Skip to content

Sentiment Analysis

Sentiment estimates the customer’s emotional tone during the conversation (positive/neutral/negative) and can be used for triage and coaching.

Choose one:

  1. Dropdown: Positive / Neutral / Negative / Mixed / Unknown
  2. Numeric score: -1 to +1 or 1–5 (less common for end users)

Dropdown is recommended for most deployments because it provides clear categories for dashboards and filters (e.g., "Show all Negative sentiment conversations").

EDITOR NOTE: sentiment model

Ask stakeholders (choose one):

  • A) Dropdown sentiment (recommended)
  • B) Numeric sentiment score
  • C) Both (score + label)

Best-guess recommendation:

  • Use dropdown for clarity and dashboards.

Configuration overview

  1. Create/verify the sentiment field (dropdown or numeric).
  2. Enable the sentiment AI Task (prebuilt) or create a tenant task.
  3. Apply filters (optional):
  4. skip very short calls
  5. Validate using Playground.

Prompt recommendations

  • Define label meanings (what counts as “Mixed”)
  • Require “Unknown” when insufficient evidence
  • Provide a short explanation citing transcript evidence