifies my job?” “I'm proud of the quality of my writing, I'm not going to hand it over to a machine.” These reactions are legitimate. Ignoring them guarantees the program will fail.

It clicked the day I understood that the work was still mine. AI was taking away the parts I didn't like doing, so I could spend more time on the ones that motivate me.

Communications manager, mid-sized food manufacturer, cohort 3 participant

My response to resistance relied on 3 levers. First: always present AI as an amplifier of people's work. Second: work exclusively on real cases from the participants' daily work. Third: give it time. I learned that it takes an average of 3 weeks of regular use for a skeptical employee to come around.

The turning points

Each cohort had its “turning point”: the session where the group as a whole moves from curiosity to conviction.

For a sales team, it came when a participant showed how they had cut the preparation of a personalized prospecting brief from 40 minutes to 8 minutes, using a structured prompt fed with CRM data. The quality was higher than what they produced by hand, because the prompt systematically included the information human intuition sometimes forgets.

For a marketing team, the turning point came from a workshop on rewriting product pages. In a 90-minute session, the group produced 35 SEO-optimized product pages, work that would have taken 3 weeks the usual way. The marketing manager immediately saw the impact on the next quarter's editorial plan.

For a communications team, building a complete industry monitoring workflow changed everything. What took 4 hours every Monday morning now took 30 minutes, with broader coverage and a more structured summary.

Concrete results after 6 months

The consolidated numbers across all 4 companies, 6 months after the first program launched:

Of these numbers, 2 deserve a closer look. The first is the jump from 12% to 74% adoption. The tools stayed the same: the practices changed. The second is the number of use cases, from 3 informal uses to 47 documented and shared cases. When teams understand the potential, they find the most relevant applications for their own work.

The estimated return on investment across the 4 engagements combined: the productivity gains are equivalent to 2.8 FTEs (full-time equivalents) per month, for a total training investment lower than the cost of a single hire.

My takeaways

AI training is a process that runs over months. One-off workshops create enthusiasm, and lasting change comes from continuity: a structured program, checkpoints, support over time. The monthly “AI clinics” proved as important as the initial sessions.

Start with the skeptics. It sounds counterintuitive, but it held up in the field. When a reluctant employee becomes convinced, they bring 5 colleagues along. When an enthusiast stays enthusiastic, they win over nobody new. Spending time on the most resistant employees is the best way to speed up adoption.

Use cases have to come from the people doing the work. The best uses of AI are the ones employees discover for themselves while working with the tool every day, and they beat anything leadership or the consultant comes up with. My role is to provide the framework, the skills and the confidence. Innovation in how the tools get used comes from the teams.

Data governance has to be addressed from the start. I built a dedicated session on data best practices into the first week of every cohort. What data can you send to an AI tool? What information is sensitive? How do you anonymize before prompting? Paradoxically, this reassuring framework sped up adoption: employees felt free to experiment within clear boundaries. In practice, I recommend setting up a simple policy with 3 categories (open data, data to anonymize, prohibited data) before the first workshop. For a hands-on look at anonymization, see my guide to GDPR and anonymization for AI.

Middle management is the key. If frontline managers aren't trained first and don't lead by example, adoption stalls. In the companies where I trained managers 1 week before their teams, the adoption rate at 3 months was 35 percentage points higher. It's the most decisive factor I've observed.

Training 100 people in 6 months taught me that AI transformation is above all a human transformation. Technology is the means. Method is the framework. What makes the difference is the confidence teams build in their ability to work differently. And that confidence is built from each person's very first step: my guide to learning Claude walks through that path, from first setup to full independence.