Quick BI provides the Q Chat feature. Powered by the Smart Q data assistant and the interactive capabilities of ChatBI, it lets you get instant data results through natural language interaction. This conversational experience makes data analysis accessible to everyone, introducing a new way to consume data. You can preview and select datasets, enter questions directly or use quick queries, engage in multi-turn chat, and view your conversation history in the Conversation list on both PC and mobile clients.
This topic explains how to use Q Chat.
Q Chat is a value-added module that requires a separate purchase.
This feature is currently available only in the China (Hong Kong), Malaysia (Kuala Lumpur), and Singapore regions. Support for other regions is coming soon.
Prerequisites
You have created a dataset and completed the Q Chat configuration.
You have the required permissions for Q Chat resources. For more information, see permission management.
Access Q Chat on PC
On the Quick BI homepage, click Smart Q to open the Q Chat conversation page.
Data sources
You can query all datasets you have permission to access. These datasets are available under All and various analysis subjects.
For information about how to configure permissions for querying datasets, see Manage Q Chat Permissions.
For information about how to configure analysis subjects, see Manage Analysis Subjects.
Datasets marked with a sample tag in the upper-left corner are sample datasets.
If you do not need sample data, you can hide it. For more information, see Show or Hide Sample Datasets.
Preview and select datasets
Select a single dataset
On the Q Chat conversation page, click Select Data. You can see the datasets you have permission to access and uploaded data files. Hover over a target dataset to Preview it or Query it.
Click Preview (①) to view the dataset's field details and a data preview. You can also use quick queries or click Query in the upper-right corner to open the Q Chat chat box.
Field details
Data preview
Hover over a dimension field to view its details.
Quick query
Query
Q Chatuestions about the current dataset.
Click Query (②) to open the conversation interface and Q Chatuestions based on the current dataset.
After selecting a dataset, you can click the
icon next to the dataset to preview its fields, including key metrics and analysis dimensions.
Click the icon again to collapse the field list.
During a conversation, you can preview data and switch datasets.
Preview data
You can click the
icon next to the selected dataset to get a data preview.
Switch datasets
Click the name of the selected dataset to View available data.
Hover over the target dataset and click Query to switch to it.
You can also Preview the target dataset before you Query it.
Select multiple datasets
When your query matches multiple datasets, you can select all relevant ones to get an answer.
This section describes only how to select multiple datasets. For other operations, see Select a single dataset.
Enter a query.
The system displays datasets that may be relevant to your query. You can select the datasets you need.
Select multiple datasets and click Query based on selected data.
View the analysis results.
Q Chatuestions and get answers
Q Chat supports multiple inference and data interpretation methods. This example uses the Qwen3.7-plus large language model for inference and does not select a data interpretation method. For information about other methods, see Inference and interpretation methods.
You can enter a query in the chat box and send it. You can type your query directly, use a quick query, or use voice input.
Type a query directly
For example, enter "What is the proportion of orders for each order level?" and send the query. The system returns the analysis result and displays the analysis process on the right.
If you do not want to see the analysis process, click the
icon on the right to hide it.
After the process is hidden, the result shows only the chart.
You can then click Analysis Complete to expand the analysis process again.
As you type, the system may provide Recommended fields and Recommended questions. For example, if you type "2024", the following recommendations appear. You can select them based on your business needs.
Question clarification
When this feature is enabled, Q Chat requests clarification if your query is ambiguous. For example, it may ask for more information if a time element or metric is missing, or if multiple metrics could apply. This helps ensure a more accurate answer. For more information, see Enable or Disable Question Clarification.
In the question clarification dialog, you can change the time period and analysis metric, and then click Start analysis.
You can also Modify based on plan to rephrase your query.
Alternatively, you can choose to Skip and answer directly.
Favorite a question
You can favorite up to 30 questions.
Click the
icon to the left of a question to Favorite question.
After a question is favorited, you can find it under Quick query.
If no dataset is selected, you can click a favorited question to ask it directly.
Click the
icon again to unfavorite the question.
Copy a question
Click the 
icon to the left of a question to Copy question.
Use a quick query
Click the
icon in the lower-left corner of the query box to open the quick query interface. You can then click a question from your Favorites, Recommended, or Recent list to ask it.
For questions in Favorites, you can select Automatically switch dataset. This option is selected by default. When selected, clicking a favorited question switches to its associated dataset before asking the question. If cleared, the query runs on the currently selected dataset.
Alternatively, click a recommended question above the query box to ask it.
Click the
icon to get a new set of recommended questions.
After each turn in a conversation, you can also select one of the three Recommended Questions below the chart to ask a new question.
Use voice input
After you enable the voice query switch, you can use voice input to Q Chatuestions.
Click the Voice input icon to open the voice conversation interface.
In the voice conversation interface, ask your question.
The system recognizes your speech and converts it to text.
You can click the
icon on the left to Cancel input, or click the 
icon on the right to Switch to text input.Click Send or press Enter to get the analysis result.
View results
While viewing the results, you can use the following features.
Data filtering (①): The query conditions used for the current analysis are displayed in the chart. You can change them.
Switch chart type (②): Change the visualization type to better suit your analysis needs. The available chart types vary depending on the data.
View AI data retrieval process (③): Check whether the retrieved data meets your requirements. You can view both the business logic SQL and the executed SQL.
Cause analysis and trend prediction (④)
Cause analysis
NoteCause analysis is supported for scorecards and line charts that meet the following conditions:
Supported chart types: Scorecards and non-forecast line charts.
The chart contains exactly one dimension.
The dimension is a date type and there is no legend.
To run a cause analysis, the chart must be error-free.
Line chart conditions: One date dimension and one to three measures.
Scorecard conditions: One date dimension and one measure.
Click the
icon in the upper-right corner to configure Attribution report settings.
You can modify the attribution metric, data interpretation approach, and model selection. After you click OK, the system automatically updates the attribution report. For more information, see Metric Insights.
Trend prediction
NoteTrend prediction is supported for line charts and column charts that meet the following conditions:
Supported chart types: Line charts or column charts that contain a date and no legend.
The chart itself is not a prediction result.
The chart must be error-free.
The chart must have at least 12 dimension values.
Line chart conditions: One date dimension and one to three measures.
You can click the
icon in the upper-right corner to share the trend prediction.
Full screen (⑤).
Rename (⑥)
You can rename the result. You can also click the area in the red box to rename it.
Share data (⑦)
You can Copy link to share the result.
Export data (⑧)
You can export the data to a local Excel file. You can also specify a custom Export name.
NoteExport is supported only on PC, not on the mobile client.
Data interpretation (⑨)
NoteHere, you can initiate a custom data interpretation for the query results based on a specified interpretation approach and model.
You can refer to the sample interpretation approaches, enter your own, select a model, and then click Generate interpretation. The system automatically interprets the data for you.
You can Stop interpretation while the result is being generated.
You can Adopt or modify the interpretation result. Once you adopt the result, the system generates a data interpretation report as shown in the following figure.
If you want to change the data interpretation approach or model for a second interpretation, click the
icon next to Data Interpretation to modify the settings and generate a new report.
Rate (⑪)
Click
to like the result or 
to dislike it. When you dislike a result, you can provide feedback.
Copy trace ID (⑫)
Inference and interpretation methods
Methods
For both inference and interpretation, you can choose between a system built-in large language model and a custom model. You can select the appropriate methods to analyze your data based on your business needs.
Examples
The following examples show the results for the query "What is the proportion of orders for each order level?" using different inference and interpretation methods.
Example 1: Use Qwen3.7-plus for inference and no data interpretation.
The system uses the Qwen3.7-plus model to return the inference process and a chart.
Example 2: Use the built-in Qwen3-max model for both inference and data interpretation.
The system uses the Qwen3-max model to return the inference process, a data interpretation result, and a chart.
Example 3: Use the built-in DeepSeek-R1-0528 model for both inference and data interpretation.
The system uses the DeepSeek-R1-0528 model to return the inference process, a data interpretation result, and a chart.
Smart plan mode
The smart plan mode is supported. When enabled, Q Chat can answer complex, multi-step questions and engage in multi-turn chat.
Enable smart plan mode to answer complex, multi-step questions.
For the query "What are the sales data for the East China region in 2024? Please provide sales suggestions based on recent data trends.", the system integrates the data results and provides suggestions.
Enable smart plan mode for multi-turn chat.
First, ask "Which province had the highest sales amount in 2024?".
After you get the result, switch the dataset and then ask, "What is the name of the top-selling product in this province?". The system automatically uses the context from the previous turn and returns a result.
Ask a follow-up question
You can ask a follow-up question to explore the previous result in more detail. The following example demonstrates this process.
You can ask only one follow-up question per original question.
In the conversation interface, click Ask Follow-up to ask a related question, "Which province had the highest sales?".
You can enter a follow-up question based on your needs, for example: What was the top-selling product?
Click the
icon in the lower-right corner of the dialog box or press Enter to send the follow-up question and get the analysis result.
NoteThe system automatically exits follow-up question mode. To ask another follow-up question, click Ask Follow-up again.
Conversation list
You can click the
icon in the upper-left corner to view your historical conversations.
Click the
icon to start a new Q Chat session.
Mobile client experience
Configure the mobile mini program entry
Follow the steps in the figure to open the mobile editing interface.
In the editing interface, configure Q Chat for the mobile mini program.
Demonstration
You can use Q Chat on the mobile client. The operations are similar to those on PC.
Query data with real-time voice chat
The mobile client supports querying data by using real-time voice chat.
Two interaction modes are supported: Hold to Talk and Real-time Conversation. In these modes, the model uses conversational context to formulate the final query and announce the results. You can interrupt the process at any time by stopping the query, speaking over the announcement, or muting the microphone.
Procedure:
Click the
icon at the bottom of the dialog box to enter the voice chat mode.In the voice chat interface, you can ask your question aloud.
Both Real-time conversation and Hold to talk modes are supported.
When you switch to Real-time conversation mode, Q Chat continuously captures your voice input. You can Tap to mute to interrupt the input.
To resume the real-time conversation, Tap to unmute to continue your voice input.
When you switch to Hold to talk mode, you can press and hold the button to record your voice input.
After you finish speaking, Release to send the query.
After the input is complete, Q Chat automatically transcribes your speech to text, analyzes it, and displays the result. For example, for the voice input "Profit distribution for each order level", the result is as follows:
If you have further questions, you can continue the real-time voice chat with Q Chat.
Suggestions for phrasing questions
For better results, add a time constraint to your questions. For example:
Instead of "Sales proportion of each product", try "This year's sales proportion of each product".
To avoid ambiguity when summarizing multiple dimension values, specify whether you want individual or total calculations. Use keywords like "each" or "respectively" for individual results, and "total" or "in total" for a sum. For example:
Instead of "Sales in Zhejiang and Jiangsu" or "Sales from 2020 to 2023", try "Sales in Zhejiang and Jiangsu respectively", "Sales for each year from 2020 to 2023", "The total sales for Zhejiang and Jiangsu", or "Sales for Zhejiang and Jiangsu in total".
For top-N queries across multiple dimension values, specify how to group the results to avoid ambiguity. For example:
Instead of "Which product sold best in Zhejiang and Jiangsu?", try "Top-selling product in Zhejiang", "Top-selling product in Jiangsu", or "Top-selling product in each province".
Instead of "The 3 best-selling months last year in Zhejiang and Jiangsu", try "The 3 best-selling months last year in Zhejiang", "The 3 best-selling months last year in Jiangsu", or "Top 3 order months by sales for each province".
If dimension values might be ambiguous, be explicit in your question. For example:
For the query "Rank products purchased by corporate customers by sales", the term "corporate customers" might be misinterpreted. It is better to specify the dimension and value explicitly: "For customer type 'corporate', rank purchased products by sales".