The short answer.

AI training fails when it teaches the tool, because the tool is the same for everyone and the job is not. A course on ChatGPT, Claude, Copilot or Gemini shows people what the assistant can do; it does not show the accounts lead how to run the month-end report she runs. Train each person on the workflows their own job runs, on their own past work, and give them a one-sheet job aid that says which workflow to reach for and how to start it. The course is forgotten by spring. The sheet is still on the desk.

How do we train our team on AI? sets out the two layers of training and who leads them. This piece is about what happens after training: what a person has in front of them the next time the work comes round, and who helps when they get stuck.

Learning the buttons leaves the work unchanged.

Training on the tool teaches where things are: how to start a conversation, how to attach a file, how to write a clearer request. Those are real lessons, and they are the same lessons for a scheduler, a bookkeeper and a sales lead. That sameness is the trouble. Nobody leaves the session knowing which piece of their own week has changed.

So people go back to the queue they left, and the queue looks exactly as it did. The month-end report gets built the old way because the old way is the one they know works. Within a few weeks the session is a pleasant memory and the assistant is something a few keen colleagues use. How do we get people to actually use AI at work? covers why a workflow built from the team’s own method changes the week where a course does not.

The person fifteen years into the job is not the problem.

The person who has done a finance job for fifteen years knows the work better than any trainer in the room. What that person is not, and should not have to be, is in a perpetual learning state. They have a close to run, people waiting on their numbers and a standard they have held for years. Asking them to keep exploring a new tool on top of all that is asking the wrong thing of the right person.

That person needs something smaller and more useful: to know that for this report there is a workflow, where it is, and what to check before the figures go out. The expertise stays theirs. The workflow prepares the draft; they decide, as they always have.

It also helps to picture where people start. Few teams are at zero. One person is at 25 and could be at 60; another is at 40 and could be at 80. The levels of using an AI assistant at work describes the steps between. Good training meets each person at their number and moves them up through their own work.

Train on the job’s own workflows, on its own past work.

Training that holds starts from the job and works back to the tool. Each role learns the two or three workflows it runs, in hands-on sessions where people run each workflow on last month’s real cases and compare the draft with what they actually sent. The tool is learned along the way, because the work needs it.

The curriculum comes from the plan. In FullyPowered AI Planner, each team’s roles sit beside the workflows they run, so the learning path for each job is already written down: these workflows, for this role, in this order. A new person joining the role follows the same path; How do new hires learn the AI workflows their role runs? covers that case.

What a job aid does that a course does not.

A job aid keeps the lesson where the work happens. It is one sheet, letter size, folded in half, laminated if you like, that sits on the desk or goes in a bag. It lists the workflows this job uses, when to reach for each one, and how to start it.

A course asks a person to remember. A job aid asks them to look. On the day the quarterly report is due, the person fifteen years into the job opens the sheet and reads: if I want to run this report, this is how I do it. That moment is the whole point. Nobody needs to recall a session from March; the answer is folded on the desk.

A job aid is also a file. When a workflow changes, print the sheet again. The sheet is always current, because the person who owns the workflow keeps it so.

A one-sheet job aid to copy.

Fold a letter-size sheet in half and fill in these four panels for one job. Keep it to what fits.

  1. Front: the job.The role and team, named for the work. The date this sheet was printed. The one person to ask, by name.
  2. Inside left: the workflows this job uses.For each one, three to five in all: its name, the moment to reach for it (“when the month closes”, “when a new supplier invoice arrives”), and where it lives.
  3. Inside right: how to start, and what you check.For each workflow: what it starts from (the export, the file, the queue), the first step, and the points you read before anything leaves your desk.
  4. Back: when you are stuck.What to do when the draft looks wrong, what to try next, and where to look for an answer: the workflow’s owner, the team’s shared folder, the reference page for that workflow.

Three habits that keep people going when the sheet runs out.

Hands-on training teaches three habits alongside the steps, because no sheet covers every case. These are the habits good technical teachers give beginners, and they suit anyone learning a new way to do familiar work.

  • Working out what went wrong.When a draft is off, look at what the workflow started from before blaming the workflow. Most misses come from the wrong export or a missing file.
  • Keeping going when stuck.Try the next thing on the back of the sheet, then ask. Being stuck is part of learning the work, not a sign of failing at it.
  • Knowing where to look.The workflow’s owner, the team’s shared folder, the workflow’s reference page. A person who knows where answers live does not need to carry them all.

Managers carry it after the training ends.

After the sessions end, the manager is the person people ask, so the manager is trained first and equipped to answer. That is the core of change support, what an HR team knows as change management: say plainly why the change is happening and why it matters, give managers the answers, and give people a place to raise concerns and be heard.

Follow adoption by team and by workflow. Which workflows are running, in which teams, and where a team has stalled. That tells you where to send help. People are supported, never rated: nobody is scored on how fast they took to it, and a team that is stuck gets another session, not a mark against it.

One engineering firm trains its accounts team.

Take a 70-person civil engineering firm whose accounts team is three people. The accounts lead has run the firm’s books for fifteen years. The plan gives the accounts team four workflows: the month-end project cost report, supplier invoice coding, the monthly note to project managers on unbilled time, and the year-end file for the outside accountant.

Illustrative example.

  • Hands-on sessions.The team runs each workflow on last month’s real reports and invoices and compares the drafts with what it actually sent. Nobody sits through a general tour of the assistant.
  • The job aid.Each person gets a folded sheet for their own job. The accounts lead’s inside left panel lists the month-end report first, with “when the month closes” beside it.
  • The manager.The finance director is trained first, so questions about why the change is happening go to someone with an answer.
  • Following adoption.The finance director looks at which of the four workflows are running each month, not at who is fastest. When the year-end file stalls, the team gets a second session on that one workflow.

The unbilled time note began from a free workflow template in the Workflows Directory, adapted to the firm’s own projects before anyone trained on it.

Start from the plan, and the training, the job aids and the change support follow from it. FullyPowered AI Learning and Development is what we offer for this: hands-on training, a curriculum built for your company, a job aid for every job and change support, scoped to your company. How organizations plan their use of AI covers the rest of the method.