Conversation Insights

AI Toolkit • conversation_insights

Summarize the conversation, read how the customer feels and spot the issue

Overview

The Conversation Insights node reads the conversation with AI and tells you what it found: a short summary, how the customer feels, and which of your workspace's issue tags describe what they need. How the customer feels comes as a mood on a five-point scale, the emotion that stands out most (frustrated, angry, confused, anxious, satisfied or grateful), and any risks: a complaint, fraud, an outage, a legal threat, or a sign they may leave. Use it to branch on mood, to put the summary in a message or an API call, or simply to keep a record on the conversation.

How It Works

  1. Read: The node reads the conversation since it was last closed, the whole conversation, or text you write yourself (such as {{user_message}})
  2. Produce: It works out whichever of Summary, Mood and Issue you ticked
  3. Save: With Save to the conversation on, the summary is added to the conversation as a note your team can read, the mood, emotion and risks show on the conversation, and the issue tags are put on it
  4. Route: When Mood is ticked, the flow carries on through the Very positive, Positive, Neutral, Negative or Very negative handle. A run without Mood carries on through the default output. In a flow, so does a mood handle you leave unconnected. In a trigger, an unconnected mood handle ends the run

Issue tags come only from the tags your workspace allows the AI to apply. With none of those, the node finds no issues.

If the AI is unavailable, for example because your workspace has used its AI allowance, the flow carries on through the default output and nothing is saved. When you try the flow in the tester, the outputs are filled in but nothing is saved to the conversation.

Output Variable

The results are stored under the node's ID, which you can change in its properties (for example to insights):

Summary: {{conversation_insights.summary}}
Mood: {{conversation_insights.mood}}
Emotion: {{conversation_insights.emotion}}
Risks: {{conversation_insights.risks}}
Issues: {{conversation_insights.issues}}

Configuration Fields

Produce checkbox_group Required

What to work out. Tick at least one. Mood also reads the emotion and risks, and decides which handle the flow takes.

Show details

Tick any of these:

  • Summary
  • Mood
  • Issue
Read from select Required

What the AI reads.

Show details

Pick one of these:

  • Since the conversation was last closed (default)
  • The whole conversation
  • Custom text
Custom text handlebars

The text to read when Read from is Custom text. Each value you insert, such as {{user_message}}, is read as one line.

Show details

Supports Handlebars syntax:

{{variable_name}}
Save to the conversation toggle

When on, the summary becomes a note, the mood, emotion and risks show on the conversation and the issue tags are put on it. When off, the results are only available to the rest of the flow.

saveToConversation toggle

Save to the conversation

When on, the summary becomes a note, the mood, emotion and risks show on the conversation and the issue tags are put on it. When off, the results are only available to the rest of the flow.

Default: true

Output Variables

This node stores its output in a variable that can be referenced in subsequent nodes using the {{variable}} syntax.

{{variable}}.summary string

A short summary of the conversation (when Summary is ticked)

{{variable}}.mood string

very_negative, negative, neutral, positive or very_positive (when Mood is ticked)

{{variable}}.emotion string

frustrated, angry, confused, anxious, satisfied or grateful, empty when none stands out (when Mood is ticked)

{{variable}}.risks array

Any of complaint, fraud, outage, legal and churn (when Mood is ticked)

{{variable}}.issues array

Names of the issue tags that apply (when Issue is ticked)

{{variable}}.saved boolean

Whether the results were saved to the conversation

{{variable}}.error string

Error message (only present when the conversation could not be read)

Connection Points

Input
Accepts incoming connections from other nodes
Output
Connects to the next node in the flow
Very positive very_positive
The customer seems delighted
Positive positive
The customer seems happy or satisfied
Neutral neutral
The customer seems neither happy nor upset
Negative negative
The customer seems frustrated, disappointed or worried
Very negative very_negative
The customer seems angry or distressed

More in AI Toolkit

Request a demo
This form is running on UniversalChatbot.