Why should people apply AI learning between sessions?

A technique can make sense while a facilitator demonstrates it and still prove difficult to use alone. The learner's own task may contain incomplete information, a different risk threshold or an AI output that does not resemble the example.

Application between sessions creates the missing experience. It lets people discover what they can do, where they hesitate and what support the next learning interaction should provide.

Understanding a technique is different from using it at work

During instruction, the task is usually selected, the materials are ready and help is close. At work, someone must recognise the opportunity, frame the goal, choose appropriate information and decide whether AI is suitable before they can use the technique.

Generative AI adds variability. Two attempts can produce different outputs, and a plausible result may still fail the professional purpose. Effective use requires planning, monitoring and a willingness to change strategy. Those demands become visible when the learner acts in context.

A between-session task turns general understanding into a concrete decision. It can reveal that a prompt method works for drafting but not for evidence-sensitive analysis, or that the main difficulty is not the tool but access to approved source material.

One-off delivery leaves no evidence from later use

A session can include realistic practice, and that remains valuable. It cannot fully reproduce every learner's role, workflow or organisational environment. Transfer research has long treated the opportunity to perform and support in the work environment as important conditions.

Without later application, the organisation does not know whether people can recognise the relevant task, use the method away from the exercise or obtain manager support. Participants may also have no questions yet. Asking “Any questions?” immediately after a demonstration often comes before they have encountered the difficult part.

This does not make workplace application automatically better than structured training. Unsupervised use can reinforce a weak habit or create risk. The learning design must specify what can be tried and what remains outside the experiment.

Design a bounded application task

Choose one role-relevant behaviour and one timely opportunity. The task should be small enough to examine but substantial enough to require judgement. “Use AI this week” is too broad. “Compare an AI-generated summary of this approved synthetic case with its source and record one unsupported claim” is observable.

Define the purpose, permitted tool and information, required checks and stopping conditions. Early attempts should not affect a customer, formal decision or production process. Live work may be appropriate where local controls permit low-consequence use, but synthetic or sanitised material is often a safer starting point.

Give participants a short capture prompt. What were you trying to achieve? What did you do? What evidence did you use? What happened? What remains uncertain? They do not need to store whole prompts or sensitive outputs when a data-minimised observation is enough.

Provide an alternative for people without a suitable work opportunity. Uneven access is a programme condition, not evidence that an individual lacks motivation.

Bring experience back without claiming transfer too early

The next session should use the observations. Compare recurring difficulties, correct material misconceptions and let participants try a changed version. A facilitator can separate a tool-specific issue from a capability need that applies across roles.

Managers matter. They can protect time, identify safe tasks and clarify where experimentation is allowed. They should avoid turning openness about mistakes into a performance penalty, which would discourage useful reflection.

One successful attempt does not demonstrate lasting transfer. The case may have been unusually easy or the learner may have followed a memorised route. Look for application across changed tasks and later opportunities, with evidence proportionate to the risk.

Between-session application is most valuable as part of a loop. Learning prepares the attempt, the attempt creates experience, reflection makes that experience examinable, and the next session helps the learner build from it.

Example

Insurance operations professionals attend a focused session on comparing AI summaries with source material. Each person then uses an approved synthetic case and records which claim was hardest to verify.

One participant finds that the summary merges two separate dates. Another notices that a cautious source statement has become certain. The next session uses these observations to practise traceability and uncertainty on a changed case.

No live claim or customer decision is affected. The work attempt gives the group evidence that a demonstration alone could not provide.

FAQs

• Should between-session practice use live work?

  Use live work only where the task, information, tool and consequence fit local controls. Synthetic, public or sanitised material is preferable when the learner is new, the information is sensitive or the decision is consequential.

• What if someone has no opportunity to apply the learning?

  Provide a role-relevant scenario or work sample. Managers should also examine whether access, workload or permissions are blocking practice. Lack of opportunity should not be treated automatically as learner failure.

• Does one successful attempt prove the learning transferred?

  No. It is early evidence from one context. Stronger conclusions require changed tasks, later opportunities and an explanation of the judgement used, with appropriate privacy and fairness safeguards.