Analyze is where you measure what Convi is actually doing for your store — how many shoppers chat, how many conversations the AI resolves, what they ask about, which abilities fire, and how happy people are with the answers. This article walks through every chart on the Overview screen and what to do with it.
In this article
Where Analyze lives
Date range and environment
The Overview line chart
Abilities usage in conversations
Conversation topics
The Conversation funnel
Customer satisfaction over time
Message feedback (thumbs up / down)
Customers most often start chats from these pages
Sandbox vs Real shoppers — the data source
A practical reading order
Where Analyze lives
Open Analyze in the sidebar. The page subtitle reads: "What happened, what to fix, and what's working." You'll see two tabs:
Overview — every chart described in this article.
Recommendations — knowledge gaps with one-click Add answer (see Optimizing Your AI: Using the 'Optimize' Tab).
Date range and environment
The page header has two controls that apply to every chart at once:
Date range — Last week / Last 30 days / Last 90 days / All time.
Environment — Real shoppers or Sandbox.
When you switch to Sandbox, every chart shows runs from Playground, Pre-flight, and Simulations instead of real shoppers, and the page surfaces a small banner:
"Sandbox data — Showing runs from Playground, Pre-flight and Simulations — not real shoppers."
This is intentional — while you're still training, the sandbox is the only place with meaningful data. Once shoppers start chatting, switch to Real shoppers for production metrics.
The Overview line chart
The big chart at the top of the page tracks one metric over time, with the previous period rendered behind the current period in a lighter colour so you can compare at a glance.
The metric is selectable — click one of the cards above the chart to switch what's being plotted (Conversations, AI resolution, etc.). Hovering a point shows both the new-period and old-period values for that day.
Use this chart to answer: "are we trending up or down?"
Abilities usage in conversations
A donut chart showing which abilities the assistant called during the window. Each slice is one ability (Products search, Knowledge search, Order tracking, Cart management, Web search, Order cancellation, Edit shipping address, Handover to human, Email capture).
In Real shoppers mode the chart is interactive — "Click an ability to see its conversations." Clicking a legend item navigates you to Conversations with that ability pre-filtered.
Use it to spot:
Abilities you turned on but nobody uses (something is wrong with discovery, or it isn't relevant).
Abilities that dominate the donut (often the right place to invest more training).
Conversation topics
A second donut, side-by-side with abilities, showing what shoppers actually talked about. Topics come from Convi's automatic per-conversation tagging — see Understanding Conversation Topics and Analytics.
Same interaction in Real-shoppers mode: "Click a topic to see its conversations." Clicking a legend item filters the Conversations inbox by that topic.
This is the single most useful chart for "where should I invest in knowledge?" If returns dominates, you should have more FAQs about returns. If shipping dominates, the shipping policy needs to be sharper.
The Conversation funnel
A simple three-stage funnel:
Visitors ➜ Conversations ➜ AI resolution
For each step the chart shows the absolute count, the step % (how many advanced from the previous stage), and a horizontal bar sized proportionally. The subtitle reads exactly: "Visitors ➜ Conversations ➜ AI resolution."
What each stage measures:
Visitors — unique shoppers who saw your storefront in the window.
Conversations — shoppers who opened the widget and started chatting.
AI resolution — conversations the AI handled without human handover.
Use the funnel to diagnose where you're leaking:
Low Visitors → Conversations → the widget isn't being noticed; revisit Configuring Widget Content (predefined inquiries) and Customizing Your Chat Widget's Appearance (launcher position/size).
Low Conversations → AI resolution → too many chats hand over; review Train → Abilities and Train → Content, plus Analyze → Recommendations for knowledge gaps.
Customer satisfaction over time
A line chart of CSAT (average 1–5) per day, across rated conversations. Subtitle:
"Average customer satisfaction rating (1–5) per day across rated conversations."
A few realities to keep in mind:
Only conversations the shopper actually rated count. Shoppers who skip the rating prompt aren't in the average.
On small shops with low rating volume, daily noise is significant — read it as a trend, not a precise number.
Use the negative-feedback reason on each Conversations → User data card to inspect specific complaints (free-text reasons appear in a tooltip next to the satisfaction value).
Message feedback (thumbs up / down)
Below the CSAT chart is a simple two-column card showing Thumbs up vs Thumbs down counts (and percentages) on individual assistant messages, with a vs previous period badge.
This is different from CSAT — which is the shopper's overall conversation rating. Message feedback is per-message: a shopper can thumbs-down one assistant message and still rate the conversation positively overall.
Use thumbs-down volume as a quality signal: a spike means the assistant is messing up specific messages in specific scenarios. Spot-check those conversations in Conversations to see what went wrong.
Customers most often start chats from these pages
A table at the bottom of the page. Three columns:
Page URLs — the storefront page the conversation was opened from.
Conversation started — how many chats started there.
Resolved by AI — how many of those the AI resolved.
This is the highest-leverage targeting table. If the cart page is your top entry point and resolution is low, shoppers are getting stuck right before purchase — typically because of unclear shipping/returns copy. Fix the page, then watch this table.
If there's no data: "No records found for this date range. Try a different date range."
Sandbox vs Real shoppers — the data source
A reminder, because it confuses people:
Source | What's counted |
Real shoppers | Conversations opened from your live storefront by actual visitors. |
Sandbox | Conversations run from Evaluate → Playground, Evaluate → Pre-flight, and Evaluate → Simulations. |
Switching environment doesn't merge the two — you'll see only one set at a time. Pick Real shoppers once you have any real conversation; use Sandbox for pre-launch dry runs and post-launch experimentation that you don't want polluting your KPIs.
A practical reading order
When you open Analyze for the first time each week:
Overview line chart — what's the trend, vs last week?
Conversation funnel — where are people falling off?
Conversation topics — what are they asking about?
CSAT + Thumbs down — is quality drifting?
Top entry pages — where do your problem chats start?
Recommendations tab — what specific gaps should I fix today?
Three minutes of this is enough to know what to do next.