Health
Overview
The Health section in BotStudio helps monitor the chatbot’s performance, detect errors, and suggest improvements. Regularly reviewing this section ensures the chatbot functions optimally and maintains high-quality user interactions.
The Health section is located in the Analyze & Debug section of the main navigation in BotStudio.
Health Features
The Health section provides five key types of checks:
- Improvements – Recommendations for enhancing chatbot performance.
 - Errors – Identifies and highlights potential issues that could affect chatbot functionality.
 - Hygiene Checks – Ensures chatbot flows and configurations are optimized.
 - Conversation Tests – Validates chatbot responses to ensure consistency.
 - GenAI Evaluation – Evaluates AI-generated responses to fine-tune accuracy.
 
Health Page Controls
At the top right of the Health page, you will find three key buttons:
- Recompute Checks – Refreshes all health assessments.
 - Edit Health Settings – Adjusts thresholds for various checks (e.g., number of draft responses allowed).
 - View Health Descriptions – Provides details on what each health check entails.
 
If the chatbot has multiple variants or supports different languages, a selection menu allows switching between them for targeted analysis.
Diagnostics
The Diagnostics bar displays system-wide issues affecting the chatbot. Depending on the selected tab, this bar provides an overview of key findings.
- Green Bar: No detected errors.
 - Red Bar: Indicates errors that need attention, such as syntax issues or misconfigured authentication.
 
Common Issues Detected:
- Identically named nodes.
 - BotScript syntax errors.
 - Nodes requiring authentication when global authentication is disabled.
 - Missing fallback for authentication failures.
 - Compilation errors preventing chatbot deployment.
 
Screenshot Placeholder: Insert an image of a Diagnostics bar with error details.
Improvements
The Improvements tab highlights areas for chatbot optimization. This includes:
- Flows with Many Transfers – Identifies frequent chatbot-to-agent handovers and suggests improvements.
 - Nodes with Many Fallbacks – Detects nodes where the chatbot frequently fails to understand user input, suggesting enhancements to matching rules and intent classification.
 - Explore Feature – Groups common words into topics, providing insights into new potential chatbot flows.
 
Screenshot Placeholder: Insert an image of the Improvements tab with sample suggestions.
Errors
The Errors tab lists critical issues that can affect chatbot performance. These include:
- Syntax errors preventing deployment.
 - Misconfigured nodes.
 - Authentication-related failures.
 
Errors are categorized based on their severity. Some may prevent the chatbot from functioning, while others may cause minor inconsistencies.
Screenshot Placeholder: Insert an image showing an example error message with troubleshooting details.
Hygiene Checks
Hygiene checks provide recommendations for keeping the chatbot clean and efficient. This includes:
- Ensuring all classifiers are recently trained.
 - Checking for redundant or outdated nodes.
 - Identifying broken links within chatbot flows.
 
Unnecessary checks can be disabled if they are not relevant to the chatbot.
Screenshot Placeholder: Insert an image of the Hygiene Checks section with examples.
Conversation Tests
Conversation tests allow chatbot administrators to validate the chatbot’s responses using pre-recorded interactions. These tests help ensure that chatbot behavior remains consistent after updates.
Two Types of Tests:
- Test Path – Verifies that the chatbot follows the same node sequence in repeated interactions.
 - Test Content – Ensures chatbot responses remain identical when given the same input.
 
Running a Conversation Test:
- Engage in a test conversation with the chatbot.
 - Verify that responses are as expected.
 - Click Use This Conversation as a Test Case.
 - Assign a test name and specify whether to test the path, content, or both.
 - Save and run tests whenever chatbot updates occur.
 
Screenshot Placeholder: Insert an image showing the Conversation Tests section with test creation steps.
Updating Conversation Tests
If chatbot responses change over time, existing tests may need updates. If a test fails, administrators can either:
- Adjust chatbot responses to align with the expected outcome.
 - Update the test case to reflect the chatbot’s latest response behavior.
 
Screenshot Placeholder: Insert an image showing a failed conversation test with an update option.
GenAI Evaluation
The GenAI Evaluation feature helps assess chatbot-generated responses using AI models. This tool is crucial for refining chatbot accuracy and optimizing response quality.
Key Features:
- Generated Reply Quality – Evaluates response correctness based on set thresholds.
 - Articles Presented – Analyzes the quality of suggested content based on user queries.
 - Threshold Adjustments – Fine-tunes AI model performance for optimal results.
 
Running a GenAI Evaluation:
- Activate Generative AI API in the demo panel.
 - Enter a test message and add it to Generative AI Tests.
 - View evaluation results on the GenAI Evaluation page.
 
Screenshot Placeholder: Insert an image of the GenAI Evaluation interface with test results.
Adjusting Thresholds
Thresholds determine whether chatbot responses are classified as:
- Good – High confidence and accurate response.
 - Acceptable – Somewhat relevant but could be improved.
 - Unacceptable – Poor response requiring adjustment.
 - No Reply Sent – Below threshold and not displayed to the user.
 
Administrators can experiment with threshold values to optimize chatbot performance.
Screenshot Placeholder: Insert an image of Threshold settings in GenAI Evaluation.
Best Practices for Maintaining Chatbot Health
- Regularly recompute health checks to catch new issues early.
 - Use conversation tests to validate chatbot performance after updates.
 - Monitor error logs to prevent disruptions.
 - Fine-tune AI thresholds to balance accuracy and relevance.
 - Review hygiene checks periodically to maintain an optimized chatbot structure.