A four-step method for training employees on the tasks they already do, and for checking whether the training changed anything.
Most AI training plans start with a tool and end with a workshop. Employees learn what the software can do, then go back to their desks and decide on their own whether any of it applies to their jobs. A few people find uses. Most return to the way they worked before.
Your business runs on roles and the tasks each role repeats every week. An estimator builds quotes. An office manager answers customer questions and schedules work. A project lead writes status updates. When training starts from those tasks, every employee knows what they're supposed to get better at, what good work looks like, and how anyone will tell whether they improved.
This guide walks through four steps to build that kind of plan:
Map each role and its recurring tasks.
Define good output and assess current skill.
Train on the employee's own work.
Measure changed work and assign ownership.
You can start the first step with a spreadsheet and a list of your employees. Each step after that builds on the one before it.
Map Each Role and Its Recurring Tasks
Group People by the Work They Produce
Department names tell you where someone sits on the org chart. They say very little about what that person makes during a workweek. A sales department might include a rep who writes proposals and a coordinator who enters orders and tracks shipments. Those two jobs need different AI skills. Meanwhile, a customer service rep and an accounts receivable clerk in separate departments may both spend hours each week writing customer emails.
Open a spreadsheet and list every role in your business by the output it produces. Use plain labels such as "writes proposals," "answers customer questions," or "prepares monthly reports." Then add the names of the people in each role. Some employees will appear in more than one row, and that's fine.
Choose Two or Three Recurring Tasks per Role
For each role, pick two or three tasks that meet three conditions:
Frequent. The task happens at least weekly, so employees get regular practice.
Time-consuming. The task takes enough time that an improvement would be noticeable.
Reviewable. Someone can look at the finished output and judge whether it's good.
Good candidates include quotes and estimates, follow-up emails, meeting summaries, job descriptions, status updates, and first drafts of reports. Leave out tasks that happen a few times a year, along with high-stakes decisions that depend heavily on one person's judgment. Hold any task involving sensitive customer, employee, or financial data until you've set data boundaries in the next step.
Ask the people in each role which tasks they would pick. They know where their hours go better than anyone else does, and their input makes the plan easier to adopt later.
Record How Long Each Task Takes Today
Before any training happens, write down a starting point for every task you selected. Add these columns to your spreadsheet:
How often the task happens
About how long one instance takes
How many rounds of revision it usually needs
Who reviews the finished work
Estimates from the employee are enough to begin. If you want more accuracy, have people track the task for one week. You'll compare against these numbers when you measure results in step four, so keep the spreadsheet somewhere the whole team can find it.
Define Good Output and Assess Current Skill
Describe What Good Work Looks Like for Each Task
Training gives people a target only when the target is written down. For every task on your spreadsheet, add a short description of an acceptable result. Keep it practical. A good customer reply might answer the question in the first two sentences, use the customer's name, match your company's tone, and contain no promises about pricing or delivery dates that a manager hasn't approved. A good meeting summary might list decisions, owners, and deadlines on one page.
Write these descriptions with the people who do the work and the people who review it. When both groups agree on the standard, employees stop guessing what their manager wants, and managers stop rewriting drafts to match a preference nobody stated.
Set Data Boundaries Before Anyone Starts
Every employee should know three things before they use AI on company work:
Which AI tools the business has approved
What information can go into those tools
What information must stay out, such as customer personal details, employee records, financial data, and confidential contracts
Put these rules on one page in plain language. If your business operates in a regulated field, check your specific obligations before you finalize the list. People make better decisions when the rules fit on a single screen.
Name a Reviewer for Every Task
AI can produce a confident draft that contains wrong facts or misses context. Assign a person to review the output of each task, and decide when a draft needs a second look. A routine internal summary might need a quick check. A quote going to a customer should get a full review by someone who knows the pricing.
Find Out Where Each Employee Stands
Your team will start at different levels. Some people already use AI every day. Others tried it once and quit. Putting everyone in the same session wastes time for both groups.
A simple assessment works. Ask each person to complete one of their selected tasks using AI, then show you the prompt they wrote and the output they got. You'll see quickly who needs basics, who needs better habits, and who could help teach others. Sort people into three groups (beginning, regular, and advanced) and plan training for each group separately.
Train on the Employee's Own Work
Use Real Tasks as the Training Material
Generic examples make training easy to deliver and hard to apply. When an estimator spends a session writing marketing slogans, they leave with a skill they won't use on Monday. Build each session around the tasks you selected in step one, using the standards you wrote in step two.
Run sessions by role and by skill level. Beginners in the office manager role might practice turning a rough list of notes into a customer reply. Advanced users in the same role might build a reusable brief that handles the five most common customer questions. Both groups work on the same kind of task at a pace that fits them.
Use sample data or cleaned-up versions of real documents during training. Anything that would break the data rules from step two should stay out of the session.
End Every Session With Something Usable
Each session should produce a prompt, brief, or short workflow the employee can use on their next real task. If people walk out with notes and nothing else, the training was a lecture. If they walk out with a working prompt for their weekly status report, they'll use it within days, and that use is where the skill develops.
Keep sessions short. An hour focused on one task usually does more than a full day covering everything AI can do.
Build a Shared Library for Each Role
Good prompts tend to live in one person's head or one person's chat history. When that person is out sick or leaves the company, the knowledge goes with them.
Create a shared folder for each role. Store the prompts and briefs that produced good results, along with a short note on when to use each one and what to check in the output. Include the context a prompt needs, such as your tone, your product names, and your standard terms, so a new employee can reuse it without asking the person who wrote it. Review the library monthly and remove anything that no longer works.
Keep Review Inside Every Workflow
Every workflow you teach should include the review step from step two. Show employees what to check: facts, numbers, names, tone, and anything the AI may have invented. Tell them when to escalate. A rule as simple as "if the draft mentions price, contract terms, or a deadline, send it to your manager before it goes out" prevents most of the expensive mistakes.
Measure Changed Work and Assign Ownership
Compare Results at 30 Days
Thirty days after training, return to the spreadsheet you started in step one. For each task, ask the people doing it to estimate the same numbers again: how long one instance takes, how many revision rounds it needs, and how often the reviewer sends it back. Then ask the reviewer whether the quality meets the standard you wrote in step two.
Put the before and after numbers side by side. Some tasks will show clear improvement. Others will show little change, and those deserve a closer look. Maybe the prompt needs work, the task was a poor fit for AI, or the employee needs another session. Each answer points to a specific fix.
Treat these figures as working estimates. They're good enough to guide decisions about where to spend more training time, even if they wouldn't pass an audit.
Stop Counting Logins and Certificates
Attendance records show who was in the room. Login counts show who opened the tool. Certificates show who finished a course. None of those tell you whether a quote went out faster or a report needed fewer corrections.
Report results to leadership in terms of the work. "Meeting summaries now take about 15 minutes instead of 45, and managers approve most of them without changes" gives an owner something to act on. "90% of staff completed training" does not. This also helps managers and executives agree on what progress looks like, since both groups are reading the same numbers about the same tasks.
Assign an Owner for Each Role
Plans without an owner fade within a quarter. Pick one person per role to keep the plan running. That person maintains the shared library, collects the 30-day numbers, and flags problems early. The owner doesn't need to be the most advanced AI user on the team. They need to care about the work and have an hour or two each month to spend on it.
Review Monthly and Add New Tasks
Hold a short monthly meeting with the role owners. Review what improved, what stalled, and what employees asked about. Once the first two or three tasks in a role are working well, add the next task from the list. Train it, set a starting point, and measure it the same way. Over a few months, each role builds a set of tested workflows that new hires can learn in their first weeks on the job.
Every step in this plan depends on one thing: knowing what your people actually do and how well they do it with AI today. Roles and tasks give training a target. Written standards and named reviewers keep quality steady. Shared libraries keep good work from walking out the door, and 30-day measurements show whether any of it changed the work.
The assessment in step two is where most owners stall. Testing one person takes a few minutes. Testing a workforce of 50 or 200 people, then turning the results into a training budget leadership will approve, takes much longer. That's the part our AI Impact Analysis was built to handle.
You answer 35 questions about your company size, industry, priorities, and how your workforce uses AI today. Most people finish in 5 to 10 minutes, and every question is a dropdown or short answer. Your responses are converted into a profile of where your employees fall on the 10 Levels of AI Mastery, so you can see where your people cluster before you plan a single session.
The report then shows you:
8 to 15 specific AI tools your employees could build, matched to your industry and departments
An estimated dollar value of the time your workforce recaptures (or doesn't) at its current skill level, calculated from your own headcount and employee cost, plus a projection of what changes if your team moves up one level
Two or three concepts for larger custom AI applications, with cost ranges and timelines, so you can compare them against training
One recommended next step tied to your data and explained in plain language
One of our INGRAIN AI Certified Implementers reviews every report before delivery, and you'll have it within 24 hours. An optional 1-hour consultation lets you walk through the findings and decide what to act on first. The analysis costs a one-time $547.
You already know your roles and your recurring work. The analysis tells you where your people stand and what that's worth in dollars, which turns the first two steps of this plan into numbers you can bring to your leadership team.