Dashboard

Open in CMS

The Dashboard (Home) is your primary command centre for understanding conversation volume, message depth, and function-tool adoption across any time window you choose. It answers the question: Is my agent being used, and is usage growing?

Open the Dashboard first every morning to catch overnight spikes or silent outages before your users notice them.


Opening the Dashboard

Click the hamburger menu (☰) in the top-left corner to open the sidebar, then click Dashboard in the Home group.

In the sidebar, confirm Dashboard is highlighted before continuing.

The Dashboard loads with five metric cards at the top and two chart panels below.

You should immediately see the KPI cards in the top row, with analytics charts directly underneath.


Metric Cards

The five cards give you an immediate health-check of your agent. They update automatically whenever you change the period selector.

CardWhat it measuresWhat to look for
Total conversationsNumber of distinct conversation threads startedSudden drops may indicate a channel or access issue
Total messagesSum of all messages sent and receivedA stable ratio to conversations suggests predictable usage
Average conversationsConversations per hour / day / month (adapts to period)Growth over time means the agent is finding its audience
Average messagesMessages per hour / day / monthA rising average without more conversations can mean longer exchanges
Messages per conversationMean depth of each conversationIdeal depth varies by use case; a rising number can mean the agent is not resolving requests efficiently in one reply

If any card appears unusually low or flat compared to recent periods, treat it as an early warning and drill down in the charts below.

Reading the Averages

The average cards adapt their denominator to the selected period. In a 24-hour view, averages are per hour. In a 30-day view, averages are per day. In a 12-month view, averages are per month. This makes them directly comparable regardless of how long the selected window is.

A rising Messages per conversation over multiple weeks is one of the earliest signals that something in the agent is not resolving user needs cleanly. Common causes include:

  • Knowledge resources that answer part of a question but leave follow-ups unanswered
  • Mission scope that is too narrow, causing the agent to deflect edge cases
  • Interaction style that asks too many clarifying questions before answering

Period Selector

Click any of the four period options β€” 24 hours, 72 hours, 30 days, or 12 months β€” to change the time window for all charts and cards simultaneously.

The period controls are located near the top of the Dashboard and update every panel in place.

When to use each period:

PeriodBest use
24 hoursDaily morning health check. Catches overnight spikes or outages immediately.
72 hoursAfter deploying a configuration change. Shows the before/after window in a single view.
30 daysWeekly reviews. Reveals weekly patterns and sustained trend changes.
12 monthsMonthly business reporting and year-over-year comparisons.

Changing the period does not navigate away from the Dashboard β€” all panels update in place. Use this fluently to zoom in when something looks unusual.


Conversation Chart

The Conversations and messages chart shows two bars, Conversations and Messages, for each time bucket in the selected period. Both use one scale, so the conversation bars are shorter. A hint above the chart names the busiest bucket.

Read each pair left-to-right to compare thread starts versus message volume in the same bucket.

Patterns to Look For

PatternLikely explanation
Spike followed by a sharp dropViral mention or a campaign that did not convert to regular users
Consistent low volume at specific hoursRegion-specific audience; note the window for maintenance scheduling
Messages growing faster than conversationsUsers are writing more per session β€” possible confusion or incomplete answers
Both metrics flat for several weeksDiscovery problem; users do not know the agent exists or cannot access it
Conversations high but messages lowUsers open the chat and leave immediately β€” check the opening greeting

Bar Drilldown

Click any bar to open a section beneath the charts, titled with that bar (for example 29 Oct). It shows the totals for that bucket, the actions used in it, and the conversations that started in it. The conversation list loads as soon as you select the bar. Click any row to open the full conversation. Close the section with the X button.

To narrow the list, open Filters above it. Choose a part of the bucket (for example 0 to 20 min of an hour, or 06:00 to 12:00 of a day) or switch between With messages and All conversations.

Note: opening a conversation from here needs the Conversation history permission on your contributor record β€” it is on by default, and an Owner or Admin can switch it off per person. Without it the thread rows still list and still open, but the conversation itself is replaced by a warning. See Managing Agent Access.

Tip: The list shows only conversations with messages by default. This gives you a cleaner signal for engagement rate. Choose All conversations in Filters to include sessions where the user opened the chat but never typed anything.


Function-Tool Usage

Next to the conversation chart, the Actions used panel shows two things:

  1. Donut chart and ranked list β€” How often each function tool was invoked across all conversations in the selected period, with Total actions in the middle. The list shows the six most used actions; Show N more lists the rest.
  2. Action detail β€” Select an action to open a section below the charts with its calls, its share of all actions, its busiest bucket and a bar chart of its calls over time.

Why Tool Adoption Matters

The AI model decides which function tools to invoke based on their configured Name and Description. If a tool is correctly configured but ranks low in the Actions used panel, the description is probably too vague for the model to recognise when it applies.

Common tool adoption issues:

SymptomLikely causeFix
A tool with a tiny share despite frequent relevant queriesDescription too genericRewrite the description with specific trigger phrases and example scenarios
A tool dominating the chart unexpectedlyDescription too broadNarrow the description to only the scenarios where this tool is the right choice
A tool completely absent from the chartTool disabled or configuration errorCheck Custom API or plugin settings for the tool
Two tools sharing roughly equal usage for the same scenarioOverlapping descriptionsDifferentiate the descriptions to remove ambiguity for the AI

Clicking Through

Select any action to see that tool's calls over time. Close the detail with the X button. Use the same period selector as the rest of the Dashboard to compare tool adoption across time windows. A tool whose usage is trending downward after a description change may mean the new description is less effective β€” revert or iterate.


Exporting the Dashboard

Click Download PDF at the top-right of the Dashboard, next to the period control, to export the current period including all charts and metrics. The button is off while data loads, when the period has no data, or when the statistics failed to load.

When to export:

  • Before a configuration change β€” baseline snapshot to compare against after the change
  • After a configuration change (72 h later) β€” confirm the change had the intended effect
  • Weekly stakeholder review β€” attach to a report as objective evidence of trend direction
  • Monthly business reporting β€” year-over-year archive showing growth

Best practice: Name your exports consistently, for example dashboard-YYYY-MM-DD-30d.pdf, and store them in a shared folder so any team member can find the historical baseline.