AI upskilling conversations tend to start with courses. They should start with the work. Here's a four-layer, seven-step plan for building AI skills that change how teams operate.
TL;DR
Conversations about AI upskilling tend to start with courses. Which platform, which certificates, how many hours per person. Those are reasonable questions, but they come too early.
The organisations that get value from upskilling start with the work: which tasks will change, who does them, and what those people need to be able to do. The training follows from that. Here's how I'd structure it.
IBM defines AI upskilling as building employees' knowledge and capabilities so they can work effectively alongside AI in their day-to-day roles. It separates this from reskilling, which means learning a whole new set of skills for a different job. A customer service team learning to use an AI assistant to answer queries is upskilling. A data processing clerk retraining as an analyst is reskilling.
Most organisations need far more of the first than the second. The day-to-day jobs mostly stay, while the way people do them shifts.
It's tempting to buy a learning platform licence, assign a few modules, and report the completion rate. McKinsey's take on AI upskilling describes exactly that pattern: companies spend disproportionately on literacy because it's visible and easy to measure, and fewer take on adoption, which is messier.
The same piece describes how McKinsey drove adoption of its own internal AI platform, Lilli, by treating upskilling as a change journey: leaders telling the change story, new workflows, learning content, and hands-on support. That approach took the platform to 30,000 users. The lesson I take from it is that the training was one part of a broader effort to change how people worked.
There's a wider reason to take this seriously. The World Economic Forum's Future of Jobs Report 2025 expects 39% of workers' existing skill sets to be transformed or become outdated between 2025 and 2030. That's down from 44% in the 2023 edition, and the Forum's summary of the Future of Jobs Report 2025 links the improvement partly to companies investing more in continuous learning and upskilling. Steady, deliberate investment appears to help.
A literacy baseline for everyone. How the tools work, where they fail, what the rules are, and how to check output. This is the foundation, and I've written about it in more depth in our guide to AI literacy.
Role-specific skills. What a finance analyst needs is different from what a recruiter needs. This layer covers the specific tasks AI changes in each role, practised on that role's real work.
Workflow redesign inside teams. IBM calls this workflow integration: identifying where in a process AI adds value and adjusting the process so AI-assisted work fits with human-led work. This layer happens in teams, with managers, and it's where the measurable gains sit.
A separate leadership track. Leaders need to understand the tools well enough to make decisions about strategy, risk, vendors, and how saved time gets used. They also need to model the behaviour, because teams watch what leaders do more closely than what they announce.
1. Map real use cases first. IBM's guidance is to identify specific workflows and tasks where AI could make a meaningful difference before launching any programme. I'd do this team by team, with the people who do the work. A list of 10 to 20 concrete tasks gives your training something to aim at.
2. Find your starting point. Before you design anything, understand where people are. A short survey on current use, confidence, and the tools people already use (including unapproved ones) saves you from designing for the wrong audience.
3. Tier the training by role and risk. Everyone gets the baseline. People whose AI use carries more risk, such as those handling personal data, client work, or hiring decisions, get deeper training on their specific tasks and the checks they need to apply.
4. Protect time to learn. IBM is blunt that upskilling initiatives fail when employees don't have time to learn. If people are expected to fit it around a full workload, it won't happen. Block the time in calendars, and have managers defend it.
5. Build a network of champions. Peer learning often travels further than top-down training. Pick people in each team who use AI well and enjoy helping others, and give them time and recognition for it. A practical, hands-on AI workshop is a good way to launch this network.
6. Update what you measure. If performance is still judged only on volume or speed, and AI changes how that work gets done, the old measures stop meaning much. IBM recommends rethinking KPIs to reflect the value people create with AI support, and I'd agree.
7. Set a refresh rhythm. The tools change every few months. A quarterly update for each team, focused on what's new and what's working, beats an annual course that's out of date by the time people finish it.
There's no fixed number, but there are useful signals. BCG's AI at Work research found that regular use rises among employees who receive at least five hours of training and have access to in-person coaching. A single hour won't get most people to the point of changing how they work.
I'd plan for a few hours of baseline and role-specific training per person in the first quarter, followed by short, regular sessions. The pattern matters more than the total.
Course completions and quiz scores tell you about activity. To understand impact, look at three things over a quarter.
First, use: how many people use AI on the tasks you mapped, and how often. Second, workflow change: which processes now run differently, and what that's done to time, quality, or rework. Third, confidence and capability, which a short repeat survey can capture.
The second measure is the one the business feels, and it's also the hardest to fake. If you can't point to specific workflows that changed, the programme hasn't done its job yet, whatever the completion rate says.
People are understandably nervous about what AI means for their jobs. IBM recommends communicating clearly about your approach and how AI helps people do their jobs, and I'd go further: say plainly what the programme is for.
Our position at Praxes is that AI-native work means a more capable team, with repetitive and administrative work moving off people's plates. We don't run engagements aimed at reducing headcount. Whatever your organisation's position is, people can tell which kind of programme they're in, and that shapes whether they engage with it. Our guide to AI adoption covers this in more detail.
AI upskilling is building employees' skills so they can work effectively alongside AI in their current roles. It includes AI literacy, practical skills for specific tasks, and changing workflows so AI-assisted work fits into how teams operate.
Upskilling improves someone's skills for the job they already do. Reskilling prepares them for a different job. Most AI programmes are mainly about upskilling.
For an organisation, start with the work: map tasks where AI helps, then train people by role on those tasks, with protected time, internal champions, and follow-up. For an individual, practise on your own daily tasks alongside a short structured course.
A literacy baseline and first role-specific skills can be in place within a quarter. Keeping skills current is ongoing, because the tools keep changing.
It works best with shared ownership: a senior sponsor, HR or learning and development running the programme, and line managers responsible for workflow change in their teams.
If you want to see where your organisation stands before designing a programme, our free AI Readiness Diagnostic gives you a scored picture across strategy, data, people, process, and tooling.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic