Understanding the analysis
The 5-step Analysis View
The Analysis View presents generated summaries, possible patterns, possible drivers, action ideas, and conversation prompts. It is produced from the selected transcript, and each section can be incomplete or wrong.
What goes into the view #
The analysis request uses the edited transcript when one exists, otherwise the raw transcript. Text longer than 50,000 characters is truncated for the model request; the stored transcript itself is not truncated by that operation. The backend sends the text and selected output language to the configured language-model provider, validates the returned structure, and saves the result against the conversation.
The five steps are structured parts of a generated response, not five independent investigations. A missing section or empty field can mean there was not enough supported material, the model omitted it, or the response could not be used. An unsuccessful fallback is marked failed rather than presented as a healthy reading.
| Step | Question it answers | Main output |
|---|---|---|
| 1. Summary | What does the conversation say? | A generated overview. |
| 2. Behavioural Intelligence | What patterns may appear in the words? | Possible observations and interpretations. |
| 3. Root Cause | What might explain the topic discussed? | Possible triggers and contributors. |
| 4. Action Blueprint | What could be considered next? | Suggested actions and measures. |
| 5. Conversation Strategy | How might a follow-up conversation be approached? | Possible goals, questions, and listening cues. |
Read claims as possibilities #
The model is instructed to use evidence and probabilistic wording. Where available, the interface shows supporting phrases and signal-confidence labels. These do not prove that an interpretation is correct: quotes can be incomplete, transcript text can be inaccurate, and the model can infer the wrong cause from limited context.
Check the original transcript, separate direct observations from interpretations, and ask the employee rather than treating a possible driver as known. Analysis outputs do not become confirmed commitments automatically.
Features that are not actions #
Some parts of the five-step screen are presentation-only. The Signal Map and hardcoded probe questions are not live analysis features. The visible Set reminder control only toggles its appearance and does not schedule a reminder; Progress autosaved is static interface text, not proof that this screen autosaves. Use the actual supported meeting, follow-up, or conversation-memory workflows when you need a persisted record.
If a control's status is unclear, verify whether a saved record appears in the relevant workspace before relying on it.
Use this guidance thoughtfully
This page describes the current product workflow; it is not an HR policy or a promise about every deployment. Before acting on information about an employee, verify the source record, consider context the system cannot see, and use your organization's process. If a control behaves differently from this guide, check the current workspace UI and confirm the saved result rather than assuming an action completed.