When a Copperleaf Kitchen subscriber writes “I've asked about this refund three times now,” Risolva tags the tone, and a rule for frustrated or angry messages hands the chat to Dana R. before the AI replies. Topic tags and a short summary tell her the rest.
Sentiment AnyOf [frustrated, angry]Risolva keeps angry for hostility, like insults, threats or demands to escalate in strong terms. Impatience and “this still isn't working” are frustrated, and confused means the last reply didn't land. Copperleaf Kitchen sends frustrated and angry messages to Dana R., while a subscriber who asks “so if I skip, am I still charged?” stays with the AI, which explains the Wednesday noon cutoff for changing next week's box.
Once Dana takes over Megan R.'s chat, the AI summary in the info panel reads something like “Megan asked for the third time about a refund for her Sept 22 box and says no one has replied.” The summary is rewritten if the chat is reassigned and again when Dana closes it, so the last version says how it ended, such as the $72 refund she confirmed.
You start with ten general topics and add your own, like Skip a week or Box arrived warm, each with an optional description that helps the AI choose. When subscribers raise something that isn't on the list, such as gluten-free swaps, it shows up on the Suggestions tab with a count of conversations, and an admin promotes it to a topic or dismisses it.
Tagging happens in two passes: a quick read of each new message, and a fuller read of the recent thread as it grows. The summary waits until someone on your team gets involved.
While the AI is handling a chat, each incoming message gets a tone and the closest topic from your list. Handoff rules check them before the AI writes back, which is how “Upset customers go to Dana” catches Megan's refund message and holds the AI's reply. A calm follow-up on its own doesn't clear a frustrated or angry tag.
As the AI keeps replying, once a chat passes 8 messages and then after every 5 new ones, Risolva rereads up to the last 40 messages and sets up to five topics, an intent such as Complaint or Sales, a disposition such as Escalated or Resolved, and an overall sentiment. Anything discussed that isn't on your topic list goes to Suggestions instead of being forced into the nearest fit.
Taking over, assigning or closing the chat triggers a one or two sentence summary under AI summary in the conversation info panel, like the closing summary shown here, so Dana can see what Megan wants without scrolling back. On the Routing card, View signals at the time shows the topics, sentiment and intent the chat had when the rule fired.
If a tag is off, Dana clicks the chip and picks the right value, or chooses Dismiss this tag. Corrected tags show a check mark and later passes leave them alone, so the inbox topic filter, segments and lead interests all work from her version.
Names exactly as they appear in the conversation info panel, Settings > Topics, the handoff rule editor and the segment builder.
Live chat sentiment analysis classifies how a person feels in each message, separately from what they're asking. In Risolva it runs on WhatsApp, Instagram, Messenger and web chat: every incoming message gets one of five labels (neutral, positive, confused, frustrated or angry), and the conversation keeps the latest non-neutral one. Handoff rules can read that label the moment the message lands.
Book a demo and bring a few real messages, including one that made your team wince. We'll show the tone and topic each one gets, the rule that would route it, and the summary your team would read on takeover.
