ChatBI provides two session modes: deep analysis and quick query. By default, ChatBI uses the deep analysis mode, which allows you to perform multi-step exploratory analysis and data queries using natural language. It automatically generates data analysis reports with visualizations.
Overview
As a professional data analyst, ChatBI provides the following AI-driven capabilities:
Intelligent question understanding: Automatically interprets your business intent by rewriting questions, standardizing semantics, and breaking down tasks to identify your analysis needs.
Automated data analysis: Intelligently identifies target data tables and automatically generates and runs SQL queries, enabling you to perform complex data analysis without writing code.
Smart visualization: Automatically selects the most appropriate visualization, such as a line, bar, or pie chart, based on the data characteristics and analysis goals to make data trends clear at a glance.
Insight extraction: Automatically analyzes data to extract key conclusions and business insights, generating a professional data analysis report.
In-depth exploration: In deep analysis mode, ChatBI can independently design analysis paths, dynamically adjust its strategy, and self-correct. This mode supports complex scenarios like multi-table joins, root cause analysis, and anomaly detection.
Prerequisites
Before you use ChatBI sessions, complete the following:
Create a serverless resource group in the region where you use ChatBI.
Create a Datasets dataset. If the dataset is based on a data source, ensure network connectivity between the resource group and the data source.
Procedure
Access the ChatBI system.
Log in to Alibaba Cloud, then access the ChatBI intelligent data insights page in your browser. Select the access point that corresponds to the region of your DataWorks resources, such as your Serverless Resource Group and Datasets.
Select the target dataset.
In the left navigation pane, click Chat. In the lower-left corner of the session window, select the target dataset.
Enter your analysis question and send it.
Enter your question in the input box and send it. ChatBI uses deep analysis mode by default. In this mode, your first question starts the required resources (about 1 CU) on the serverless resource group. This startup process takes about 10 seconds.
NoteTo get quick results for simple queries, switch to quick query mode by turning off Deep Analysis at the bottom of the session window. Note that you cannot use both modes in the same session. Create a new session when you switch modes.
View the report and insights.
When the analysis finishes, ChatBI automatically generates a report containing visualizations and data insights.
Session modes
Comparison overview
ChatBI offers two session modes: Deep analysis deep analysis and Quick query quick query. The default mode, deep analysis, is ideal for comprehensive, multi-step analysis or root cause investigation. The following diagram illustrates the workflow for each mode.
The following table describes the key differences between the two modes.
Dimension | Quick query | Deep analysis |
Use case | Quickly retrieve preset metrics or simple data query results. | Complex, multi-step, exploratory data analysis tasks. |
Interaction | Returns data analysis and insights based on a fixed analysis template. | Leverages a large language model to progressively build analysis logic with dynamic code generation, supporting multi-round iteration and context memory. |
Workflow | Fixed workflow: Question understanding → Target table identification → Execution plan and query code generation → Data query execution → Visualization report generation. | Dynamic workflow: The large language model independently designs and orchestrates the analysis path with a self-correction mechanism, ultimately generating a visualization report. |
Complexity | Low: Suitable for single-table queries, basic aggregations, and simple condition filtering. | High: Suitable for multi-table joins, root cause analysis, time series analysis, anomaly detection, and more. |
Execution time | About 1 minute. | About 5 to 10 minutes. |
Examples | "What was the total sales in January 2026?" or "What is the trend of customer count in the East China region in 2025?" | "What is the repurchase rate trend for each product line in 2026? Identify months with abnormal fluctuations and analyze the causes." |
Deep analysis
Deep analysis is the default session mode in ChatBI. In this mode, ChatBI acts as a professional data analyst that independently plans analysis paths, dynamically orchestrates analysis steps, and automatically adjusts its strategy based on intermediate results. It ultimately produces a data analysis report with visualizations and conclusions.
How it works
Unlike a fixed workflow, deep analysis mode uses a large language model to autonomously drive the entire analysis process:
Autonomous analysis planning: After receiving your question, ChatBI independently designs a multi-step analysis plan.
Dynamic code generation and execution: ChatBI progressively generates and executes SQL or Python code based on analysis needs. Each step is dynamically generated based on prior results and may involve multi-table joins, data cleansing, statistical analysis, and other operations.
Self-correction and iteration: After executing code, ChatBI checks the results. If it detects data anomalies, query errors, or incomplete results, it automatically adjusts the analysis strategy and re-executes without manual intervention.
Insight extraction and report generation: After the analysis is complete, ChatBI automatically summarizes results from each step, extracts key conclusions, and generates a data report with visualizations.
Use cases and examples
Deep analysis is suitable for complex analysis tasks that require multi-angle, multi-step exploration, such as:
Root cause analysis: "Sales dropped 15% month-over-month this month. Analyze the main reasons for the decline."
Multi-dimensional comparison: "Compare customer retention rates across regions and product lines, and identify the worst-performing combinations."
Anomaly detection: "Analyze the order volume trend over the past 12 months, identify abnormal fluctuations, and provide possible causes."
Comprehensive report: "Generate a business performance analysis report for Q1 2026."
Quick query
Quick query is a lightweight session mode in ChatBI that follows a fixed analysis workflow to quickly return data query results. It is suitable for scenarios where you have a clear data query need and want to get results in the shortest time.
How it works
Quick query follows a fixed 5-step pipeline, with deterministic processing logic and output at each step:
Question understanding: Parses the semantics of your question and extracts query intent and filter conditions.
Target table identification: Matches the most appropriate data table from the current dataset.
Execution plan generation: Generates one or more SQL queries based on the identified target tables.
Data query execution: Runs the SQL code and retrieves the query results.
Report generation: Generates a report with visualizations and analysis conclusions based on the query results.
Use cases and examples
Quick query is suitable for straightforward analysis needs with a clear objective that can be answered with a single query, such as:
Metric query: "What was the total sales in January 2026?"
Trend analysis: "What is the customer count trend in the East China region over the past 6 months?"
Ranking and distribution: "Top 10 product categories by sales."
Conditional filtering: "List of customers with order amounts greater than 10,000 yuan last month."