An AI upskilling program in 90 days: a week-by-week plan

A 90-day AI upskilling program has four phases: a baseline, training, supported practice and a reassessment. The baseline and the reassessment use the same measures. Without that, nobody can say whether the program changed anything. This plan covers 13 weeks for one to three teams, which is a sensible size for a first run.

Why a short course is not enough

Most employees have not had enough training. Boston Consulting Group surveyed more than 10,600 workers in 11 countries in 2025. Only 36% felt adequately trained in AI use (BCG). Only 51% of frontline employees were regular AI users, a figure BCG described as stagnant.

BCG also found that people who received five or more hours of training were significantly more likely to become regular users. In-person training with coaching made a larger difference (BCG). A single webinar does not reach five hours. This plan budgets about eight hours of training per person, plus coached practice on real work.

The quality of use matters as much as the amount. A 2025 study by KPMG and the University of Melbourne found that 66% of employees rely on AI output without checking its accuracy (KPMG). A program that raises use without teaching people to check output raises risk too.

Weeks 1 and 2: scope and baseline

1. Pick the teams. Choose one to three teams with a manager who will protect time for the program.
2. List the tasks. For each role, list five to ten recurring tasks where AI could help, such as drafting, summarizing or analysis. The AI workflow map template gives a format.
3. Set the rules. Publish which tools are approved and what data may go into them. Use the team AI use policy template if you have no policy yet.
4. Measure skills. Give every participant the same skills assessment and record the results by person and by skill.
5. Measure the work. Pick three tasks per role and record current time per task, rework rate and errors found in review. Use a sample of real work, not estimates.
6. Agree the decision rule. Write down, with the sponsor, what result in week 12 would justify extending the program.

Weeks 3 to 6: training on real tasks

Run four weekly sessions of about two hours each. Teach each skill on the tasks listed in week 1, not on generic examples. Use the baseline results to spend more time on the weakest skills.

• Week 3, asking and context. How to write a clear request, and what background, data and limits the tool needs. See asking well and giving context.
• Week 4, checking the work. How to verify figures, quotes and citations, and when not to trust an answer. See checking the work.
• Week 5, keeping it safe. What data stays out of which tools, and how to disclose AI use. See keeping it safe.
• Week 6, putting it together. How to turn a one-off prompt into a repeatable workflow with a check at the end. See putting it together.

End each session with homework: apply the skill to one real task that week, and bring the result to the next session.

Weeks 7 to 10: supported practice

Practice on real work is the longest phase. Each participant picks two of their listed tasks and does them with AI every time they come up for four weeks.

• A weekly 30-minute clinic. Participants bring one result that worked and one that did not. A champion or trainer leads it.
• Manager review. Managers review a sample of AI-assisted work each week, using the output verification checklist.
• A shared library. Working prompts and checks go into one shared document, with the task they belong to.
• A help channel. Questions get an answer within a working day.

Champions and managers carry this phase. See how to run an AI champions program and AI training for managers.

Weeks 11 and 12: reassessment

1. Give every participant the same skills assessment as in week 2.
2. Measure the same three tasks per role in the same way: time, rework and errors found in review.
3. Ask each participant which of their two practice tasks they still do with AI, and why.
4. Collect the shared library and count the workflows that other people now reuse.

Week 13: decide what happens next

Write a short report for the sponsor. Show the change in each skill, the change in each task measure, and the workflows now in use. State what did not work.

Then make one of three decisions. Extend the program to more teams. Repeat it with changes for the same teams. Or stop, if the measures did not move. Measuring AI training ROI explains how to turn these measures into a cost case.

Time budget per participant

• Assessments: about one hour in total, split between weeks 2 and 12.
• Training sessions: about eight hours across weeks 3 to 6.
• Clinics: about two hours across weeks 7 to 10.
• Practice: no extra hours. Participants do tasks they already do, with AI.

Managers need extra time for weekly review in weeks 7 to 10. Agree that time with them before week 1.

Common ways these programs stall

• Generic content. Training built on examples from other jobs does not transfer to the participants' own tasks.
• No protected time. Practice drops first when deadlines arrive.
• No checking. Use rises, errors rise with it, and the program loses trust.
• A changed measure. A new assessment in week 12 makes the result impossible to read.
• No decision rule. Without an agreed threshold, the week 13 report becomes a debate about what counts as success.

Using Emeri for the baseline

Emeri offers a free 15-minute AI skills check that reports results by skill. Teams can use it as the week 2 baseline and the week 12 reassessment. The teams page covers group use. Any consistent assessment will do, as long as the same one is used both times.