The company had purchased AI licences for the entire organisation seven months earlier. The CEO told me this at our first meeting with a mixture of pride and bewilderment — pride in the decision, bewilderment that nothing had really happened since. Licences were activated. People had access. And yet the workflows looked exactly as they had in 2022.
The first thing we did was not build a course or write a strategy. The first thing we did was listen. We conducted one-on-one conversations with 23 employees across the four departments — sales, marketing, customer service and product. The questions were simple: Are you using AI in your work? What for? When do you stop using it? What do you wish it could do?
The answers were surprisingly consistent — and surprisingly revealing. 17 of the 23 employees were already using AI regularly. None of them knew their colleagues were doing the same. One sales consultant had built a complete set of prompts for writing proposals that she had never shared with anyone. A product manager was using AI to analyse customer feedback — but in his personal ChatGPT account, not the company's licensed system, because he didn't know it existed.
This phenomenon — shadow AI — is more widespread than most leaders imagine. And it creates two problems simultaneously. On one hand: risk. Data pasted into personal AI accounts is data the company doesn't control. Customer information, confidential strategy documents, pricing data — all of this can theoretically end up in an external model trained on the input. On the other hand: wasted potential. All these individual experiments, all these prompts, all these gains — they never multiplied, because no one shared them.
The problem wasn't the technology. It wasn't the lack of licences. It was the lack of shared culture and structure. People didn't lack access — they lacked confidence to experiment visibly. They were afraid of looking incompetent if they asked an AI for help. They were uncertain about what was and wasn't permitted. And they lacked a forum where they could share and learn from each other.
We designed the solution in three layers. The first layer was governance: a simple, two-page AI policy that answered the questions people actually asked. What may we use AI for? What may we not? What data is okay to enter? The policy was deliberately concise — we wanted people to read it, not file it.
The second layer was knowledge sharing. We set up an internal AI guild — a monthly forum where teams shared prompts, use cases and failure stories. Not PowerPoints about strategy, but concrete examples: "I used this prompt to summarise client meetings and it saves me 40 minutes a week." It sounds simple. It works extraordinarily. People learn best from each other, not from consultants.
The third layer was department-specific playbooks. Not one generic guide for everyone, but four targeted documents — one per department — with concrete use cases, recommended prompts and clear examples of good output. The sales team got a playbook on proposal writing and meeting preparation. Customer service got one on categorising enquiries and drafting responses. Marketing got one on brief structures and content ideas. Product got one on analysing feedback and prioritisation.
Implementation took three months. And it wasn't frictionless. The biggest resistance didn't come from the employees — it came from a middle manager in the sales department who was worried AI would take people's jobs. That's a concern that deserves respect and a genuine answer, not dismissal. We spent time addressing it: AI is a productivity tool, not a replacement tool. What we're trying to free up is time for the things humans are best at — judgement, relationship, creativity. It took time to build that trust. But it happened.
After six weeks, the AI guild started generating its own energy. Employees who hadn't been invited asked if they could join. The sales consultant who had never shared her prompts gave a 20-minute demonstration to the entire sales team. The manager who had been sceptical sat in the front row taking notes.
Results didn't come as a sudden leap, but as a gradual and sustained change. After three months we measured time usage again — same method, same questions. On average 11 hours freed per employee per week. For a team of 80 people, that equates to over 880 working hours per week that can be used for something else.
But the real gain wasn't in the numbers. It was in the culture. AI had stopped being something people did secretly in their browser tabs. It had become something they talked about at the lunch table, something leadership celebrated, something that was part of the company's identity. That's the kind of transformation that doesn't disappear when the consultant goes home.
If I were to point to the most important lesson from this project, it's this: AI adoption is not a technology problem. It's not an access problem. It's a culture problem — and culture problems are solved with patience, structure and psychological safety, not more licences.