Skip to content

Find answers in Analytics

Measure conversations, deployments, forms and content, and drill into the details.

Last updated

Analytics shows how your agents, deployments, forms and content perform. Each view has its own entry in the sidebar, and the tabs at the top of the page switch between them.

The views#

ViewWhat it measuresHeadline numbers
ConversationsHow conversations went, with breakdowns by sentiment, resolution, CSAT, signal temperature, business category, channel, audience, model and countryConversations, Signals, Signal rate, Visitors, Avg messages / conv, Helpful rate, Avg CSAT, Satisfaction
DeploymentsWidget traffic and engagement per deploymentVisitors, Conversations, Hits, Quarantined hits
FormsForm activity by trigger, page, country and deviceSubmitted, Skipped, Subjects, Completion rate
ContentWhat visitors do with your content in the chat widgetViews, Opens, Reads, Link clicks, Searches, Visitors
AI ModelPer-reply latency, tokens, models, tools and finish reasonsTurns, Conversations, Vendor cost, Avg latency, Total tokens
AI UsageRaw usage quantitiesMetered events, Total tokens, Pages crawled, Vectors embedded, Images generated, Paid provider units, Logical storage

Quarantined hits are set aside and left out of every other total. Searches on Content also shows how many searches found nothing, a quick way to spot a missing article. Content counts the chat widget only: to count portal page views, add a Google Analytics measurement ID to the portal (see Create a portal).

Pick a range and scope#

Choose Last 24 hours, Last 7 days, Last 30 days (the default), Last 90 days or a Custom range. Last 24 hours and custom ranges of up to two days show hourly data; longer ranges show daily data. Days are in UTC. Select a day in a chart to see that day hour by hour.

Filter Conversations and AI Model by agent and deployment, and Deployments and Content by deployment. AI Model also filters by model, tool and finish reason, and Forms by form and source.

On Conversations and AI Model, the first filter picks who you're looking at, starting with Visitors · AI deployment. Visitors and your team are reported separately, never averaged together:

ChoiceWhat it covers
Visitors · AI deploymentVisitors on your deployments
Team · all internalEveryone in your workspace
Team · AI AgentYour team's agent conversations
Team · AI ChatYour team's AI Chat conversations

Read the numbers right#

  • Helpful rate comes from visitors' thumbs up and down on replies: direct feedback.
  • Sentiment, resolution, CSAT and business category are AI-inferred from the conversation, and labeled that way. CSAT is on a 1 to 5 scale.
  • Satisfaction is a score from 0 to 100 that combines helpful rate (50%), CSAT (30%) and the share of positive sentiment (20%). When one of them is missing, the others are reweighted.
  • Visitors adds up each UTC day's unique visitors, so someone who comes back on three days counts three times. Read it, and the conversation totals on Deployments and AI Model, as an upper bound.
  • Vendor cost on AI Model and the quantities on AI Usage aren't what you were charged. For charges, see Understand credits, charges and usage.

Drill into conversations#

On Conversations, select Browse conversations to open a list that keeps your current filters. Narrow it further by deployment, agent, type, audience, channel, sentiment, CSAT, outcome, temperature and signal.

Select a conversation's title to read it. Visitor conversations open in Inbox, AI Chat conversations in AI Chat, and other agent conversations under the agent.

Download the data#

Conversations, Browse conversations, AI Model and AI Usage have a Download menu for CSV or Excel (XLSX).

How fresh is it?#

The info icon at the end of the tab row shows Data through: the UTC time data has been processed through. Newer buckets may still change after the next processing run.

In Browse conversations, set All outcomes to Escalated to human or Abandoned to find questions your agent didn't settle, then fill the gap in Library.

Was this helpful?

Still need help?

Ask the community