Quick Answer: More training won’t fix it. The employees who aren’t using AI meaningfully already know it exists. The real problem is that most people don’t yet know how to describe the workflow of their own job well enough — in the specific, step-by-step way — to tell AI what to do.
What’s actually happening
A 2023 Harvard Business School and Boston Consulting Group study with 758 consultants found something telling: AI made consultants 40% more productive on tasks it was good at — but actively made them worse (by 19 percentage points) on tasks it wasn’t equipped to handle. The problem wasn’t the tool. It was that people couldn’t tell the difference, so they trusted the output when they shouldn’t have.
Your super-users figured this out through trial and error. They know which parts of their job AI can handle (drafting, summarizing, formatting) and which parts require their own judgment (client context, unwritten rules, institutional history that lives nowhere in any document). That knowledge takes time to build. A one-hour training session doesn’t build it.
The middle of your organization isn’t resistant to AI. They’re rational. They tried it, the output wasn’t good enough for their actual work, and they went back to doing it themselves. That’s not a mindset problem — it’s a fit problem nobody has solved for them yet.
What actually works
- Stop measuring adoption by who opened the tool. Measure whether the output quality improved on specific tasks. Those are different things.
- Pick one task per team and engineer it completely. Find the thing a team does ten times a week. Work out exactly what the prompt needs to say, what background context it requires, and what a good output looks like. Then document that and share it.
- Use your super-users as guides, not cheerleaders. Their value isn’t enthusiasm — it’s that they’ve already done the hard work of figuring out how AI fits their specific job. Have them show peers on their actual work, not demos.
- Give people protected time to experiment. Microsoft’s Copilot rollout across 300,000+ employees found that manager-approved time to experiment — not training content — was what actually moved people from curious to capable.
The one thing to do next
Pick the single task your team repeats most. Write down every piece of context that task requires — the stuff that’s obvious to a ten-year veteran but invisible to anyone else. That’s your prompt. Build from there.
Read the full analysis: The Detail Gap: Why Your AI Strategy Stalls in the Middle