General Manager
“You need to communicate more clearly” gives a learner a problem without showing them where to begin. I would rather point to the moment they offered a delivery date, the information they had at that point and the question they could ask before making the same commitment again.
That is the job I give feedback in Practice-Native Learning, the methodology I am developing at Altaius. It should connect the decision record to another useful attempt. A fluent paragraph about leadership is not enough.
Tannenbaum and Cerasoli's 2013 meta-analysis covered 46 samples and 2,136 participants. It reported an average debriefing effect of d = 0.67 and highlighted the importance of alignment and structure. This was evidence across the interventions studied, not a predicted Altaius result. [1]
There is also a reason to be careful. Kluger and DeNisi's 1996 review analysed 607 effect sizes and 23,663 observations. Feedback improved performance on average, but more than one-third of the interventions reduced it. That does not make feedback undesirable; it makes the design and focus of feedback important. [2]
Suppose a participant promises to resolve a complaint before confirming who has authority to authorise the remedy. This is an illustrative example, not a client record. The feedback could identify the exact commitment, the missing authority check and the risk introduced by proceeding without it.
I would then ask what the learner believed at the time. Perhaps they thought the customer needed an immediate answer. Perhaps the scenario failed to make the authority boundary discoverable. The discussion should leave room for both possibilities.
An explanation that assumes the scoring rule must be right can conceal a design defect. The learner needs a route to question inaccurate evidence or an unreasonable interpretation.
I would organise a short debrief around these prompts:
The prompts are an Altaius design suggestion. They are not a validated instrument or a mandatory script. The facilitator can follow the significant decision rather than asking every question mechanically.
The next attempt should test the proposed change. If the learner says they will check authority before offering a remedy, a different case can reveal whether they do so when the pressure comes from another direction. Repeating a memorised answer would offer weaker evidence.
If an AI coach supplied the decisive question, that help belongs in the record. The learner may have made progress with support, which is useful, but the result is different from independently recognising the need to ask it.
I would also separate observation from interpretation. “The participant asked about approval after making the offer” is an observation. “The participant consistently neglects governance” is a much broader inference. One episode does not justify that leap.
For managers, access should serve an agreed development purpose. Sharing an entire conversation when a specific excerpt will support the discussion may expose more information than the task requires. Agree the audience, scope and review process before using the record.
Before relying on feedback, inspect whether it quotes the correct speaker, uses the relevant scenario state, explains its criteria and proposes an action the learner can actually take. Review disputed or inconsistent cases. If the system cannot establish what happened, it should say so.
The end of a debrief is not a polished summary. It is a clearer next attempt, grounded in an event the learner and reviewer can both examine.
Read the Altaius definition of Practice-Native Learning
Previous in this series: Designing Decisions for Practice-Native Learning
