Every conversation in Convi is automatically tagged with a summary, one or more topics, and a user satisfaction signal. These power the inbox filters and most of the charts under Analyze. This article explains where the signals come from and how to use them.
In this article
What gets tagged automatically
Topics — what they are and where to see them
Filtering by topic in the inbox
Conversation topics in Analyze
Abilities usage in Analyze
CSAT (customer satisfaction over time)
Knowledge gaps and recommendations
Sandbox vs Real shoppers
What gets tagged automatically
When a conversation finishes (or in many cases, while it's still running), Convi attaches several pieces of metadata to it:
Title — a short label like "Return policy question".
Summary — one or two sentences describing what happened.
Topics — one or more topic chips drawn from the conversation's content.
Sentiment and satisfaction signals — derived from the shopper's words and any thumbs-up/down they left.
Abilities used — which tools the assistant called during the chat (Knowledge search, Order tracking, Handover to human, etc.).
Started from — the page URL the chat was opened from.
You see all of this in the conversation's Details pane (right side of the inbox) and aggregated under Analyze.
Topics — what they are and where to see them
Topics are short labels Convi assigns to a conversation based on its content. The available set is populated from your shop's actual conversations, so you'll see categories like Shipping, Returns, Sizing, Promotions, Orders, Product — whatever your shoppers are actually asking about.
A single conversation can have multiple topics when the shopper jumps between subjects (e.g. Returns and Shipping in one chat).
Filtering by topic in the inbox
In Conversations, open the filter bar and pick Topic. You can select one or many. The list immediately narrows to conversations tagged with at least one of the selected topics.
This is the right tool when you want to answer questions like:
"How many shoppers asked about returns this week?"
"Show me every chat about international shipping."
"Did we ever resolve that promo-code edge case?"
Conversation topics in Analyze
Open Analyze → Overview. One of the cards is Conversation topics — a breakdown of which topics shoppers are asking about, with counts. Hovering or clicking a topic deep-links into the Conversations inbox with that topic pre-filtered ("Click a topic to see its conversations").
This is the highest-leverage chart on the page: if one topic dominates, that's where to invest knowledge or abilities.
Abilities usage in Analyze
Right next to topics is Abilities usage in conversations — how often each enabled ability was actually called. Clicking an ability deep-links into the inbox with that ability pre-filtered ("Click an ability to see its conversations").
Use this to spot:
Abilities you turned on but nobody is using (you might be hiding them, or they're not relevant).
Abilities that fire a lot but resolution rate is low (worth reviewing the prompts and scenarios under Train → Abilities).
CSAT (customer satisfaction over time)
The Customer satisfaction over time chart shows average CSAT (1–5) per day across rated conversations. Subtitle: "Average customer satisfaction rating (1–5) per day across rated conversations."
A few practical notes:
Only conversations the shopper actually rated count. If shoppers don't see the rating prompt (or skip it), they don't move the average.
The chart is sensitive to small shops with few daily ratings — read it as a trend, not as a precise daily KPI.
Negative free-text feedback is surfaced on the conversation's User data card so you can act on specific complaints.
Knowledge gaps and recommendations
In Analyze → Recommendations you'll find Knowledge gaps — "Questions shoppers asked that Convi couldn't confidently answer." Each item shows the inquiry, an impact rating (High / Medium / Low), and one-click actions:
Add answer — turn the gap into a custom FAQ in Train → Content.
Test — open the question in Evaluate → Playground to see how Convi handles it now.
Dismiss — remove the item if it's not relevant.
This is the most actionable analytics view in Convi: it converts unanswered questions directly into knowledge you can publish.
Sandbox vs Real shoppers
Analyze pages have two Environment modes, surfaced by a badge at the top:
Real shoppers — data from your actual storefront conversations.
Sandbox — runs from Playground, Pre-flight, and Simulations.
When you're in Sandbox mode you'll see a banner: "Showing runs from Playground, Pre-flight and Simulations — not real shoppers."
This separation matters for two reasons:
You can stress-test your training (Simulations) without polluting your real metrics.
Until you have enough live conversations, the Sandbox view is the only place with meaningful data.