Introduction

The Conversations module is where you see every interaction between users and your AI agents.

It gives you full visibility into live and past conversations across chat and voice channels, along with insights like intent, sentiment, location, and fallback events. From here, your team can monitor performance, improve responses, and step in manually when needed using Human Takeover.

What the Conversations Module Is Used For

The Conversations module allows you to:

• View all live and past conversations

• Monitor user intent and sentiment

• Identify where the AI struggled or failed

• Perform Human Takeover in live conversations

• Retrain the AI using real interactions

• Review analytics across channels

This is the control center for managing real-world AI interactions.

Where to Find Conversations

To access conversations:

1. Open your organization dashboard

2. Click on Conversations

This view includes conversations from:

• Website chat

• Website voice

• Shopify storefront

• VoIP (toll-free calls)

• Playground testing

All conversations are tracked in one place.

Conversation List View

Each conversation in the list shows key details at a glance, including:

• Date and time

• Channel (chat, voice, Shopify, VoIP, Playground)

• User location

• Detected intent

• Sentiment

• Topics covered

• AI fallback label (if triggered)

This helps you quickly identify conversations that need attention.

Opening a Conversation

Clicking on a conversation opens the full transcript.

Inside a conversation, you can:

• Read the entire interaction

• See AI and user messages

• Review voice transcripts for calls

• Identify where the AI performed well or struggled

This view is useful for audits, reviews, and training improvements.

Understanding Live vs Past Conversations

Live Conversations

Live conversations are currently active and happening in real time.

For live conversations, authorized users can:

• Monitor the interaction as it happens

• Take over the conversation manually

• Assist the user directly

Past Conversations

Past conversations are completed sessions.

For past conversations, you can:

• Review transcripts

• Analyze intent and sentiment

• Identify fallbacks

• Retrain the AI for future improvements

Human Takeover

Human Takeover allows a human agent to step in and handle a conversation when needed.

This is useful when:

• The AI cannot answer a complex question

• A sensitive issue requires human judgment

• The user explicitly requests human support

How Human Takeover Works

During a live conversation:

1. Open the live conversation

2. Click Human Takeover

3. The AI stops responding

4. A human agent continues the conversation manually

The user experiences a seamless transition from AI to human.

After Human Takeover

Once a human takes over:

• The AI remains paused for that conversation

• All messages are still logged

• The session is tracked like any other conversation

After the session ends, you can review and retrain the AI if needed.

AI Fallbacks

An AI fallback occurs when the agent is unable to respond confidently.

Fallbacks usually indicate:

• Missing or insufficient training data

• Queries outside the agent’s defined scope

• Ambiguous user input

Fallbacks are clearly labeled in the conversation view.

Using Fallbacks to Improve the AI

Fallbacks are valuable learning signals.

From a fallback conversation, you can:

• Identify what information is missing

• Add new training via Text, Website, or Documents

• Use Instant Retrain to correct responses immediately

Over time, this significantly improves agent accuracy.

Instant Retrain from Conversations

You can retrain the AI directly from conversations.

Instant Retrain allows you to:

• Correct incorrect AI responses

• Add missing context

• Improve future responses instantly

Corrections are saved as Text training and applied immediately.

Conversation Analytics & Insights

Each conversation includes insights such as:

• Detected intent

• User sentiment

• Topics discussed

• Channel source

These insights help you:

• Understand user needs

• Identify common questions

• Improve AI performance and coverage

Voice & VoIP Conversations

Voice and VoIP conversations appear alongside chat conversations.

They include:

• Full transcripts

• Intent and sentiment analysis

• Fallback labels

• Analytics and insights

This allows consistent monitoring across all channels.

Best Practices for Managing Conversations

• Monitor conversations regularly

• Act quickly on repeated fallbacks

• Use Human Takeover for sensitive cases

• Retrain the AI using real conversations

• Review analytics to identify trends

Active conversation management leads to better outcomes.

Common Mistakes to Avoid

• Ignoring fallback conversations

• Overusing Human Takeover instead of improving training

• Not reviewing conversation insights

• Forgetting to retrain after repeated issues

The goal is to reduce human intervention over time while improving AI quality.

When to Use Human Takeover

Use Human Takeover when:

• A conversation becomes complex or sensitive

• The AI lacks sufficient context

• A user explicitly asks for a human

Use AI training improvements to reduce repeated takeovers.

What’s Next?

Once conversations are running smoothly, you can focus on scaling and optimization.

Next User Guides:

• Insights & Analytics

• AI Actions & Automation