Skip to main content

Insights

Overview​

Insights automatically extract specific, structured information from your conversation history. They grade 100% of conversations and reveal how your AI Agent performs at scale. Instead of sampling a small set of dialogs, you gain a complete picture to track outcomes, identify issues, and prioritize improvements.

Insights use a large language model (LLM) to interpret each dialog in context and return actionable results. You can use these results to personalize follow-ups, enrich your CRM, and power analytics.

Types of insights​

Insights include two types: system insights and custom insights.

System insights​

System insights are predefined, high-quality metrics available out of the box. They are tuned for different AI Agent use cases (for example, debt collection, support, upsell) and provide a reliable foundation for performance analysis.

  • AI CSAT (customer satisfaction score): A score from 1 to 5 estimating user satisfaction based on the full conversation.
  • Goal achieved: Indicates whether the AI Agent accomplished its task. Possible values: SUCCESS, PARTIAL, FAIL.
  • Sentiment: Indicates how the user feels about the conversation. Possible values: POSITIVE (😊), NEUTRAL (😐), NEGATIVE (😞).

Custom insights​

Custom insights let you define and extract business-specific data. Each custom insight includes:

  • Name: A human-readable title (for example, "Promise to pay date").
  • Description: Clear guidance for the LLM describing the exact information to extract (for example, "The date the user promises to pay the outstanding balance").
  • Type: The expected data type:
    • Boolean: A boolean value (true or false).
    • Datetime: A date and time value (for example, 2025-01-01 12:00:00).
    • String: A string value (for example, Has two unpaid invoices).
    • Integer: An integer value (for example, 123).
    • Number: A number value (for example, 123.45).
    • Email: An email address (for example, john.doe@example.com).
    • Phone number: A phone number (for example, +1234567890).
    • Categorical: A string from a predefined list (for example, willing, neutral, unwilling).
  • Categories: Required when type is categorical; provide a list of allowed values (for example, eager, neutral, reluctant).

Categorical​

This type is similar to boolean but instead of just true or false, it is more nuanced and can be any of the categories you provide.

For example, if you want to extract the user's willingness to pay, you can set the type to categorical and provide the following categories: willing, neutral, unwilling.

Insight processing​

When a conversation expires, Moveo triggers an extraction job that sends the conversation history to the insight extractor powered by an LLM. The extractor analyzes the full context and returns values for both system and custom insights enabled for the AI Agent.

How to create custom insights​

Custom insights are linked to a specific AI Agent. Create them in the AI Agent's settings:

  1. Go to Build → AI Agents.
  2. Select the AI Agent.
  3. Open Review → Logs.
  4. If Insights are disabled, enable the toggle.
  5. Click Create insight.
  6. Enter a name to identify the insight in the Logs table.
  7. Add a description that guides the LLM on what to extract.
  8. Select the data type.
  9. If the type is categorical, add the allowed categories.
  10. Click Create.

How to view insights​

When a conversation expires and Insights are enabled, values appear as a new column in the Logs table. To continue the example from the previous section, we can see how the insight value shows differently for two different conversations. In the first conversation the user owes more than 100,sotheinsightvalueis‘true‘.Inthesecondconversationtheuseroweslessthan100, so the insight value is `true`. In the second conversation the user owes less than 100, so the insight value is false.

Insight columns appear in the AI Agent's Logs table only, not in Analytics.
Last updated on