Does Auto-Learning consume credits?

Yes. When Auto-Learning is enabled, the system analyzes conversations and generates improvement suggestions, which consumes additional credits.

What happens if the A/B test determines that B loses?

If version B (with improvements) does not outperform version A (control), the improvements are discarded and your current prompt remains unchanged.

Can I reverse an applied upgrade?

Yes. In the Version History, you can roll back to any previous version, as long as it is not currently in an A/B test.

How long does it take for an improvement to be applied?

The improvements first go through an A/B test that lasts at least 2 days and requires at least 20 interactions on each version. Only if B wins are the improvements applied permanently.

WAzion Self-Learning

What is Auto-Learning?

Auto-Learning is a system that automatically analyzes conversations with your customers and detects patterns to improve AI responses.

The system generates improvement suggestions and tests them through A/B tests before permanently applying them to your prompt.

How it works

1

Conversation analysis

The system analyzes the interactions and measures the “similarity“ between what the AI suggests and what the agent actually sends.

2

Suggestion generation

When it detects that the agent frequently modifies the AI’s suggestions, it generates improvement suggestions with priority HIGH, MEDIUM, or LOW.

3

Automatic A/B Test

When there are 3 or more HIGH suggestions, an A/B test is created. Your current prompt (A) receives 80% of the traffic and your improved prompt (B) receives 20%.

4

Automatic decision

After 2 days and at least 20 interactions per version, the system decides which one is better. If B wins, the improvements are applied permanently.

How to activate Auto-Learning

1 Open the WAzion Dashboard
2 Go to the “Auto-Learning“ section
3 Activate the “Enable Auto-Learning“ switch
4 Press “Save

Having Auto-Learning enabled consumes additional credits for each conversation analyzed.

Key Performance Indicators (KPIs)

Processed Interactions

Total interactions analyzed by the system.

Average Similarity

Percentage of similarity between the AI’s suggestions and what the agent actually sends. The higher, the better the AI performs.

Applied Improvements

Total number of improvements that have passed the A/B test and have been applied to the prompt.

Next Update

Remaining interactions for the next analysis (Micro: 10, Medium: 50, Macro: 500).

Quality Metrics

Average Iterations

How many times the agent edits before sending. The goal is 1.0 (direct sending).

Perfect

Percentage of responses sent on the first attempt without modifications.

Multiple Attempts

Percentage of responses that required more than one attempt.

Pending Suggestions

Suggestions are classified by priority:

HIGH Important improvements that are tested first

MEDIUM Moderate improvements

LOW Minor improvements

When there are 3 or more accumulated HIGH suggestions, the system automatically starts an A/B test.

A/B Tests

How does the A/B system work?

1. When there are 3+ HIGH improvements, an A/B test is created automatically

2. Your current prompt (unchanged) = Version A - receives 80% of the traffic

3. Your prompt + improvements = Version B - receives 20% of the traffic

4. After 2 days and 20 interactions, the system decides which one is better

5. If B wins: the improvements are applied permanently. If A wins: your prompt remains the same

Version History

Each time improvements are made, a new version of the prompt is created. From the history you can:

• View all previous versions with date and number of improvements
• Compare differences between versions (button “Diff“)
• Rollback to a previous version if the improvements do not work as expected

Versions marked as “TESTING“ cannot be reverted until the A/B test is completed.

Available charts

Evolution of Similarity

Line chart showing how the average similarity has evolved over the past 30 days.

Category Distribution

Pie chart showing which categories of conversations are most frequent.

Similarity by Category

Horizontal bar chart showing the average similarity in each category. The categories with the lowest similarity are where the AI needs the most improvement.

Self-learning

What is Auto-Learning?