Instant Retrain & Feedback

Introduction

Instant Retrain and Feedback allow you to improve your AI agent using real conversations.

Instead of waiting for manual retraining cycles, Exei lets you correct responses instantly and capture user feedback directly from conversations. These features help your AI learn faster, reduce errors, and improve accuracy over time.

What Feedback Does

Feedback allows users and internal teams to indicate whether an AI response was helpful.

Feedback is collected in the form of:

• Likes

• Dislikes

Feedback helps identify:

• Incorrect or incomplete responses

• Poorly worded answers

• Areas where training needs improvement

Where Feedback Comes From

Feedback can be given:

• During live conversations

• During Playground testing

All feedback is tied to the exact conversation and response where it was provided.

Where to View Feedback

Feedback is visible in:

• Individual conversation views

• Insights (on supported plans)

This allows teams to review feedback in context and take corrective action.

What Instant Retrain Does

Instant Retrain allows admins to immediately correct an AI response and teach the agent how it should respond in the future.

Instead of editing training data separately, you can fix issues directly from real conversations.

How Instant Retrain Works

When you use Instant Retrain:

1. You select the AI response that needs correction

2. You provide the correct or improved response

3. The correction is saved as Text training

4. The AI learns instantly and updates its behavior

Future conversations benefit from this correction immediately.

When to Use Instant Retrain

Use Instant Retrain when:

• The AI gives an incorrect answer

• The response is missing important details

• The tone or wording needs improvement

• A fallback response occurs due to missing information

Instant Retrain is the fastest way to fix real-world issues.

Instant Retrain and Fallbacks

Fallbacks are strong signals for retraining.

When a fallback occurs:

• Review the conversation

• Identify missing knowledge or rules

• Use Instant Retrain to add the correct response

Over time, this significantly reduces fallback frequency.