Most SMBs we work with have done some form of Copilot training, and most are quietly disappointed with what it produced. Usually it was a 90-minute webinar from the reseller, maybe a follow-up lunch-and-learn, and then radio silence. Four weeks later, two people are getting real value out of the tool and the other twenty-eight have quietly drifted back to old habits.
The problem isn't the content. It's the shape. Copilot is a skill, not a feature. You don't learn skills from a webinar; you learn them from doing the thing, being observed, getting better. Here's how to design training that actually works in a small business, without hiring a learning and development team you don't have.
Drop the one-off training day
The standard model - a single big training event, followed by a deck and some recorded videos - has roughly the same effect on adoption as a gym membership has on fitness. A small spike, no lasting change. People go back to their inboxes on Monday and never apply what they learned.
What works instead is short, frequent, and contextual. Twenty minutes, every week, for six to eight weeks. Always tied to something the person is actually trying to do that week. Always with the chance to ask questions. That's it.
Train by role, not by feature
Microsoft's training materials are organised by feature: 'Copilot in Word', 'Copilot in Excel', 'Copilot in Teams'. That's useful for the people who built it. It's not how anyone in your business thinks about their week. Your sales lead doesn't wake up wondering how to use Copilot in PowerPoint. They wake up wondering how to get the Q2 pipeline review deck done before Thursday.
Train by role. For each team, identify the three or four most painful recurring tasks and build the training around those. Sales: proposal drafting, meeting follow-ups, account research. Finance: variance commentary, board pack drafting, policy lookups. HR: job description drafting, interview question prep, policy Q&A. Operations: supplier comparison, SOP drafting, incident summaries.
When someone learns 'how to write a board commentary in 20 minutes instead of 90', they remember it and they use it the next day. When they learn 'Copilot in Excel features overview', they forget it by Thursday.
Use working sessions, not training sessions
The single most effective format we've found is the 'bring your own work' session. Thirty minutes, six people from the same team, each brings one real task they need to do that week. The facilitator (an internal champion, not an external trainer) works through one person's task with everyone watching, then the others try theirs while the facilitator floats. By the end, six real pieces of work have been done with Copilot and six people have a tangible memory of what good use looks like.
This format does several things at once. People learn from each other's tasks, not just their own. The use cases are immediately relevant. And it solves the 'I don't know what to use it for' problem that kills more rollouts than any other.
Build a small internal library
Within four weeks of starting, you'll have a stash of prompts that genuinely work for your business. Save them. A simple SharePoint page or pinned Teams channel is enough. Categorise by team and task. When someone new joins or someone new gets a licence, the library is their first stop. This compounds: every month it gets better and the time-to-productivity for new users gets shorter.
Don't outsource this. Internal prompts written for your business are dramatically more useful than the generic 'top 50 Copilot prompts' content floating around online.
Make a champion in every team
A central training team in an SMB doesn't scale. Instead, identify one person per team who's naturally curious, technically comfortable, and respected by their colleagues. That's your champion. Give them a small amount of protected time (an hour a week), early access to anything new, and explicit licence to teach their team. Champions are the single biggest predictor of whether Copilot sticks in a team. Pick them carefully.
Measure training, not just usage
Track two things alongside the usage data. First, did people attend the working sessions - if attendance falls below 60%, your sessions aren't useful enough, change the format. Second, can people in the team articulate, unprompted, two or three specific things they use Copilot for? If the answer is no by week six, the training isn't landing and you need to adjust before usage tails off.
What about the people who won't engage
There will be some. Usually 10-20% of any team. Two responses both work, depending on the role. For people whose job is mostly Copilot-helpable (heavy office work, lots of writing and analysis), a direct conversation about expectations is fair: this is how we work now. For people whose role genuinely doesn't benefit much, redirect the licence elsewhere. Don't try to drag people who don't want to be dragged - it poisons the wider rollout.
The honest summary
Copilot training that works is not a training day - it's a working pattern. Short weekly sessions, organised by role and real tasks, run by an internal champion, with a growing library of prompts that actually fit your business. None of that requires an L&D team or an external partner. All of it requires the discipline to keep showing up for six to eight weeks while the habits embed. Do that, and you'll go from 30% adoption to 80% in a quarter. Skip it, and you'll be writing off the licences by Christmas.