Conversation Intelligence overview

This feature is only available to paid customers as an add-on. To upgrade, contact your Customer Success Manager or sales@heap.io

Overview 

Conversation Intelligence brings AI-analyzed phone, chat, and email interactions you’re having with customers directly into Heap. This gives Product teams the ability to see customer conversations and customer behavior together in the same place, with no new tool to learn and no warehouse query to write.

This can help you answer questions like:

  • What's driving customer contact, and how can I use that to understand my funnel?
  • What aspects of our recent release are users struggling with?
  • Which product areas generate the most customer friction - and what kind?
  • Where should we invest? Self-service, training, or product fix?
  • What's the impact of product issues on customer adoption, usage, and retention?

How it works

Once Conversation Intelligence is enabled on your account, you will see a new event type called Conversation in your Heap environment. 

Each Conversation event includes conversation attributes derived from AI models that describe the support interaction, such as the Contact Driver,  Root Cause, Sentiment, and other conversation-level signals. See Conversation event schema below for the full list. 

We support conversations from the following CRMs: Zendesk, Intercom (includes Fin), Salesforce Service Cloud, and Ada.

Setup

Contact your CSM to purchase the Conversational Intelligence add-on and set it up. 
 

Conversation event schema

Identifiers & Source

Property Description Example
Conversation ID Unique identifier for the conversation (Loris DB ID)
Source System CRM platform the conversation was imported from Zendesk, Intercom, Five9
Source Ticket ID Ticket ID in the original source system
Conversation Type High-level conversation channel grouping Email, Chat, Voice
Channel Specific communication channel WhatsApp, Facebook, Instagram, SMS, Chat, Email, Voice, Phone
Conversation Created Timestamp when conversation/ticket started in source system 2026-02-14T09:32:00Z

Timing & Effort Metrics

Property Description Notes
Handle Time Total handle time in seconds Channel-specific
First Response Time Seconds between conversation creation and first agent response
Average Response Time Average response time across all agent messages Chat-only
Number of Exchanges Back-and-forth exchanges between customer and agent
Number of Customer Messages Count of messages from the customer
Number of Agent Messages Total count of agent messages
Number of Bot Messages Count of messages sent by bot agents
Number of Human Agent Messages Number of human agent messages
Bot Only Is number of human agent messages is 0

Contact Driver Classification (3 Levels)

Property Level Description Example
Contact Driver Area of Business L1 Broadest category of business area driving support contacts  Subscription Management, Order Issues, Security, Reservations
Contact Driver Category L2 Mid-level grouping of related contact reasons within the business area Subscription Cancellation, Refunds. Identity Theft incident, Manage reservation
Contact Driver L3 Primary reason for contact support; what they were experiencing or needed help with, from the customer's point of view Cancel Subscription – Refund Request, Data leak, Change flight
Topics   Separated list of topics covered in the conversation Cancel subscription, cancel charges, refund on the same conversations, Upgrade seat

Note: Contact Drivers are dynamic  - names can be changed and assignments to conversations can be updated over time; Topics not always assigned.

Quality & Sentiment

Property Description
Conversation Quality Score (CQ) Numeric quality score of the solving agent’s interaction - 1-5 star scale.
First Sentiment Sentiment score of the first customer message. A 5-point scale from −2 (most negative) to +2 (most positive), where 0 is neutral.
Last Sentiment Sentiment score of the last customer message. A 5-point scale from −2 (most negative) to +2 (most positive), where 0 is neutral.
Sentiment Delta Change from first to last — negative = customer left angrier; positive = process improved their experience

Resolution

Property Description Values
Resolution Status Whether the issue was resolved Resolved, Not Resolved, Cannot Determine
Resolution Reasoning Generate resolution reasoning

Intelligence & Classification

Property Description Example
Root Cause The company-side diagnosis of what caused the issue - how the business owns and categorizes the underlying operational problem Alert and Notification issues management, Identity Theft and Fraudulent Activity Concerns
Root Cause Class Root cause grouped into a standardized category
Frequently Asked Question Class Customer questions classified into FAQ categories How can I freeze my credit report and ensure all my credit is securely frozen?
Primary Bucket Product or Policy Opportunity classification
Topics Topics covered in the conversation (|| separated) Wrong address, Discount request
Product Lines Product-line mentions detected (|| separated) Identity Guard, Product - Parental control
Competitors Competitor mentions detected (|| separated) T-Mobile, AmEx, Verizon, Trustworthy
Agent actions || separated list of agent actions Empathy, Reassurance, Recognize the issue raised, verification, Policy explanation

Markers & Tags

Property Description
Reasons to Review Conversation markers flagging notable events: Escalation, Profanity, etc.
Conversation Markers All conversation-level markers (|| separated)
Platform Tags Platform-specific tags from the source system (|| separated)

Example

Let's say you want to know which Contact Drivers are delivering the worst customer experience and need attention from leadership. 

Create a new Usage over time chart. On the Conversation event, add an Event Filter: Conversation Quality Score < 3 AND/OR Last Sentiment < 0. Group by Contact Driver, and then rank by average score.

Save your chart and click Subscribe to send this information to stakeholders at a regular cadence. 

 

 




 

Last updated
Powered by Zendesk