Build a coaching evidence packet around the behavior and its context, including both supporting examples and information that could change the interpretation. The goal is to make the discussion specific and reviewable. A dashboard score alone does not show what happened, why the behavior mattered, or whether the agent had a workable alternative.
Level AI describes interaction-based coaching. AWS documents evaluations alongside contact evidence. Treat these product descriptions as evidence-access leads; they do not establish that every recording, screen view, or relevant policy will be available in your implementation.
What should the packet include?
Start with the current criterion and its plain-language meaning. Include the relevant interaction excerpt with enough surrounding context to avoid distorting the customer's request. Record the policy version, contact reason, and any known workflow constraint. Add a constructive example where the desired behavior is visible.
| Packet element | Question it answers |
|---|---|
| Criterion and version | What expectation applies? |
| Interaction evidence | What was said or done? |
| Context and constraints | What choices were available? |
| Contrasting example | What does the desired behavior look like? |
| Agent perspective | What information may be missing? |
| Open question | What must be resolved before a plan? |
This is a proposed preparation checklist, not a list of native software fields. Keep the packet concise enough to discuss; link to supporting evidence rather than burying the behavior in a long transcript dump.
How should examples be selected fairly?
State the selection rule and review more than one opportunity when the decision concerns a pattern. Include relevant successes as well as failures. Check whether examples come from comparable contact types and whether the same rubric applies to each.
Avoid using an isolated emotionally charged exchange to support a broad conclusion. If the evidence is limited to one case, frame the session around that case. Allow the agent to identify missing context or a different interpretation before moving to a development commitment.
What must be verified before the meeting?
Open each evidence link with the permissions expected for the meeting participants. Confirm that timestamps, transcripts, and evaluation notes refer to the same interaction. Check that a cited policy actually supports the feedback. Resolve a disputed or obsolete criterion before treating it as the basis for an improvement plan.
Handle sensitive customer and agent information through the organization's established process. Prepare only the material needed for the coaching purpose, and avoid copying unnecessary data into new documents merely for convenience.
How should evidence become an action?
Ask the agent to explain the decision, then identify whether the next step concerns skill, knowledge, or an operating constraint. Keep the evidence separate from the conclusion so another reviewer can understand the reasoning. Use the action-plan template only once the behavior and its cause are sufficiently clear.
When reviewers disagree on the underlying criterion, use QA calibration rather than resolving a scoring problem through repeated coaching.
How should a recorded-call evidence packet preserve a disputed decision?
In an illustrative packet, a scorecard marks “next step not explained.” The recording contains an explanation, but the transcript drops a sentence. Attach the recording timestamp, transcript discrepancy and criterion version; keep the finding disputed until a reviewer resolves it. Do not build a coaching plan around the unverified score.
| Packet item | Illustrative entry |
|---|---|
| Finding | Next-step explanation marked missing; review pending. |
| Evidence locator | Contact reference and recording interval with sufficient preceding context. |
| Conflicting evidence | Transcript omission and agent explanation preserved. |
| Decision owner | Reviewer resolves the criterion result before the manager assigns practice. |
For calls selected from automated QA, retain the model or rule version as well as the scorecard version when available. Ask the manager to open the evidence under their normal access before the session. If playback is missing or the excerpt cannot support the finding, mark it unavailable and obtain a reviewable example. A complete packet makes a discussion auditable; it does not establish that the original finding is correct.
What are the limits of this guide?
Sources were checked October 20, 2026. No coaching evidence system was tested. Availability, synchronization, retention, and access differ by deployment and require confirmation. The checklist does not replace employment policy or a formal dispute process. Incomplete recordings or ambiguous criteria should remain explicit limitations rather than being filled in by generated explanations.