> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hubtalk.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# 18. Post-call analysis

> Structured fields from the conversation, two tiers of analysis and the success criterion.

# Post-call analysis: what we learn after the call

> [!NOTE]
>
> **Module goal:** get structured fields out of the conversation that reports and campaign decisions can be built on.

![The Post-call analytics window: summary, sentiment and success criterion](https://mintcdn.com/hubtalk/UuNkjJ26eB-xpPwv/images/tutorial/tut-35-settings-post-analysis_en.webp?fit=max\&auto=format\&n=UuNkjJ26eB-xpPwv\&q=85\&s=e7aa266da311604db3dae2bd4f281bb4)

The Post-call analytics window (Settings → Quality & integrations): summary, sentiment (including the caller's own) and a success criterion in your own words.

> [accordion] Show the whole panel
>
> ![The Post-call analytics window: summary, sentiment and success criterion](https://mintcdn.com/hubtalk/UuNkjJ26eB-xpPwv/images/tutorial/tut-35-settings-post-analysis_en.webp?fit=max\&auto=format\&n=UuNkjJ26eB-xpPwv\&q=85\&s=e7aa266da311604db3dae2bd4f281bb4)

![Custom extraction fields in the Post-call analytics window: four fields](https://mintcdn.com/hubtalk/UuNkjJ26eB-xpPwv/images/tutorial/tut-36-settings-custom-fields_en.webp?fit=max\&auto=format\&n=UuNkjJ26eB-xpPwv\&q=85\&s=0f6cb6ec1f0a8e4552f2386b567c31f7)

Custom extraction fields — your own fields over the full transcript: visit\_confirmed (boolean), new\_slot (string), refusal\_reason (string), callback\_requested (boolean).

> [accordion] Show the whole panel
>
> ![Custom extraction fields in the Post-call analytics window: four fields](https://mintcdn.com/hubtalk/UuNkjJ26eB-xpPwv/images/tutorial/tut-36-settings-custom-fields_en.webp?fit=max\&auto=format\&n=UuNkjJ26eB-xpPwv\&q=85\&s=0f6cb6ec1f0a8e4552f2386b567c31f7)

The difference from Extract Variables: post-call analysis runs **after the call over the full transcript**, so it sees the whole conversation and does not affect its speed. Field instructions, enum options and the **Success criterion** can use call variables `{{var}}` — "did the customer confirm the visit on `{{appointment_date}}` at `{{appointment_time}}`". An unknown or empty variable is substituted with an empty string.

All the settings are in the [post-call analytics](https://docs.hubtalk.ai/v4/platform/agent-settings/analytics.md#post-conversation-analytics) section of the **Quality & integrations** window; first turn on **Enable post-conversation analytics** in the section header.

## Results language and field examples

**Language** in the header of the **What to compute** card sets the language of the summary, explanations and text fields. It is independent of the cabinet and agent languages; the empty value, **agent language**, means the agent's language.

Tip: choose the language in which your staff and the systems receiving the webhook will read the results.

* **Kazakhstan (`kz`)** — usually Russian (`ru`); Kazakh (`kk`) if reports are read in Kazakh. Example field: `satisfaction` (enum) with the options «высокая», «средняя», «низкая».
* **USA (`us_tcpa`)** — English (`en`). Example field: `satisfaction` (enum) with the options `high`, `medium`, `low` and the instruction "Did the customer confirm the visit on `{{appointment_date}}` at `{{appointment_time}}`?".

Tip: the field name (`visit_confirmed`, `satisfaction`) is the key in reports and the webhook, so it is convenient to keep it in Latin letters in any market. An enum value is one of the options in the list, so write conditions on the field with exactly the same text as the option.

## Two tiers and when the deep one kicks in

The mass tier runs on every call that passed the content threshold — it uses a fast model (**Analytics model (mass tier)**, the system setting by default). The deep tier runs only on **Deep tier triggers**: sentiment (for example negative), outcome (**Completed** or **Interrupted**), a condition on an extracted field (for example `satisfaction == low`, or `satisfaction == низкая` for a Russian-language agent) and **Sampling** — a percentage of calls. With no triggers and zero sampling the deep tier does not run. Without a **Deep analytics model** of your own, the server deep model is used; with none set at all, the triggers never fire. Tip: this way the strong model works on dozens of calls, not tens of thousands.

The **Content threshold** block filters out empty conversations before any model is called: **Min user turns** (1 by default), **Min duration** (0 by default — the check is off) and **Skip dispositions**. By default every disposition except **Silence (auto-end)** is checked: such a call usually contains part of a conversation and is analyzed. Failed dials and calls dropped before the answer are never analyzed and have no chips of their own.

## Success criterion and Goal achieved

The `call_successful` flag is determined by the analytics model when **Call success evaluation** is on and the **Success criterion** is filled in, for example "the visit is confirmed or a new time is booked".

> [!WARNING]
>
> **An empty criterion means no evaluation.** If the **Success criterion** is empty, success is not evaluated even with **Call success evaluation** on: `call_successful` is not recorded, the campaign's **Goal achieved** card counts no contacts, and the **Goal not achieved** condition in the retry rules never fires.

The **Goal achieved** card in the [campaign details](https://docs.hubtalk.ai/v4/platform/campaigns#campaign-details) shows for how many contacts post-call analysis marked the call successful (`call_successful`), and the analysis coverage. The card is shown only when the campaign agent has post-call analytics and **Call success evaluation** on. The same flag is the **Goal not achieved** condition in the campaign's retry rules, see [module 23, Outbound campaign](https://docs.hubtalk.ai/v4/tutorial/campaign.md).

> [!NOTE]
>
> **Tip: why the Success criterion is in words**
>
> `call_successful` is a metric a manager understands without explanation. Word the criterion so that borderline outcomes are decided in advance: if "call me back later" is not a success, say so — then such a contact returns to dialing under the **Goal not achieved** condition.

## Where the results are visible

* **History and call details** — on the **Overview** tab: **Summary** with the summary, sentiment and success badge, **Call results** with your fields, see [Call details](https://docs.hubtalk.ai/v4/platform/campaign-history/call-details/transcript.md#post-call-analytics). In History, calls are filtered by **Call success**.
* **Reports** — the **Call successful** and **Summary** columns and the custom post-analytics fields, see [Report columns](https://docs.hubtalk.ai/v4/platform/campaign-history/report-columns.md).
* **Dashboard** — the **Goal reached** and **Analytics field** widgets, see [Dashboard](https://docs.hubtalk.ai/v4/platform/dashboard.md).
* **Webhook** — the **Analytics ready (call\_analyzed)** event with the `call_analysis` block: `summary`, `sentiment`, `custom_fields` and `call_successful`, see [Webhooks](https://docs.hubtalk.ai/v4/webhooks.md).
* **Campaign** — the **Goal achieved** card.

## Which agent version decides what to extract

The set of fields is taken from **the agent version the call ran on**. If you publish a new version while a call is still in progress, that call is analyzed by its own, earlier version, so editing the fields on the fly is safe. The **Re-run analytics** button in the call details takes the settings from the agent's current draft.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.