Thinking of Upskilling Your Team in AI? Start Smaller Than You Think.

I stepped onto the scales the other day and sighed. My wife was nearby and said: "Why are you sighing? It's not like you've changed your diet or exercised, what did you expect?" Cold, but fair. That's kind of what learning about AI is like. Nothing changes if you don't actually do something different.

So you've been reading about AI developments and decided now's the time to upskill your team. The natural instinct is to organise company-wide training. Get everyone in a room, run through the possibilities, surely people will figure out how to apply it themselves.

The problem isn't that people don't understand AI conceptually. It's that they don't see how it fixes their actual Tuesday afternoon when they're manually copying data between spreadsheets for the fourth time this week.

Here's what I've found actually works: pick one department, automate one genuinely tedious workflow, then step back and let people talk about it. Smaller than you think, more specific than feels comfortable.

Start with the team doing the worst work

Not the most strategic team or the most visible one. Find whoever is doing the most soul-crushing repetitive task in your organisation. Finance reconciling expense reports. HR processing leave requests. Operations copying information between systems because two platforms refuse to talk to each other.

These become your early adopters not because they're particularly tech-savvy, but because they're genuinely desperate for help. Turns out desperation beats enthusiasm for actual implementation every time.

Build one thing that actually works

Skip the grand strategy sessions and roadmaps for now. Do a quick sense check of what everyone's level of AI is. Most likely you'll get some members of the team still "skeptical" of the technology whilst others may already be far along the automation journey.

Pick one specific task and automate it properly. Show the finance team how to build an n8n workflow that pulls expense data, validates it against policy, and populates the approval form. Or set up a Claude Projects workspace that drafts responses to the same twelve vendor questions they answer every week.

The important bit is making it simple enough that they can modify it themselves when requirements inevitably change. Template it, document it with screenshots, test it until it's genuinely easier than their current process. If it's not actually easier, they won't use it, and you've wasted everyone's time.

Why this approach matters

Most AI training fails because it treats technology adoption as purely an education problem. Send everyone to a seminar, surely they'll figure out applications themselves. But people are already busy. They default to existing processes because changing behaviour requires more effort than sitting through a training session provides.

A working automated workflow that saves real time is different. It's immediately useful, which makes it immediately adopted, which makes other people curious. You're not asking them to learn something for future benefit. You're giving them something that makes today easier than yesterday.

The scales in my bathroom haven't budged because I keep expecting different results from the same behaviour. Organisations do exactly the same thing with technology training, running identical company-wide sessions and wondering why adoption rates don't improve.

Maybe it's time to try something smaller. Fix one genuinely annoying thing for one team. Watch what happens when people realise this might actually make their jobs better rather than just different.