Skip to main content

​
Intelligent Monitoring

Each call is automatically distilled into structured fields. Define your own Custom Call Outcomes in the app, and we’ll populate them from the call transcript—reliably and in a strict schema. It turns messy call transcripts into simple fields you can trust, so you can search, filter, and act on calls without rereading them.
At a glance
  • Add fields per agent under Post-processing → Custom call outcomes
  • Choose a type (Boolean, Text, Number, Category)
  • Select Reason to add an optional explanation for each value
  • View results in Call History and receive them via webhook
Call Analysis Outcome Both Types telli provides both System Call Outcomes (built-in fields managed by the telli team like Summary, Dialogue, Transfer, Voicemail) and Custom Call Outcomes (fields you define and manage yourself). Both types appear in Call History and webhooks under the call_analysis and call_outcome keys.

​
How it works

A dedicated extraction model analyzes the conversation transcript and fills your fields:
  • Schema-strict: output always matches the field type you selected.
  • No guessing: if a value isn’t clearly supported by the transcript, it’s null.
  • Robust to STT: built for speech-to-text imperfections and everyday phrasing.
If Reason is enabled for a field, we also return a short explanation for the chosen value.

​
Create a field

  1. Go to Agents → [agent] → Post-processing → Custom call outcomes → Add.
  2. Name the field (use snake_case, e.g., lead_conversion).
  3. Pick a Field type.
  4. Write clear Instructions (see templates).
  5. Save. New calls will be analyzed for this field.
Call Analysis Edit Self Serve Field types
  • Boolean (True/False): yes/no or binary states
  • Text: short free-form text
  • Number: integer or decimal
  • Category: one option from a list you define (use labels your team understands)

​
Plan before you add fields

  • What decision will this field drive? (reporting, routing, QA, follow-ups)
  • Which type is simplest for that decision?
  • For Category, define a small, clear option set.
  • Keep the Instructions unambiguous and focused on what to extract.
  • Confirm downstream systems handle new fields and null safely.

​
Instruction templates (copy/paste)

  • Boolean — lead conversion Set to true only if the customer clearly agreed to be contacted again (explicitly or by accepting next steps). Otherwise set to false.
  • Category — lost reason Select the single best explanation for why the customer is not proceeding.
    • Choose NO_INTEREST if the customer is no longer interested in the product.
    • Choose PREVIOUSLY_CONTACTED if the customer already been contacted about this topic
    • Choose TOO_EXPENSIVE if the customer expresses reluctance about the price of the product
  • Text — follow-up timing Return the agreed follow-up time concisely (e.g., Thursday 09:00).
  • Number — quantity requested Return the total number of units requested.
  • Text — insurance provider (string example) Return the exact insurance provider the customer says they are a member of (e.g., Insurance A, Insurance B, Insurance C).
Best practices
  • One decision per field.
  • Prefer Category over Text when you’ll filter or chart results.
  • Include brief examples in the instructions if edge cases exist.
  • Remember that the analysis is solely based on the call transcript.

​
Data you’ll receive

Each analysis field follows a consistent data structure: 
{
  "call_outcome": {
    "lead_conversion": {
      "value": true,
      "reason": "Customer confirmed they want to be contacted."
    },
    "lost_reason": {
      "value": "NO_INTEREST",
      "reason": "Customer explicitly mentioned that they are not planning to buy right now."
    },
    "insurance_provider": {
      "value": "Insurance A",
      "reason": "Customer stated their insurer explicitly."
    }
  }
}
Notes:
  • Reason appears only if enabled for that field.
  • Fields may be null when the transcript doesn’t clearly support a value.