An AI workshop can teach a useful skill without giving people a dependable way to use it in ordinary work. When enthusiasm does not turn into practice, inspect the first independent task after the session. Could the person access the approved tool, recognize a suitable opportunity, remember how to begin and produce something the team could accept?
For a learning lead, those questions change the follow-up. A missing skill may need practice. A missing permission needs a decision. A task that never occurred needs a different observation window. Another workshop is helpful only when it addresses the problem you actually found.
Separate a demonstration from independent work
A demonstration makes the next step visible while a trainer is there. Daily work asks the employee to choose that step later, with different inputs, competing demands and responsibility for the result. Completing the session does not establish that this handoff happened.
In a September 2026 Reddit discussion, a Reddit user described attending several consultant-led AI workshops that ended with a basic Copilot prompt followed by a more detailed one. They wanted to know what came after seeing the improved response. Replies pointed out that beginners might still need that introduction. In a contrasting reply, another Reddit user said their company had helped a team of 14 build agents they could use while learning. That provider account did not report whether the team continued using them afterward.
The distinction is useful without treating either account as representative. A workshop may introduce a skill, demonstrate it or support practice. You still need to inspect what happens when the trainer is absent.
Nor is short training inherently ineffective. A 2026 preprint by Benjamin M. Chen and Hong Bao reported on 164 law students who completed a randomized mock-exam study. Conditions included optional AI access with or without a short training video and quiz. AI use was 41% in the trained group and 26% in the untrained group. This was one institution and one immediate exam task, not a study of sustained workplace adoption. It challenges a blanket dismissal of brief instruction while leaving the follow-through question open.
Ask the workshop owner which result the session was designed to produce. Then ask the line manager where that capability should next appear. Attendance, satisfaction and an impressive demonstration answer different questions from independent performance.
Inspect the first eligible task
Start with a task people already need to complete. An eligible opportunity means the task occurred, the use was permitted and the person had the inputs and access needed to attempt it. Someone who never encountered that opportunity should not be counted as having failed to transfer the learning.
A review of training-transfer research by Rebecca Grossman and Eduardo Salas identifies workplace support, opportunity to perform and follow-up among the factors associated with applying training. It is not an AI-specific experiment or proof of one follow-up schedule. It gives learning leads a reason to investigate the environment around the skill, as well as the lesson itself.
Use the next task to separate the possible breaks in the handoff:
| What you observe | What to investigate | A useful response to test |
|---|---|---|
The person cannot begin without the trainer's example | Can they recognize the task and reconstruct the steps? | Practise that step and provide a short, findable job aid |
No suitable task has occurred | Does this role encounter the use often enough? | Choose a relevant opportunity or wait for the task to recur |
The person cannot access the approved account or inputs | Which access or permission is missing? | Resolve it before judging the person's performance |
The draft is inaccurate or expensive to check | Which errors occur, and what does correction require? | Practise verification or reconsider whether this use is worthwhile |
The person produces a useful draft but nobody accepts it | Who may review and use this output? | Agree the review responsibility and acceptance standard |
A job aid can be a short instruction beside the work, with an approved starting example and the checks that matter. It should help someone recover the process, rather than ask them to remember which slide contained it. Do not infer a particular rate of forgetting from a person needing a reminder.
If access is available but the tool interrupts the work, use the guide to distinguishing usability problems from training needs. If the person needs a broader progression of skills, use the role-specific learning-path guide. Here, the immediate question is where the first real attempt stopped.
Give the first attempt an owner
Suppose Jamie Lee coordinates purchasing for a food-packaging manufacturer. After a workshop on drafting supplier emails, Jamie's next suitable task is a delivery discrepancy. The approved purchase order says 120 cartons; the receiving note records 100. Jamie needs to ask the supplier to confirm the missing 20 cartons and the expected delivery date.
The workshop example used a tidy sample record. This task requires Jamie to choose the permitted information, preserve the quantities and avoid turning an unanswered question into a promised delivery date. The buyer, Claire Bennett, will review the email before it is sent.
Claire asks Jamie to attempt the draft, then show where help was needed. The response follows the difficulty:
- Unclear permission: Claire takes the question about permitted records to its owner.
- Missed discrepancy: They practise comparing the draft with the source quantities.
- Invented delivery promise: They correct the draft and assess whether checking it costs more than the existing method.
That review preserves the task's consequence. A polished email that promises an unconfirmed date can create a supplier dispute. The learning outcome is a usable, checked message, not merely a fluent response.
In a September 2026 LinkedIn article, trainer Reyhaneh Khalilpour argues that instruction needs ownership and permission to change work. Her article promotes an approach to follow-up and describes an unnamed case, so it does not establish an adoption rate or a validated timetable. The practical question it raises fits this example: who will make room for the attempt and act on what it reveals?
Before the workshop ends, identify a line manager or work owner who can answer that question. A trainer can teach a technique but may have no authority to approve the employee's inputs, change a supplier process or accept the output.
Keep the next task within reach
Follow-up should be close enough to a real opportunity that the employee can show the work. A next-day check may suit a daily supplier-email task. A month-end reconciliation may require a later check. The calendar should follow the work, rather than impose the same interval on every role.
A practical sequence is:
Follow up on the first independent AI task
Choose the task
Name the next ordinary task, permitted inputs and required output.
Prepare the attempt
Confirm access, a findable starting aid and who will review the work.
Inspect the output
Let the employee show the attempt, checking effort and point of difficulty.
Adapt the support
Resolve the observed barrier, then check a later comparable task.
Keep the employee doing the work. A manager who takes over every difficult prompt may produce a good result while hiding the step the employee still needs to learn. Coaching can explain a decision, provide feedback and then return the next attempt to the person.
Use the workshop's examples as starting aids, with the tool, permitted inputs and review rules still correct. If the real task differs materially, update the aid or practise that difference. Sending the slide deck again is unlikely to answer which part of a delivery discrepancy Jamie should check.
This sequence is a way to organize an investigation, not a proven treatment with a guaranteed effect. Record what changed and look at the later task. If the same error persists, the response has not yet solved the problem.
Count opportunity and accepted output separately
A learning dashboard can make a successful session look like successful transfer by keeping only the attendance denominator. Retain the route from participation to work instead.
Suppose eight purchasing coordinators complete the workshop. During the observation window, six encounter an eligible supplier-email task. Four attempt the approved method independently, and three produce an email accepted after the required review.
| Observation | What it tells you |
|---|---|
Eight completed the workshop | Participation, not evidence of later use |
Six had an eligible task | The opportunity denominator for this window |
Four of those six attempted it independently | About 67% attempted among those with an opportunity |
Three of the four attempts were accepted | 75% met the agreed standard after review |
A later comparable task is checked | Whether the capability persists beyond the first attempt |
Keep three follow-up decisions separate:
- No opportunity: Retain the two people in the record, without claiming they tried and failed.
- No attempt despite an opportunity: Ask the two eligible people what prevented the attempt.
- An output that was not accepted: Review the error and the effort needed to check or repair it.
“Accepted” should mean the output met a defined work standard after the required review. Record material corrections and reviewer effort so an acceptance count does not conceal expensive rework. Where practical, compare the effort and quality with the existing method on comparable work. These small observations alone do not establish that the workshop caused a productivity gain.
For Jamie's task, Claire can check whether the quantities remain correct, the delivery date remains a question and the email needs substantial rewriting. On the next discrepancy, she can see whether Jamie performs those checks independently. That is a more useful next decision than scheduling another general introduction because a usage chart remained flat.
The people and change guide connects this follow-up to the wider support, responsibilities and working conditions behind adoption.
Questions and answers
How soon should you follow up after an AI workshop?
Follow up at the next real opportunity to use the taught skill, with access, permitted inputs and an output reviewer arranged beforehand. Daily work may allow an early check; a task that happens monthly needs an observation window that includes it. Inspect the employee's independent attempt and checking effort, then revisit a later comparable task.
Use the first-task follow-up sequence to prepare that check. There is no single interval established here as best for every workshop or role.
When is another AI workshop the right response?
Another AI workshop is useful when observation reveals a specific capability gap that relevant practice can address. For example, a purchasing coordinator may need practice comparing the generated supplier email with the source quantities. A repeated general introduction will not resolve a missing account, an unapproved input or a task that never occurs.
Start with the handoff diagnosis. If the need extends beyond one task, build it into a role-specific learning path with appropriate checking responsibilities.
How do you measure whether AI training transfers to work?
Look for independent performance on eligible real tasks, with the output assessed against an agreed standard. Count who had an opportunity separately from who attempted the method and whose output was accepted. Record correction and reviewer effort, then check a later task to distinguish a first attempt from continuing capability.
The purchasing-team measurement example shows why attendance and use need different denominators. A small follow-up record can guide support; it does not by itself prove a causal productivity gain.
What if employees cannot use the approved AI tool after training?
Identify the missing account, permission, input or accessible way to do the task, and give the unresolved condition an owner. Keep the existing permitted process available while it is resolved. An employee blocked from making a legitimate attempt should not be described as having failed to apply training.
Once access is usable, inspect the real task. If the difficulty is navigating the tool or fitting it into the process, the usability and training diagnosis helps choose the next intervention. If the use remains unsuitable, record that decision rather than treating more logins as the outcome.
