What problem is the training meant to solve?
Write the problem in one sentence before anything else. Typical versions: licences the organisation pays for are barely used; staff are using personal AI accounts and nobody knows what goes into them; a new AI policy exists but nobody has practised it; a team has asked for help with a specific workload. Each points to different training and different measures. Some problems are not training problems. If staff have no approved tool, or no permission to change how they work, a workshop will not fix it, and the case should say what has to change first.
What does staff AI training actually cost?
The provider's fee is the visible cost and often not the largest. Count the hours everyone spends in the session, the time spent beforehand collecting tasks and checking access, and the follow-up time that turns a session into a habit. Add venue and travel where they apply. Licence costs belong in the case only if the training depends on buying them; many organisations can train on a tool they already have. Show each line separately so leadership can see which ones scale with headcount.
Hypothetical example: a team of 12 attends a half-day session of 3.5 hours, which is 42 staff hours. Add an assumed 6 hours of preparation across the team lead and IT, and 1 hour each of follow-up practice, another 12 hours. That is 60 staff hours. At an assumed average cost of $60 an hour, the staff time alone is $3,600 before any provider fee. Your own figures will differ.
Which benefits can you state honestly before training?
Before training, you cannot know how much time a particular team will save, and a number in the case invites a challenge you cannot answer. You can state benefits that rest on current facts. The National AI Centre notes that some staff are likely already using AI at work without their employer knowing, and advises investing in training and policies. The OAIC recommends, as best practice, not entering personal information into publicly available AI tools, which only works if staff know the rule. The National AI Centre also notes that work health and safety laws require training so people can work safely, including when using AI.
- Less use of personal accounts for work, once an approved tool and rules exist.
- Use of licences already paid for, visible in the vendor's usage reporting.
- One shared rule on what information stays out of AI tools.
- Staff who can say what they must check before an AI draft is used.
How do you size the opportunity without inventing figures?
Use your own tasks instead of industry averages. List the recurring tasks the team says AI might help with, and for each record how often it happens, roughly how long it takes, how bad a mistake would be and whether the input is already digital. Ask the people who do the work for the estimates and label them as estimates. A task that happens daily, takes an hour and has a checkable output is a better candidate than a monthly report with legal consequences. After training, time a few real instances to replace the estimates with measurements.
What will leadership ask, and how can you answer?
Expect a handful of predictable questions, and prepare a short factual answer to each instead of a longer document. The National AI Centre's foundations guidance suggests assigning a senior leader as the overall owner of AI governance, so name that person in the case: leadership will want to know who is accountable for the rules the training teaches. Keep every answer tied to something you can check later, and leave out claims that depend on results you have not measured.
- Why now? Point to the current state: licences, unapproved use or a new policy.
- Why not let people learn on their own? Unmanaged learning produces unmanaged rules.
- Why an external provider? Explain what you lack internally, such as time or facilitation.
- What if nobody uses it afterwards? Name the follow-up and the manager's role.
- How will we know it worked? Name the measures agreed before the session.
Which measures should be agreed before approval?
Agree the measures at the same time as the budget, because a baseline cannot be collected afterwards. In Microsoft 365 organisations, the Copilot usage report in the admin centre shows enabled users, active users and the active users rate over 7, 28, 90 or 180 days, so record those figures before the session. Add a short survey of confidence and current use, and keep a few samples of current work on the tasks you plan to practise. Then set review points at which the same data is collected again and compared.