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

# Understand analytics and activity

> Move from workspace trends to the conversation, run, or call that produced them

Dubot separates the workspace-level Overview from the activity attached to each product object.
Use Overview to find a pattern, then open the relevant Wizard, Guidance theme, Resource Center
conversation, action, or capability to diagnose it.

## Start with Overview

Open **Overview** and choose 7, 30, or 90 days. Depending on the features enabled and available
data, the page can show:

* sessions, median turns, resolution rate, and unique users;
* action success, errors, latency, and action health;
* sessions over time and common themes;
* the agent-surface funnel;
* Guidance performance; and
* token use and estimated spend.

Overview currently reports Production activity. A cached or unavailable panel is not the same as
zero activity.

<Frame caption="Start with workspace trends, then open the supporting session, run, or action record.">
  <img src="https://mintcdn.com/dubot/g8qfTI_1l8Lbh02C/assets/product/analytics-overview.png?fit=max&auto=format&n=g8qfTI_1l8Lbh02C&q=85&s=8bde7e3210aa686597757c54f47ce362" alt="The workspace Overview with session, resolution, action, user, theme, and cost trends" width="1440" height="900" data-path="assets/product/analytics-overview.png" />
</Frame>

## Open the supporting activity

| Start here                      | Inspect next                                                          | Use it to answer                                                        |
| ------------------------------- | --------------------------------------------------------------------- | ----------------------------------------------------------------------- |
| **Wizard → Activity**           | Conversation and transcript                                           | What the customer asked, what the agent used, and how the session ended |
| **Guidance → Analytics**        | Theme, session, and run timeline                                      | Which journeys complete, stop, error, or go quiet, and at which step    |
| **Resource Center → Analytics** | Opens, viewers, card clicks, searches, chats, and conversation detail | Whether people open the panel and which help path they use              |
| **Action → Activity**           | Calls, outcome, user, agent, duration, and failure detail             | Whether a specific operation is reliable and available                  |
| **Capability → Activity**       | Calls from the actions in that capability                             | Whether the attached bundle is being used successfully                  |

For an action, open **Activity** to compare its success rate, latency, and recent calls. Filter the
list before opening the failed or slow record that needs investigation.

<Frame caption="Use outcome and latency filters to find the specific calls that need investigation.">
  <img src="https://mintcdn.com/dubot/g8qfTI_1l8Lbh02C/assets/product/action-activity.png?fit=max&auto=format&n=g8qfTI_1l8Lbh02C&q=85&s=dcbc6b7e7e8c134d0baec708fe98ec2a" alt="Action Activity with success, latency, call volume, outcome, user, and agent details" width="1440" height="900" data-path="assets/product/action-activity.png" />
</Frame>

## Diagnose before changing configuration

<Steps>
  <Step title="Identify the pattern">
    Use the date range, theme ranking, action-health table, or feature metrics to find a repeated
    outcome. Do not optimize around a single unusual session.
  </Step>

  <Step title="Open the underlying record">
    Read the full transcript, run timeline, or action call. Confirm the user, agent, page, duration,
    and result belong to the behavior you are investigating.
  </Step>

  <Step title="Find the responsible version">
    Check whether the activity came from the live published configuration or a later draft tested
    by an operator. Draft testing is not customer adoption.
  </Step>

  <Step title="Change one dependency">
    Update the smallest responsible object, test it, and publish deliberately. Revisit the same
    activity view after enough comparable sessions have accumulated.
  </Step>
</Steps>

## Interpret the evidence carefully

* A rendered surface is not the same as a started session.
* A tool request is not the same as a successful customer-side operation.
* A completed assistant turn is not necessarily a resolved customer outcome.
* A draft tested by an operator is not customer adoption.
* Per-theme Guidance counts can overlap because one run may carry several themes.
* Cached or unavailable analytics panels should not be treated as zero.

Guidance outcomes include completed, stopped, error, abandoned, and active. Review the run timeline
before changing a guide based only on its aggregate completion rate.

<Note>
  Resource Center Analytics does not retain the customer's search or question text as an aggregate
  metric. Ask conversation detail is available separately to authorized workspace users.
</Note>

<Warning>
  Product activity is operational evidence, not a contractual security or compliance audit log.
  Confirm retention, export, access, and compliance requirements from the current customer
  agreement and security materials.
</Warning>

See [Preview and publish safely](/operate/publishing) for the version model. When the evidence
identifies a live configuration problem, use
[Recover or roll back configuration](/operate/recovery).