Witivio Documentation

# Analytics Dashboard

The Analytics dashboard is the only place that looks across your whole subscription at once — every agent you have, combined or filtered down to just the ones you pick. Unlike General Statistics, which you open from inside a single agent, Analytics compares performance across time periods and across agents: Are users coming back more? Are they happier than last month? Is one agent carrying most of the volume?

Quick Navigation

Access this from the top-level Admin Center > Analytics entry in the sidebar — it only appears when you're not already inside a specific agent, since it's a subscription-wide view, not a per-agent one. It's marked Preview.

# What You See

Analytics Dashboard - AI agents filter, KPI tiles, and the Users/Messages comparison charts, each broken down by agent (2020, IT Pilot, HR Pilot) against the previous period

# KPI Cards at the Top

Each shows a metric with a comparison to the previous period:

  • Active users: The change up or down (e.g., "4 active users, +100% vs previous period" means you had 2 users last period, now 4)
  • Messages: Total messages sent this period, with % change
  • Satisfaction rate: Your overall feedback score, with trend
  • Sessions: How many conversation threads happened, with trend

How to read it: A green + is good (more engagement, higher satisfaction). A red - means things got worse. Every KPI is compared to the previous period of the same length — a 30-day selection is compared to the 30 days before it, a "Last 12 Months" selection to the 12 months before that.

# All Agents, or Just the Ones You Pick

At the top of the page, the AI agents filter lists every agent in your subscription. This is the control that makes Analytics different from General Statistics:

  • Leave it empty (placeholder: "All agents") and every figure on the page — KPIs, charts, feedback, most-used content — is the combined total across your entire subscription.
  • Select one agent and the page behaves like a single-agent report, scoped to just that one.
  • Select several agents to build an ad-hoc group — for example, two pilots you want to compare side by side, or every agent except a retired one.

Pick your agents, adjust the date range if needed, and click Apply — the whole page reloads scoped to your selection. There's no saved "default group": every visit starts from all agents unless you filter again.

# The Two Comparison Charts

Two cards sit side by side: Users — selected period vs. previous period and Messages — selected period vs. previous period. Both charts work the same way. The selected period is drawn as stacked columns, one bucket per point in time (hour, day, or month — see below), and inside each column, one color segment per agent currently in your filter. A grey line overlaid on top traces the previous period's total, so you see both breakdowns at once: how the total is trending, and which agent is driving it.

  • Filtered to one agent: the column is a single color — this is just that agent's trend against its own previous period.
  • Filtered to several agents (or "All agents" with more than one agent in your subscription): each agent gets its own color in the legend. A column that grows because one color segment grows tells a very different story than one where every segment grows evenly.
  • The grey line staying above the columns means the current period is behind the previous one; below means you're ahead.

Bucket size follows your date range automatically:

Selected range Bucket
Same day / up to 24 hours Hourly
Up to ~2 months Daily
Longer (e.g. the "Last 12 Months" preset) Monthly

What to look for:

  • Is one agent's segment shrinking while the total holds steady? Its users may be migrating to another agent, or it needs attention.
  • Is messaging (segment height) growing in step with users, or are users flat while messages climb? The latter means fewer people are having deeper conversations — not necessarily bad, but worth knowing which it is.

# Feedback Analytics Section

Scoped to whichever agents you've selected above, just like the rest of the page.

# Satisfaction Rate and Total Feedbacks

Both are shown with their own variation badge against the previous period.

Example: "100% satisfaction rate" with "31 total feedbacks" means all 31 pieces of feedback were thumbs-up or neutral, zero thumbs-down.

When to worry:

  • Satisfaction is 100% but feedback count is low (e.g., 5 total feedbacks) → Not enough data yet to be confident
  • Satisfaction dropping while feedback count rising → Users are rating you worse AND engaging more (something broke)
  • Satisfaction high but fewer feedbacks → Users aren't rating your answers; they might not care about quality

# User Feedbacks Donut

Breaks the same feedback down into Positive, Negative, and Neutral (feedback nobody explicitly rated).

Example: An 80/10/10 split = 80% positive, 10% negative, 10% neutral. Near-100% positive with almost no neutral is the strongest signal — it means people are actively rating, not just leaving the default.

# Most-Used Documents and Questions

Below the charts, two ranked lists — Most-used documents and Most-used questions — show your top 20 most-cited pieces of content over the selected period, most-cited first. Because this page can span several agents, each entry names the agent it belongs to right under its title, so a document or FAQ cited from two different agents shows up as two separate rows. Every title is a clickable link that opens the source directly in a new tab (a document, or the FAQ editor); if the agent that owned a cited source has since been deleted, it's labeled "Deleted agent" instead of a name.

Why it matters: when you're comparing several agents, this is the fastest way to see whether they're answering from shared content or from completely separate knowledge bases — and which single piece of content is doing the most work across your whole subscription.

# How to Use This Dashboard

# Week-to-Week Tracking

Check this if you just launched something or made a major change. Did engagement spike? Did satisfaction drop?

Example workflow:

  1. Launch a new knowledge base section on Wednesday
  2. Check Analytics on Friday: Did message volume increase? Satisfaction stable?
  3. Check again next Monday: Is the spike holding or was it one-day traffic?

# Comparing Pilots or Agents

If you're running multiple agents (for example, one pilot per department) and want to see them side by side:

  1. Open the AI agents filter and select the specific agents you want to compare — leave it empty first if you want to see the combined baseline across everyone.
  2. Set the time period (e.g., "Last 30 days") and click Apply.
  3. Read the stacked columns: each selected agent gets its own color segment in the Users and Messages charts — the tallest, most consistent segment is your best performer.
  4. Narrow the filter to just the two or three you're comparing if the legend gets crowded — reapplying with fewer agents makes each one's segment easier to read.
  5. Pick the winner: expand that agent's promotion or content strategy first, apply the lessons to the others.

If overall satisfaction dropped:

  1. Check when it happened: Use the date range filter to narrow down the drop
  2. Check what changed: Did you update a FAQ? Roll out a new agent feature? Policy change?
  3. Fix it: Update the content that broke, then watch Analytics next period to see if satisfaction recovers

# Real-World Example

Scenario: Like in the screenshot above, your subscription has three agents — "2020", "IT Pilot" and "HR Pilot" — and you leave the AI agents filter on All agents. The KPI tiles show the combined total for the last 30 days: 4 active users, 32 messages, 100% satisfaction. The Users and Messages charts break that same total into one color segment per agent, and most of the activity actually landed in a single spike a few weeks back, driven mostly by IT Pilot and HR Pilot together — the combined KPI tiles alone don't tell you that.

What this tells you:

  • The subscription-wide numbers look healthy, but the activity isn't spread evenly across agents or across the period — it's concentrated in a short burst.
  • Without filtering, you can't tell whether "2020" is even being used, or whether it's IT Pilot and HR Pilot doing all the work.

Your next move:

  • Re-filter Analytics to just "2020" to check whether it has any activity of its own, isolated from the two pilots.
  • Re-filter to IT Pilot and HR Pilot together to compare their segments directly and see which one drove the spike.
  • Once you know which agent and which week, cross-check its own General Statistics dashboard for the detail Analytics doesn't show (categories, most asked subjects, action plans).

# Tips

  • Compare against yesterday's data, not last month's. One day tells you if something broke. One month tells you if your strategy works.
  • Link satisfaction to feedback. If 30 people rated the bot and 100% gave thumbs-up, that's strong. If 2 people rated it, that's noise.
  • Watch message volume, not just user count. 100 new users sending 1 message each is signup spam. 100 new users sending 5 messages each is real adoption.
  • Archive your Analytics snapshots. Take a screenshot monthly so you have a record. The dashboard doesn't keep history forever.

← Subscription Management →