How Voice IntELIEgence works
A voice application is one saved configuration for one kind of call. Every conversation, transcript, output and export job belongs to exactly one application, and each application has its own menu on the left. Dashboard comes first, then Conversations, then the insight pages such as Analytics, then Settings, Personas, Import, Export and Access. Which items you see depends on the application's settings and on what you are allowed to do in it. Chat is reached from Ask in Chat at the top of the application.
What an application decides
| What it decides | Where you set it |
|---|---|
| Where conversations come from | Upload conversation on Conversations and Dashboard, and connectors on Import |
| What is masked before any AI step | Settings, Redaction (PII), through a Smart Redact from IntELIEDocs |
| Which summary prompt each call gets | Settings, Templates & Routing |
| Which outputs are produced | Settings, Outputs & Grading: Summary, Action items, Topics & sentiment, Talk-time stats and Grade |
| What a grade is scored against | The Grading rubric on Outputs & Grading |
| Whether agents are named from their voice | Settings, Details, Speaker identification, and the voices enrolled on Personas |
| Whether calls feed a knowledge base and chat | Settings, KB & Chat |
| Whether a person must approve each call | Settings, Details, Require human review |
| Where outputs are delivered | Export |
| Who can do what in this application | Access |
How a conversation is processed
Each recording becomes a conversation and runs through up to seven stages in order: Transcribe, Speaker map, PII redact, Classify, Understand, KB embed and Export. Transcribe and Understand always run. The others run only when the application uses them: Speaker map while speaker identification is on (it is on by default), PII redact and Classify once you switch them on, KB embed once a knowledge base is attached, and Export once a destination exists. Settings, Details draws this as the Pipeline, with a stage that will not run drawn dashed.
The status badge on each conversation tells you where it is, and the same stages run along the top of a conversation's own page. See Conversation statuses.
Where review fits
With Require human review set to Human review (the default), a conversation that finishes processing waits as Waiting for review until someone approves it on its own page. Set to Automated, Elie approves it for you and it goes straight to Completed. Before approving, a reviewer can rename speakers, edit the summary and the action items, or send the call back through part of the pipeline with Flag & Retry. Approving locks the conversation for good. See Reviewing a conversation.
Settings decide which pages exist
Topics & Sentiment appears in the menu only while the Topics & sentiment output is on, Grading only while Grade is on, and Personas only while Speaker identification is on. Ask in Chat appears once a knowledge base is attached and chat is switched on. Switching one off removes it outright rather than greying it out.
Every save pins a new version
Saving a Settings section, an export destination or a payload mapping creates a new version of the application, which is why Settings says Changes apply to new conversations only (a new version is pinned). A conversation keeps the version it started under, so an edit never rewrites a summary, grade or transcript that already exists. To bring an older call that is not yet approved up to date, send it through in full with Re-sync under Flag & Retry.
Next
- Creating a voice application - the three-step wizard.
- Voice application settings - every section in order.
- Conversation statuses - what each badge means.