Skip to main content
An outcome describes what happened; extracted variables capture business facts; QA evaluates conversation quality. Configure these separately and inspect a completed run before building downstream rules around their values.

​
Configure call-disposition extraction

  1. Open the agent’s Settings → General.
  2. Enable Extract call disposition at the end of the call.
  3. In the configuration dialog, keep suitable defaults and add custom Disposition Code / description pairs. Use unique, meaningful codes such as appointment_confirmed with a description explaining exactly when it applies.
  4. Select Enable extraction or Save options, then Save General Settings. Closing the options dialog alone does not publish the workflow.
  5. Test a conversation for each intended outcome and inspect the run. Publish when the classification matches the task.
With extraction disabled, Athena keeps the disposition recorded by the call-ending event. With it enabled, classification chooses from the configured options at the end. These codes are recorded before organization-level mapping. Tools such as End Call and Transfer Call can also set an outcome; check the final value when combining them with extraction.

​
Map outcomes to your own codes

Open Platform Settings → Preferences and enable Disposition mapping, then Configure mapping. Map Athena outcome codes to the codes expected by your reporting or PBX integration, save the mapping, and save preferences when required by the form. Mapping affects webhooks, run filters, reports, and external PBX write-backs. Coordinate consumers before changing it. Turning the toggle off stops applying the mapping while preserving configured entries. Verify a run’s final displayed/exported value and its receiver payload instead of assuming the unmapped and mapped code are interchangeable.

​
Configure QA Analysis

  1. On the workflow canvas, add QA Analysis. It runs after the conversation and has no conversation edges.
  2. Enable the node and set System Prompt to the quality criteria you want reviewed. Start from the default instructions and preserve the expected structured output when customizing them.
  3. Set Minimum Call Duration (seconds), Include Voicemail Calls, and Sample Rate (%). The sample rate is 1–100; use 100 for a controlled test so a sample does not accidentally skip it.
  4. Use Use Workflow’s LLM, or select the QA provider/model and required authentication fields. Azure also requires its endpoint.
  5. Save, test an eligible call, and inspect the run’s QA results or annotations. Also verify that intentionally ineligible calls are skipped. Publish the tested definition.
QA instructions support {node_summary}, {previous_conversation_summary}, {transcript}, and {metrics} placeholders. Do not replace these single-brace QA placeholders with the double-brace context syntax used in workflow templates. QA requires a usable transcript and model access. Sampling, minimum duration, voicemail exclusions, or processing failure can explain a missing review. A QA score is an automated assessment, not proof that a call met every business requirement.

​
Read daily reports and export CSV

  1. Open Reports to view Daily Reports.
  2. Choose a date and All Workflows or one agent. Confirm the displayed time zone; its default comes from organization preferences.
  3. Review total runs, transfers, outcome distribution, and duration distribution.
  4. Choose Download CSV when the selection has data. The daily export contains phone number, disposition, duration in seconds, and a workflow-run URL.
  5. Open a few corresponding runs to check the selection and outcome interpretation before using the export in another system.
Use Settings → Report within an agent for that workflow’s report, and the campaign report for campaign-specific results. These scopes are separate from an organization daily report. Changing the time zone can move calls across date boundaries. See Run review for transcripts, activity, media, and split-track audio.