HelpStackDocs

Guides

Analytics

The Analytics page (at /analytics) gives you a live picture of your support workload and your website's visitors. It has two parts: a conversation overview that summarizes today's inbox activity, and visitor analytics that show who is on your website right now and where they're coming from. This guide explains every metric so you know exactly what each number means.

Sections#

Who can see analytics#

All roles — OWNER, ADMIN, AGENT, and VIEWER — can view analytics. See Team & roles for the full permission breakdown.

Conversation overview#

At the top of the page, the conversation overview summarizes today's inbox at a glance:

  • Open — the number of conversations currently in the open state.
  • Pending — the number of conversations currently pending.
  • Resolved today — how many conversations were resolved today. This card also shows a last-30-days total so you can compare today against the recent trend.
  • New today — how many new conversations came in today, along with the percentage change versus yesterday, so you can see whether volume is rising or falling.

Visitor analytics#

The visitor section shows real-time and recent activity from people browsing your website. It's built from your website chat widget's visitor tracking, so it reflects live traffic rather than historical conversation data.

The time-range selector#

A range selector lets you choose the window the visitor metrics are calculated over:

  • 1h — the last hour
  • 6h — the last six hours
  • 12h — the last twelve hours
  • 24h — the last day
  • 2d — the last two days
  • 7d — the last seven days
  • 30d — the last thirty days

Changing the range updates the stat cards and the activity chart so they all reflect the same window.

Visitor metrics are now retained for up to 30 days, so the longer ranges (2d / 7d / 30d) fill in as history accumulates going forward — a freshly-upgraded inbox won't show 30 days of data until 30 days have passed.

Visitor stat cards#

  • Active now — a live count of visitors currently on your site.
  • Peak traffic — the highest number of simultaneous visitors seen in the selected range.
  • Average active over range — the average number of active visitors across the selected window.
  • Total unique visitors — how many distinct visitors you had over the selected range.

Active visitors list#

Below the stat cards, the active visitors list shows each visitor currently on your site with:

  • IP — the visitor's IP address.
  • Channel — which channel/source they're associated with.
  • Current URL — the page they're looking at right now.
  • Time on site — how long they've been browsing.
  • Conversation link — if the visitor already has a conversation, a click-through takes you straight to it in your inbox.

This is handy for spotting someone who's lingering on your pricing or checkout page — you can see if they've started a conversation and jump in.

Activity over time#

The activity-over-time bar chart shows the peak number of visitors per time bucket across your selected window. Each bar represents a slice of time, and hovering over a bar shows a tooltip with the time and the visitor count. The chart respects the selected time window and shows timestamps, so the buckets line up with whichever range (1h / 6h / 12h / 24h / 2d / 7d / 30d) you've chosen.

Use it to find your busy periods — if traffic peaks at a particular time of day, that's when you'll want agents online.

Top channels, pages, and referrers#

Three ranked lists round out the visitor analytics, each showing the top 5 with visitor counts:

  • Top Channels — which channels are bringing in the most visitors.
  • Top Pages — the most-visited pages on your site.
  • Top Referrers — where your visitors are coming from before they land on your site.

Where the live data comes from#

All of the live visitor data — Active now, the active visitors list, the activity chart, and the top lists — comes from your website chat widget. If you haven't added the widget to your site yet, the visitor analytics won't have anything to show. The conversation overview, by contrast, reflects conversations across all of your channels.

Advanced analytics#

Growth plan and above unlocks a deeper Advanced Analytics view further down the page: response and resolution times, an AI-vs-human breakdown, per-channel and per-agent performance, usage trends, and an AI performance section. On a lower plan, this area shows a locked preview with an upgrade prompt instead of your real numbers.

AI performance

AI performance shows how AI-drafted replies get reviewed, and why they get rejected when they do:

  • AI drafts generated — the total number of replies the AI drafted in the selected time range.
  • Draft outcomes — every draft lands in one of four buckets: approved as-is, edited, then sent (an agent changed the wording before sending it), rejected, or pending review.
  • AI approval rate — the share of decided drafts that were approved or edited-then-sent. An edited-then-sent draft still counts as approved here — the agent chose to send the AI's draft, just with changes. Drafts still awaiting review are excluded from this rate, so it only reflects drafts an agent has actually acted on.
  • AI drafts per day — a daily count of drafts generated, so you can see volume over time.
  • Approval rate over time — the daily approval rate as a trend line, so you can see whether AI quality is improving or slipping.
  • Top rejection reasons — when an agent rejects a draft, they pick a reason (wrong information, wrong tone, too long, missing detail, hallucinated, didn't use tools, or other). This list ranks those reasons by how often they come up, so you know what to fix first.

Advanced Analytics has its own range selector (7 / 30 / 90 days) — separate from the visitor-analytics range above — and AI performance always follows it, along with the same Growth plan and above gating as the rest of Advanced Analytics.

Ask about your numbers

Advanced Analytics has an Ask box: a question in ordinary words, answered from your own data.

It reads only aggregate figures, never the text of any conversation, and it can only see this organization's numbers. Every figure it gives comes from the same queries that draw the charts above it.

The important part is what it does when it cannot answer. Ask it whether customers are more frustrated than last year and it will tell you that tag-based sentiment only starts on the day tagging was switched on for you, rather than inventing a trend. It is built to refuse rather than guess, because a wrong answer here would read exactly like a right one.

Good questions to start with:

  • Which tags are growing this week?
  • What is driving our slowest resolutions?
  • Which channel got busier recently?
  • What did the AI get wrong most often last month?
  • How many people asked about refunds?

Questions about what customers wrote are matched by meaning, not spelling, so a customer writing "money back" is counted by a search for refunds, and so is one writing in another language. It says which method it used; for a conversation not yet indexed it falls back to a literal word match and tells you so.

It compares periods by rate rather than raw totals, so a 30-day window and a 7-day one are put on the same footing and it says which dates each side covers.

Tags

If you use tags, Advanced Analytics reports on them:

  • Conversations tagged — what share of conversations carry at least one tag. A low number means the reporting below is drawn from a slice, not from everything.
  • Tags over time — a daily line per tag, all on one scale. This is the section worth checking after a release: a tag that steps up is usually a real event — a firmware push, a delivery problem, a page that broke.
  • Per-tag table — how many conversations carry each tag, its share of all conversations, and the average time those conversations took to resolve. Shares deliberately do not add up to 100%: a conversation can carry several topics at once.
  • Removed — how often your team takes a tag off after the AI applied it. This is the number that tells you a tag's description is wrong. If agents strip refund request from four in ten conversations, the AI is reading that description differently than you meant it, and rewriting it in Settings → Tags fixes more than correcting the tags by hand ever will.
  • Category breakdowns — tags grouped by your own categories, so a Sentiment split and a Product-area split are read separately rather than as forty bars in one list.

A dash means not measured rather than zero: no resolution time until conversations with that tag are closed, and no removal rate until the AI has applied the tag at least once.

Related#

Edit this page on GitHub