The NLU (Natural Language Understanding) Engine Analytics dashboard offers a detailed view of how effectively your NLU layer interprets user input and generates responses. It captures intent detection rates, response quality metrics, and system performance data—empowering you to fine-tune your models for more accurate, context-aware conversations and continuously improve user satisfaction.

1. From the left sidebar, click Analytics to expand the menu.
2. Select the NLU Engine tab at the top of the analytics page to load the relevant charts and tables.

A bubble chart illustrates which intents the engine recognizes and their relative frequencies:

• Bubbles: Each bubble corresponds to a distinct intent (for example, book_test_drive, hungry_user).
• Size: The area of the bubble grows with the number of times that intent was detected over the chosen period, making high-volume intents instantly identifiable.
• Labels: Intent names are clearly displayed on or near each bubble to help you correlate patterns without cross-referencing tables.

A bar chart tracks the activation frequency of each intent over time:

• X-Axis: Lists intent names in a consistent order for easy comparison.
• Y-Axis: Quantifies how many times each intent was triggered, enabling you to spot usage spikes or declines at a glance.
• Bars: Vertical bars rise in proportion to intent volume, with tooltips showing exact counts when you hover over them.

Performance metrics for the Generative AI component, each shown with a current percentage and its trend over time:

• Faithfulness: Measures how closely AI responses adhere to source data, ensuring factual integrity.
• Answer Relevancy: Assesses whether the generated answers address the user’s question directly and meaningfully.
• Contextual Precision: Evaluates the AI’s ability to maintain conversation context across multiple turns.
• Hallucination: Tracks the rate at which the AI introduces unsupported or irrelevant content.

A consolidated bar chart displays the average across all four KPIs, giving a quick snapshot of overall generative quality.

A comprehensive log of every processed query, with key details:

• Namespace: Indicates the context or domain (e.g., “sales”, “support”) under which the query was handled.
• Query: Captures the exact user input text for precise auditing.
• Answered: A binary flag (Yes/No) showing whether the engine supplied a valid response.
• Response: Displays the AI’s returned text, letting you verify correctness immediately.

A pie chart summarizes the answered vs. unanswered distribution, so you can quickly identify coverage gaps and prioritize training for unhandled user questions.

• Inferences (Count): Logs the total number of inference calls the NLU engine made within the selected timeframe, helping gauge system load.
• Response Time (ms): Reports the average processing time from user utterance to AI reply, measured in milliseconds, highlighting any latency issues.

• Date Range Selector: Choose from presets like “Last 7 days” or “Past Month,” or define custom start and end dates to focus your analysis.
• Namespace Filter: Drill down to a specific domain or model namespace to isolate performance by use case.
• Answer Status Filter: Toggle between answered, unanswered, or all queries to zero in on gaps in coverage.
• Refresh: Click the refresh icon to reload with the latest available data without leaving the page.
• Download: Export tables and charts as CSV or image files for external reporting and deeper offline analysis.

• Optimize Training Data: Use Intent Insights to pinpoint low-volume intents that may need more labeled examples to improve recognition.
• Monitor for Drift: Watch Intent Trends for any abrupt increases or decreases, which could indicate shifting customer needs or gaps in your model.
• Refine Generative Models: Track GenAI Analytics KPIs regularly to spot rising hallucination rates or dips in relevancy, and then retrain or adjust prompts accordingly.
• Address Coverage Gaps: Review NLU Engine Queries for frequent unanswered questions, then update your intent definitions or knowledge base so the engine can handle those queries next time.