Data Visualization¶

The assistant can supplement its textual and tabular answers with interactive charts to help identify trends, distributions, and patterns within the data. Clicking in the action bar at the bottom of a response opens a dedicated side panel where the visualization is rendered.

Activation and Triggering

The availability of charts depends on the chatbot’s configuration. If the visualization feature is enabled in the settings, the assistant evaluates the retrieved data and prepares a chart if it adds analytical value. Users can also explicitly ask for a visualization in their query (e.g., “Show me a pie chart of…”) to ensure a specific graphic is created.

Visualization Type Capabilities

The assistant supports multiple types of visualizations to suit different data structures. Users can let the assistant choose the best format or specify their preference for a specific chart type.

Pie chart showing the distribution of loans by state.¶

Example of a bar chart showing loan amounts by state.¶

Interactivity and Exploration

The chart panel offers several interactive features to enhance data exploration:

• Tooltips. Hovering over any data point, bar, or slice displays a tooltip with the exact values and labels.

• Dynamic Navigation. For time-series or large datasets, charts may include navigation sliders at the bottom, allowing users to zoom in on specific periods or segments.

• Legend Interaction. Clicking on legend items can toggle the visibility of specific data series for better focus.

Interactive line chart showing averages over time with a navigation slider.¶

Example of the navigation slider narrowed down to focus on a specific time period.¶

Data visualization availability

It is important to note that a chart may not be generated for every response. The assistant only provides a visualization when the retrieved data is compatible with a graphical format and when the representation is considered truly helpful for the user’s understanding.

When to use the Deep Query tool¶

Deep Query is intended for complex analytical questions that require multi-step reasoning, such as comparisons, trend analysis, segmentation, or investigations that cannot be reliably answered with a single query. It helps users go beyond simple metrics to understand why something is happening, not just what is happening.

The following examples show the types of questions where Deep Query delivers the most value:

• Identify the top-performing product in terms of both revenue and customer satisfaction. Consider total revenue, total quantity sold, and average customer rating over the past 12 months. Break down the product’s monthly performance to detect any seasonality or sales spikes. Also analyze what types of customers are purchasing it most often (for example, by age and region). Compare its performance with other products in the same category. If possible, summarize key themes from customer reviews to explain why this product might be performing well.

• Analyze the best-performing course on the platform in the last 6 months. Use metrics such as number of enrollments, average completion rate, student ratings, and re-enrollment rates (students who took multiple courses from the same instructor). Identify monthly trends in engagement and completion. Break down performance by course category and difficulty level, and show which student segments (for example, age groups and countries) are most engaged with this course. Compare it to other top-3 courses in the same category and suggest factors driving its success based on reviews and completion behavior.

Deep Query should not be used for simple questions or quick lookups. In the following cases, the standard Ask mode is usually more appropriate:

• The answer can be obtained with a single query or aggregation (for example, totals, averages, or simple filters).

• The question requires no comparisons, segmentation, or time-based reasoning.

• The user is looking for a quick factual answer rather than a detailed analysis.

Using Deep Query in these scenarios may result in longer execution times and unnecessary computational costs without adding analytical value.

Submitting Deep Query Questions¶

The chatbot provides an interface to submit complex analytical questions using the Deep Query feature. The typical workflow is as follows:

1. Submitting the analysis question using Deep Query

   You can type the analytical question into the chatbot input field. To execute a Deep Query analysis, select Deep Query from the drop-down button instead of the default Ask option.

   

   Submitting a question using Deep Query.¶

2. The assistant request clarifications for the analysis

   Upon submission, the question is sent to the LLM. Before performing the analysis, the assistant requests clarifications to ensure the user’s intent, interest, and scope are clearly understood.

   

   The assistant requesting clarifications for a submitted Deep Query question.¶

3. Providing clarifications to the assistant

   Answer the assistant’s questions in the chat to refine the scope and objectives of your Deep Query analysis.

   

   Providing clarifications to refine the Deep Query analysis.¶

4. The assistant returns the Deep Query analysis

   Once clarifications are provided, the assistant executes the Deep Query analysis and returns an answer. This response includes a summary of findings.

   

   The assistant answers with the Deep Query analysis results.¶

5. Generating a Deep Query report from the previous answer

   Above the answer, two buttons are available to generate the Deep Query HTML report from the current analysis:

   • Download HTML Report – Click to download the HTML report to your device.

   • Open in the Report Viewer Panel – Use to display the HTML report in the chatbot’s left panel for quick inspection, without downloading.

   

   Buttons to download or view the Deep Query report from the analysis answer.¶

   While the assistant is generating the report, you can click at any time to interrupt the report generation process.

   Once the report generation is complete, the assistant message bubble containing the Deep Query response is highlighted in green, as shown in the following image.

   

   Assistant message bubble highlighted in green, indicating that the Deep Query report has been successfully generated.¶

   The full report structure and content are described in the Deep Query Report section.

Deep Query Report¶

A Deep Query report translates a multi-step database analysis in a structured, shareable format, for both technical and non-technical readers. It explains what was analyzed, how the analysis was carried out, what evidence supports the results, and what actions to take based on the findings. The sections of a Deep Query report are described below.

1. Introduction

   Purpose:

   The Introduction sets the context and purpose of the analysis, defines what question is being answered, and clarifies the scope and boundaries of the report.

   What you will find here:

   • The background of the analysis and why it was conducted

   • The main objective or question the analysis aimed to answer

   • The scope and limitations of the report (what is included and what is not)

   Key characteristics:

   • Short and concise (typically one paragraph)

   • Written in clear, non-technical language

   • Does not include results, conclusions, or recommendations

   

   This figure shows the Introduction section of a Deep Query report.¶

2. Executive Summary

   Purpose:

   The Executive Summary provides a high-level overview for readers who want to understand the outcome quickly without reading the full report.

   What you will find here:

   • A brief restatement of the analysis objective

   • The most important findings from the analysis

   • The key recommendations derived from those findings

   • One or two main charts or visuals that best illustrate the core results

   Key characteristics:

   • Concise and results-focused

   • Written for decision-makers

   • Can be read independently of the rest of the report

   • Summarizes conclusions, but does not explain detailed steps

   

   This figure shows the Executive Summary section of a Deep Query report.¶

   

   This figure presents an illustrative graph titled “Distribution of Interest Rates for Approved Loans” located in the Executive Summary section.¶

   

   This figure presents an illustrative graph titled “Average and Median Interest Rate by Credit Score Band” located in the Executive Summary section.¶

3. Detailed Analysis

   Purpose:

   This is the core of the report, where the analysis is explained step by step.

   What you will find here:

   • A detailed walkthrough of how the analysis progressed

   • Explanation of intermediate findings and how they led to later insights

   • Interpretation of data trends, patterns, or anomalies

   • Visualizations (charts or tables) shown alongside the specific insights they support

   Key characteristics:

   • The longest and most detailed section

   • Follows the logical flow of the analysis as it was conducted

   • Visuals are embedded where they are discussed, not grouped separately

   • Explains how conclusions were reached, but does not summarize or conclude

   

   This figure shows the Detailed Analysis section of a Deep Query report.¶

   

   This figure shows a geographic analysis in the Detailed Analysis section.¶

4. Methodology

   Purpose:

   The Methodology describes how the analysis was performed (data sources, steps followed, metrics/definitions used, and any constraints), focusing on reproducibility and transparency.

   What you will find here:

   • The overall analytical approach used

   • Data sources that were queried or analyzed

   • Tools, queries, and techniques used during the analysis

   • The sequence of steps followed, from data extraction to final results

   Key characteristics:

   • Focuses on process, not outcomes

   • May include technical elements, but explained at a high level

   • Helps establish transparency and trust in the analysis

   

   This figure shows the Methodology section of a Deep Query report.¶

5. Recommendations

   Purpose:

   The Recommendations section lists the top actionable next steps derived from the findings, prioritizing impact and feasibility and grounding each recommendation in the reported results.

   What you will find here:

   • Up to three clear, actionable recommendations

   • Each recommendation is directly supported by data or findings from the analysis

   • A brief explanation of why each recommendation matters and what impact it may have

   Key characteristics:

   • Practical and specific (not generic advice)

   • Focused on feasibility and business impact

   • Explicitly grounded in evidence presented earlier in the report

   

   This figure shows the Recommendations section of a Deep Query report.¶

6. Conclusion

   Purpose:

   The Conclusion recaps the key insights and their implications, clearly stating the final answer to the original question without introducing new information.

   What you will find here:

   • A recap of the most important insights from the analysis

   • A summary of their implications for the original question or objective

   • A cohesive closing narrative that ties the report together

   Key characteristics:

   • Summarized in a set of concise paragraphs

   • Does not introduce new data, findings, or recommendations

   • Reinforces understanding rather than adding detail

   

   This figure shows the Conclusion section of a Deep Query report.¶

7. Appendix

   Purpose:

   The Appendix provides a transparent, detailed reference of the technical execution behind the analysis.

   What you will find here:

   A high-level summary of the overall Deep Query execution, such as:

   • Total analysis time

   • Number of analysis iterations

   • Total tool calls executed

   • Models used for planning and execution

   • A list of all tool calls executed during the analysis

   • For each tool call:

     • The tool name and unique Tool ID

     • The input parameters and their values

     • The output returned by the tool

     • When applicable, the VQL query executed and a brief explanation of its intent

   Key characteristics:

   • Designed as a reference section, not narrative reading

   • More technical than the rest of the report, but still structured and readable

   • Does not introduce new insights or interpretations

   • Allows advanced users, auditors, or reviewers to understand exactly how results were produced

   

   This figure shows the Appendix section of a Deep Query report.¶

   

   This figure shows the execution details of a database agent tool in the Appendix.¶

   

   This figure shows a chart generated by the database agent tool.¶