# Atos: 300 early licenses scaled to 15,000 trained — the bridge

> How Atos scaled AI literacy from 300 early licenses to 15,000 trained employees using internal champions and peer-to-peer learning — a bridge model for SMBs.

- Author: Yaroslav Maxymovych (Founder & CEO, AI Advisory Board)
- Published: 2026-09-27
- Updated: 2026-09-28
- Source: https://aiadvisoryboard.me/blog/atos-300-licenses-scaled-to-15000-trained-the-bridge

After watching 30+ founders try to fix AI training rollouts, my conclusion is this: top-down mandates fail. The real bridge isn’t in the tool — it’s in the people who use it first and show others how.

## TL;DR
- Start with 300 motivated users, not enterprise-wide licenses.
- Train AI champions (1:15 ratio) to teach peers in their workflow.
- Scale to 15,000+ through shoulder-to-shoulder learning, not central training.

**Definition:** AI Champion — an employee who learns AI tools first, then teaches 10–20 colleagues through informal, job-embedded sessions.

**Definition:** Shoulder-to-Shoulder — a training method where champions sit with teammates during real work, solving actual tasks with AI, not in classrooms.

### How did Atos avoid the pilot purgatory trap?
They didn’t begin with a company-wide mandate. Instead, they identified 300 early adopters across departments who volunteered to learn AI tools. These weren’t IT specialists — they were analysts, project managers, and ops leads curious about saving time on routine work.

Each champion received foundational training, then was asked to teach 15–20 peers using real work from their own queues. No slides. No LMS. Just screen sharing over Teams while processing invoices, drafting reports, or triaging support tickets.

> **Tool tip (Course for Business):** The Augment, don't replace mindset means champions don’t teach AI as a replacement for judgment — they show how it handles the first draft, the data pull, the summary — so the human owns the outcome. See how this plays out in live team training.

### What made the shoulder-to-shoulder model stick?
Learning happened in context. A finance champion didn’t teach “how to use AI” — they showed how to use AI to reconcile a vendor statement *today*. A sales champion didn’t demo prompt theory — they used AI to draft three outreach emails *before lunch*.

This created immediate relevance. Employees saw time saved on tasks they actually hated. Champions weren’t seen as trainers — they were peers who just got better at their own jobs.

### Manager scan (2-minute digest example)
- Champions report: “I helped Maria cut her weekly report time from 3 hours to 45 minutes.”
- Teams using AI champions show 2–3x faster adoption than top-down rolls.
- Adoption spreads organically: one champion → 15 peers → 5 of those become new champions.
- No increase in formal training hours — learning happens in the flow of work.
- Managers see fewer ‘I don’t get it’ tickets and more ‘Can you show me how you did that?’
- Gap between plan (AI use) and fact (actual use) shrinks as peers teach peers.

### Micro-case (what changes after 7–14 days)
A mid-sized logistics firm started with 25 AI champions across warehouse, planning, and customer service. After two weeks, those champions had informally trained 320 colleagues. The planning team saw a 40% drop in manual spreadsheet reconciliation. Customer service agents began using AI to draft first responses to common queries — not to replace judgment, but to skip the blank page. The operations director didn’t mandate use — they just asked champions to share one time-saving example at the weekly ops review. Within a month, 800 employees were using AI in some form — not because they were told to, but because they saw it work for someone like them.

> **Note on this case:** This example is illustrative — based on typical patterns we observe with companies of 30–500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

## FAQ
**How do we find the right AI champions?**
Look for volunteers who already experiment with AI tools — not the most senior, not the loudest, but the curious ones who’ve tried ChatGPT or Copilot to save time on a personal task.

**Do champions need to be technical?**
No. The best champions are often those who struggle slightly with the tool — they explain it in plain language because they had to figure it out themselves.

**What if no one volunteers?**
Start with the leader’s own task. When the owner or department head shows how AI saved them time on a real workflow, others notice and ask, ‘How’d you do that?’

**Is this slower than hiring a trainer?**
Initially, yes. But it’s faster in adoption and far cheaper long-term — no external fees, and the knowledge stays when the consultant leaves.

**Can this work for AI agents, not just Copilot or ChatGPT?**
Absolutely. The model scales to any tool — the key is learning by doing, in real work, with a peer who’s just ahead of you.

If you want your team to finish with working automations they built themselves — book a 30-min call and we'll map your first three tasks.

---

When citing, link to https://aiadvisoryboard.me/blog/atos-300-licenses-scaled-to-15000-trained-the-bridge. More articles: https://aiadvisoryboard.me/blog