Insights

Complete reference for the Insights section — every analytics screen in the product, what each metric actually measures, and which role can open it.

The section is hidden from agents. Beyond that the gating is per-tab and is not uniform, so read Who can see what before assuming a colleague sees the same tabs you do.

For the fast path, use the Quick Start.

Section map#

/analytics redirects to /analytics/overview. Every /analytics/* tab is one page switching on the URL, which is why the date range persists as you move between them.

Who can see what#

The Insights group is hidden from agents entirely. Within it:

Note the asymmetry: supervisors do not see most of Insights. They get Call Centre and Agent Analytics, and that is it. Editors are the reverse — everything except those two.

Only /agent-analytics is guarded at the route level. The /analytics/* tabs rely on the sidebar hiding them plus server-side authorisation, so a pasted URL may render a tab your role would not normally offer while the data behind it stays authorised.

The date range#

Enlarge

One Date Range control — two date inputs separated by to — sits at the top of most tabs and is shared between them. It defaults to the last 30 days ending today, the end date cannot be set past today, and there are no preset buttons.

Three tabs behave differently:

Timezone

Chat, call, SMS, and monthly figures are computed in UTC — day boundaries are UTC midnight regardless of where your business is.

Call Centre is the sole exception: it is computed in your organization's timezone, the one set on Setup → Organization. This is correct — a calling floor's "day" is a local day — but it means the Call Centre tab and the Calls Overview tab can attribute the same call to different days.

Exports

There are no exports in Insights. No CSV, no Excel, no PDF, on any tab. The only download-adjacent controls are pagination and the Call Centre transcript-mining button. To get rows out of the product, export leads from Call Centre or a data collection from Audience.

Overview#

Total Widgets ignoring the date range is deliberate but easy to misread — it is a current count, not "widgets created in this period".

Two charts follow: Trends Over Time, an area chart with Sessions, Messages, and Unique Visitors series, and Widget Performance, a horizontal bar chart of Sessions and Messages per production widget. Test widgets are excluded from Widget Performance.

Widget Analytics#

Enlarge

Nothing renders until you pick a widget from the Select Widget grid — a grid of buttons, not a dropdown.

Satisfaction prefers support-message ratings and falls back to the conversation satisfaction rating. N/A means no ratings exist, which is a different statement from a low score.

Four charts: Messages by Day, Top Intents (percentages, from the AI's classification of each conversation), Device Breakdown (donut), and Location Breakdown (by country).

Top Intents is the most actionable panel in the section. It is a ranked list of what people came to ask, and a high-volume intent your knowledge base answers badly is the clearest possible instruction about what to write next.

Calls Overview#

Enlarge

Every call in the range — AI-handled and human-handled, inbound and outbound, across all providers.

Empty state: No Call Activity Yet — Call analytics will appear here once your call widgets start receiving or placing calls.

Charts: Calls Over Time (Total Calls, Completed, Failed), Talk Time (minutes), Call Widget Performance, and donuts for Status Breakdown, Sentiment, and Provider Breakdown.

Sentiment comes from the AI's post-call classification. It is a useful trend and a poor absolute — read the direction of travel, not the exact split.

Call Widget Analytics#

Pick one voice agent from Select Call Widget; each entry is labelled with its direction and id. With none created you get No call widgets yet — create one to see per-widget analytics.

First row of tiles: Total Calls, Completed (with a completion percentage), Failed, Avg Duration (with the total underneath).

Second row: Incoming, Outgoing, Unique Callers, and LLM Tokens — the last carrying {n} tools • {n} barge-ins underneath.

LLM Tokens, tool calls, and barge-ins are the diagnostic row. A high barge-in count means callers are interrupting the agent, which usually means its turns are too long. Tool calls tell you whether the agent is actually using the abilities you gave it.

Then Calls by Day, Talk Time by Day (minutes), Top Intents, and donuts for Status Breakdown and Sentiment. Empty: No calls for this widget in the selected date range.

Call Centre#

Enlarge

The only tab about human calling, and the only one in your organization's timezone. It covers rep calls only — AI calls are excluded.

Two prerequisites: your role must be owner, admin, or supervisor, and the workspace must have call-centre phone numbers. Without numbers the tab errors rather than showing zeros. The maximum range is 366 days.

A caption under the header states the range and timezone explicitly — "{n} day(s), counted in {timezone}. Rep calls only." — which is worth reading before comparing anything here to another tab.

Summary and queue tiles

Wrap-up filed is a data-quality metric disguised as a performance one. Every other figure on this page — dispositions, pipeline, win rate — depends on reps filing wrap-ups. If this is low, distrust the rest of the tab.

A second row appears only when calls were offered to a queue: Service level, Answered, Abandoned, Average wait.

Panels

Speed to lead measures lead creation to first human call, for leads created in the range. Never rung is the number to act on — it is a direct count of leads nobody called.

By rep shows target attainment as {n}% of target, calculated against each rep's daily target multiplied by the number of days in the range. A short range including a weekend will therefore understate attainment.

Pipeline → Open leads by time in status buckets as under a day, 1-3 days, 4-7 days, 1-4 weeks, over a month. It is a snapshot of now, not of the range.

What callers said does not run automatically — it reads transcripts on demand, and reports how many it read.

Empty state: No rep calls in this range. Load failure: Could not load call-centre insights.

SMS#

Covers SMS volume and money. The window is derived from the shared date range and clamped to between 1 and 365 days.

A Pending charges by carrier table follows, with Carrier, Pending, Period sent, Period segments, Period charges, and Lifetime billed, plus an Unattributed (legacy) row for older traffic. Carriers appear as AI Assistant, Twilio, MSG91, Jio, and Stub (test).

Charts: SMS volume over time (Sent, Failed, Charges on a dual axis), Segments by carrier (period), Sends by source (period).

Two banners matter. If SMS is suspended you get SMS sending is suspended with a reason — resolve it in billing before anything else, because sends are being dropped. With no traffic at all you get No SMS activity yet, pointing you at Integrations to configure a provider.

If the charts are empty but the tiles are not, widen the range: No SMS sent in the selected window. Try widening the date range.

Agent Ratings#

All-time, not filtered by the date range.

Average Rating is a mean of means, so an agent with two ratings weighs as heavily as one with two hundred. Read the per-agent list rather than this tile when the counts are uneven.

Three panels: Agent Performance (Daily, Weekly, Monthly resolved counts), Rating Distribution (agents per star band), and Agent Ratings (per-agent stars with a rating count and resolution counts). A separate Agents Without Ratings panel lists the rest.

Ratings come from customers rating resolved support conversations. Empty: No Agent Ratings Yet — Agent ratings will appear here after customers rate resolved queries.

Survey Responses#

Post-chat surveys, over the shared date range. Configure the survey itself from the Survey action on a widget card in Chat.

A channel filter offers All channels, AI-only chats, and Agent-resolved chats.

NPS is the standard calculation — scores of 9 and 10 are promoters, 6 and below are detractors — so it ranges from −100 to +100 and is not a percentage of anything.

The AI vs agent split is the most strategically useful number here: it tells you what share of resolved conversations needed a human, which is the real measure of whether your knowledge base and workflows are working.

Responses are listed 25 at a time with Previous / Next, each row pilled with its CSAT, NPS, and channel, and expandable to All {n} answer(s).

Monthly Insights#

Its own month and year selectors, defaulting to the current month; month boundaries are UTC.

Four cards, each with a month-over-month delta: Total chats, Total calls, Avg CSAT (out of 5), Avg NPS.

Then Chat sentiment breakdown and Call sentiment breakdown pies, and ranked lists for Top chat keywords, Top chat topics, and Top call intents.

Empty states are specific and worth knowing: pies show No sentiment data yet, and the ranked lists show No data yet — needs enriched conversations. The second one means the AI enrichment pass has not run over your conversations, not that you had no conversations.

The emailed version

Separately from this tab, a scheduled job compiles the previous calendar month and sends it out on the 1st of each month at 09:00 (Asia/Kolkata) with the subject Monthly Insights — {organization} — {month}. It can go by email and optionally to Teams or WhatsApp.

It only runs for organizations that have scheduled reports enabled, and the recipients come from that same configuration — see Setup → Reports.

Agent Analytics#

A standalone page at /agent-analytics, for supervisors and owners, and the only Insights page that is guarded at the route.

Heading Agent Analytics — Monitor performance metrics for all agents in your organization. There is a Refresh button and a search box (Search agents by name or email...).

Table columns: Agent, Daily Resolved, Weekly Resolved, Monthly Resolved, Active Chats. Ten rows per page with Previous / Next and a Showing {a} to {b} of {n} results caption. No export.

Empty: No agents found matching your search.

How it differs from the other agent views

There are three separate "agent performance" screens in the product, and they measure different things:

None of them is a superset of the others. An agent who looks idle in Agent Analytics may have spent the day on the phone, which only shows up under Call Centre.

Why these numbers differ from the Dashboard#

Both are correct; they answer different questions.

See Dashboard for the other side of this comparison.

Troubleshooting#

Related#

• Insights quick start
• Dashboard — the summary these pages sit behind
• Call Centre — the floor the Call Centre tab measures
• Chat — widgets and the post-chat survey
• AI Calls — call widgets and call logs
• Support — CRM ticket analytics
• Setup — organization timezone and scheduled reports