Skip to main content

Post-call analytics

Post-call analytics is a section of the Quality & integrations card in the agent settings. The Post-call analytics row shows its status — On · N fields or Off; a click on the row opens the Quality & integrations window on this section. The navigation on the left switches between the window’s sections: next to it are agent objectives, observers and webhooks. How changes are saved. Analytics is edited in a copy: the window’s Save writes it to the draft together with the edits of the other sections, and closing the window with unsaved edits asks Discard changes?.

Post-Conversation Analytics

Post-call analytics runs offline after a conversation ends. It can analyze the full transcript and produce a summary, sentiment signals, custom fields, and outcome evaluations without delaying the call. The main switch, Enable post-conversation analytics, is in the section header. While it is off, the section shows only a hint. While it is on, the section is four cards: What to compute, Custom extraction fields, Models & deep tier and Content threshold. The Post-call analytics window in Quality & integrations: what to compute, the success criterion and custom extraction fields

What to compute

  • Call summary — a short account of the conversation.
  • Sentiment — the overall sentiment of the call. Under it, two checkboxes:
    • Caller sentiment separately — a separate verdict for the customer, alongside the overall one;
    • Sentiment trajectory over the conversation — how the mood moved from turn to turn.
  • Call success evaluation — whether the call met its goal; the Success criterion field under it holds the wording, for example “meeting scheduled / customer agreed”.
  • Language in the card header — the language of the results. Independent of the cabinet and agent languages; agent language means the agent’s language.

Custom extraction fields

The Custom extraction fields card adds fields of your own, pulled out of the full transcript exactly as the Extract Variables node does. The fields are a table: Field name, Type (a tag tinted by type) and Extraction instruction. A field is edited right in its table row: a click on the row or on the pencil opens the editor in place. + Field in the card header adds an empty editor at the end of the table. The editor holds:
  • Field name;
  • Type — string, number, boolean, enum or list;
  • Extraction instruction — the question the extractor answers from the transcript;
  • Enum options — for the enum type: each option is a pill with ✕, + option or Enter adds one. While there are none, the editor says the extractor has nothing to choose from.
Save field puts the field into the table; it is unavailable while the name is empty or taken, or an enum has no options. Cancel drops the edits, Remove field deletes the field.

Models & deep tier

A diagram at the top shows the path of a call: every call goes through the mass tier; a call that matched a trigger or the sample goes on to the deep tier.
  • Analytics model (mass tier) — the faster, non-reasoning model used for the bulk of the offline analysis. It inherits the system setting by default; Override for this agent sets a model of your own.
  • Deep analytics model (optional) — a stronger model for selected calls. The deep pass also runs offline and does not affect live call latency. Override for this agent sets a deep-tier model of your own; reasoning model (OK offline) — reasoning does no harm here: nobody is waiting on the answer in real time. Without a model here the server deep model is used; with none set at all, the triggers never fire.
The analytics answer on both tiers is capped at 1536 tokens. If there are many fields and the answer does not fit, raise the cap to 2048 with the max_tokens field of the overridden model; a value outside the range is brought to the nearest bound. A cut-off answer does not turn into an empty result: the analytics task ends with an error and its reason. The Deep tier triggers block decides which calls go for a second pass:
  • By sentiment — positive, neutral or negative; each is a chip that turns on with a click.
  • By outcome — Completed or Interrupted, also chips.
  • By extracted fields — conditions on your custom fields: the field is picked from the list, then the operator (==, != or contains) and the value; + condition adds a row, ✕ removes it.
  • Sampling — the share of calls sent to the deep tier regardless: a slider and a number field, from 0 to 100% of calls.

Content threshold

The Content threshold card filters out empty conversations before any model is called:
  • Min user turns — how many times the customer had to speak.
  • Min duration — in seconds; 0 turns the check off.
  • Skip dispositions — which outcomes to leave alone, as chips: Voicemail, Voicemail, no message left, Declined by a call screener, Call screener timed out, IVR gate (buttons), No speech, Too short, Silence (auto-end) and Internal error.
Calls with no one on the line are never analyzed and have no chips: a failed dial, busy, no answer, an inbound call the caller dropped before the agent answered, a cancelled call, a call that got no agent or no audio from the caller. The line Calls with no one on the line are never analysed under the chips is the reminder. Such a call gets a skipped status with its reason in analytics, and call_analyzed carries no outcome for a conversation that never happened. The result of the analysis reaches your endpoint as the call_analyzed event — see Webhooks.