Buying a library of AI courses is not an upskilling plan. People finish a module, return to their inbox and work the way they did before. An upskilling plan connects specific roles to specific skills, then gives people time and real work to practice on.
This guide covers skills development: helping employees use AI to do their jobs better. It assumes your AI usage rules already exist and are trained separately. Policy training answers "what am I allowed to do?" This plan answers "how do I do it well?"
The Six Steps at a Glance
| Step | Key question | Output |
|---|---|---|
| Step 1: Start from outcomes | What work should improve? | Two to four target outcomes |
| Step 2: Map roles to skill levels | Who needs which depth? | Role-skill matrix |
| Step 3: Assess the starting point | Where are people now? | Baseline by role |
| Step 4: Design pathways | How will each role learn? | Pathway per role, with practice tasks |
| Step 5: Make room to practice | When and on what? | Protected time, sandbox, champions |
| Step 6: Measure applied results | Did work change? | Applied-work measures and reviews |
Step 1: Start From Business Outcomes, Not Tools
Pick two to four outcomes where AI could plausibly help and where improvement would matter to leaders. Write them as work changes, not tool adoption.
• Weak: "Everyone uses the AI assistant weekly"
• Better: "Support agents draft first replies faster without lowering quality scores"
• Better: "Analysts produce first-draft monthly reports with fewer manual steps"
Outcomes decide which roles go first. They also give you something to measure in Step 6 other than course completions.
Step 2: Map Roles to Skill Levels
Not everyone needs the same depth. A simple four-level scale keeps planning manageable:
• Level 1, Aware: understands what AI tools can and cannot do, and the basic risks of wrong answers
• Level 2, User: uses approved tools for everyday tasks with good prompts and checks output
• Level 3, Builder: designs repeatable AI workflows for a team, such as templates, assistants or automations
• Level 4, Specialist: builds, integrates or evaluates AI systems as part of their job
Template: Role-Skill Matrix (Illustrative)
| Role | Target level | Priority skills |
|---|---|---|
| All staff | 1 | What AI does well, checking facts, when not to use it |
| Customer support agents | 2 | Drafting replies, summarizing tickets, tone control, fact-checking against the knowledge base |
| Financial analysts | 2 to 3 | Data cleanup prompts, formula help, reviewing AI-drafted commentary, building report templates |
| Operations team leads | 3 | Mapping workflows, building and maintaining team prompt libraries, measuring time saved |
| Software engineers | 3 to 4 | AI coding assistants, testing AI-written code, retrieval and evaluation basics |
| Data and ML engineers | 4 | Building, evaluating and monitoring LLM applications |
Fill this in with role owners. They know which tasks eat time and where errors would hurt.
Microsoft Learn's module on planning an AI skilling journey takes a similar approach: assess readiness, then design role-based journeys tied to day-to-day work. CIPD's AI skills planning guide also suggests matching the plan to your organization's AI adoption maturity.
Step 3: Assess the Starting Point
Skip long surveys. A short baseline per role is enough:
• Self-rating against the four levels, with one example task per level
• A practical task: for example, ask support agents to draft a reply to a sample ticket with AI, then score accuracy and edits needed
• Manager input on where people already use AI informally
Practical tasks matter because self-ratings drift. Some people overrate their skill; others underrate it because they use AI quietly.
Step 4: Design Pathways With Practice Built In
A pathway is a sequence of learning and practice that moves a role from its current level to its target. Each step should end in a real work task, not only a quiz.
Sample Pathway: Customer Support Agent, Level 1 to Level 2 (Illustrative)
Week 1, foundations: short session on how AI assistants produce answers and why they can be wrong. Practice: compare three AI drafts against the knowledge base and mark errors.
Week 2, drafting: prompts for first replies with tone, length and policy constraints. Practice: draft replies to ten past tickets and compare with the replies actually sent.
Week 3, summarizing: summarizing long ticket threads for escalation. Practice: summarize five escalations; a team lead checks for missing facts.
Week 4, live use with review: use AI drafts on real tickets, with a peer reviewing a sample. Practice: share one prompt that worked well in the team prompt library.
Proof of skill: a reviewer scores a sample of live replies for accuracy and tone. Passing moves the agent to Level 2.
Notice that every week ends with a task from the agent's real job. That is the difference between upskilling and content consumption.
Choosing Formats
| Format | Good for | Watch for |
|---|---|---|
| Self-paced courses | Level 1 foundations at scale | Low completion without protected time |
| Live cohorts | Levels 2 to 3, peer learning | Scheduling across time zones |
| Hands-on labs | Levels 3 to 4, building | Needs a safe sandbox and real data rules |
| Peer champions | Spreading team-specific tips | Champions need time and recognition |
| Prompt libraries | Reusing what works | Needs an owner to prune outdated entries |
Step 5: Make Room to Practice
Content alone rarely changes how people work. They also need time and permission to practice.
• Protect time. Agree a weekly learning block with managers, and protect it during busy periods
• Provide a sandbox. Give approved tools and safe test data so people can practice without risking real data
• Name champions. One or two people per team who share tips and collect questions
• Reward applied use. Recognize people who share a workflow that saved their team time
Make sure the usage rules are clear before practice starts. If people are unsure what data they can use, they will either not practice or practice unsafely.
Step 6: Measure Applied Results
Measure at three levels and report all three to leaders.
| Level | Example measures |
|---|---|
| Learning | Pathway progress, practical-task scores |
| Applied work | Share of tasks done with AI support, quality scores on AI-assisted work, prompt library contributions |
| Business outcome | The Step 1 outcomes, such as reply time with steady quality scores or report preparation time |
Set a baseline before training starts so you can compare. Expect learning measures to move first, applied work next and business outcomes last.
Our guide on measuring corporate AI training effectiveness goes deeper on metrics and reporting.
A 90-Day Plan for a First Wave
Days 1 to 15: agree outcomes with leaders, pick one or two pilot roles and complete the role-skill matrix for them.
Days 16 to 30: run baseline tasks, set up the sandbox and champions, and finalize pathways.
Days 31 to 75: run pathways with protected time. Review progress every two weeks and fix what is not working.
Days 76 to 90: score proof-of-skill tasks, compare applied-work measures with the baseline and decide which roles join the second wave.
Common Mistakes
• Starting with tools instead of outcomes. Tool adoption is easy to count but says little about value.
• One course for everyone. Roles need different depths and examples from their own work.
• No practice time. Learning without protected time rarely changes how people work.
• Mixing skills and policy training. Keep the rules short and separate so neither program gets diluted.
• Measuring completions only. Track applied work and the business outcomes you chose.
For a view of external training options, see our guide to AI training for employees.
Our Own Product Note
Disclosure: this is our own service. LuMay AI Learning Labs offer hands-on, mentor-led AI projects for teams. We are not ranked or recommended in this guide.
Next Step
Choose one outcome and one pilot role this month. Fill in its row of the role-skill matrix, run a baseline task and draft a four-week pathway.
References
• Microsoft Learn, Plan your organization's AI skilling journey with Microsoft 365 Copilot: https://learn.microsoft.com/en-us/training/modules/plan-ai-skilling-journey/
• CIPD, AI skills planning: Practical guidance for people professionals (Apr 14, 2026): https://www.cipd.org/en/knowledge/guides/ai-skills-planning/
