Asking Data

This document introduces the main functions of FineChatBI for you to understand and experience. You are advised to follow the document steps first, and then explore FineChatBI on your own.

Version

Experience Path

The following figure shows the workflow of FineChatBI.

Two entries are available to enter FineChatBI.

• Directly suffix the FineBI access URL with webroot/decision/ai/conversation/resource/web/html.

• Enter FineBI and click the Ask Data button, as shown in the following figure.

The following table lists conversation management areas.

Asking Questions About Preloaded Data

Note:

The admin needs to configure preloaded data first. For details, see Preloading Configuration.

You can directly select an analysis subject from the data list above the query box to ask questions. Alternatively, you can click Data Overview to expand the data list, click Preview to preview an analysis subject, and then make a selection.

After clicking Data Overview to expand the data list, you can click Preview to view detailed data within an analysis subject to help check whether this analysis subject contains required data.

Asking Questions About a Custom Subject

If the required analysis subject is not in the preloaded data list, you can also manually select it. But you can manually select only those analysis subjects on which you have editing permissions.

Subject/Field Confirmation

Enter your question in the input box to start query. Since the system may match multiple subjects/fields, you need to confirm the corresponding subject and field.

(1) Subject confirmation: When the question (for example, "Sales amount per store in California last year") triggers the system to match multiple analysis subjects marked by the Sales Scenario label, the system will display these subjects for you to select from. After the target analysis subject is selected:

(2) Field confirmation: If a question contains uncertain fields, FineChatBI will display the fields in the subject for you to confirm.

(3) Determine whether to save the confirmation record:

• If you click Confirm, the confirmation result of the subject and field will not be saved. You will need to reconfirm them the next time you ask this question.

•  If you click Confirm and Remember My Choice, the confirmation record of the subject and field will be saved, eliminating the need for reconfirmation the next time you ask this question.

(4) Confirmation record deletion: To delete previous subject or field confirmation records, you can click User Preference and delete historical confirmation records to restore to the initial state, as shown in the following figure.

Multi-Round Conversation

The system supports multi-round conversations, automatically linking contexts to provide coherent answers to subsequent questions:

Scenario one: Unchanged dimensions, with only the analyzed indicators to be switched

Scenario two: Unchanged dimensions, with only filter conditions to be switched

Result Accuracy Verification

FineChatBI will generate analysis ideas above the component, including data tables, filter conditions, summary methods, and other information. If they are all correct, the results are correct.

• Check if the used data table is correct.

• Check if the filter conditions are correct.

• Check if the group fields, the indicator fields, and the summary methods are correct.

Intelligent Rewriting

Note:

Admins can use this function only after configuring LLMs. For details, see LLM Configuration.

When you cannot get an accurate response to a natural language question from FineChatBI, you can click Unsatisfied with the result? Regenerate.

In this case, the system intelligently rewrites the natural language question to help FineChatBI obtain a more standard question. For example:

• Question asked in the natural language: "The best store"

• Standardized question after rewriting: "Store name with the highest sum of sales amount"

Chart Type Switchover

You can freely switch chart types to display analysis results according to your needs.

Supported chart types include Table, Column Chart, Pie Chart, Partitioned Column Chart, Multi-Series Line Chart, Multi-Series Line Chart, KPI Card, and Scatter Chart, as shown in the following figure.

Data Interpretation

FineChatBI's Data Interpretation function provides more intelligent data insight services. After the chart is generated, click Data Interpretation.

• Data performance analysis: The system will interpret the data performance in detail according to the chart content based on the asked question.

• Related external factors: The function will automatically associate the external information that affects the data (if any), provide a reasonable explanation for the chart data performance, and grasp the underlying reasons behind the data.

• Recommended subsequent questions: Based on the current analysis direction, the system will intelligently recommend a series of further questions for exploration, guiding you to dig deep into data and expand analysis dimensions.

Note:

Admins can use this function only after configuring LLMs. For details, see LLM Configuration.

Questions recommended by the data interpretation will appear below the data interpretation, according to which you can continue your analysis, as shown in the following figure.

Drill Analysis

FineChatBI allows drill analysis of generated charts.

For example, if you find that store 107 has the highest sales amount, you can further drill down the data to analyze the sales amount of each product category in store 107, as shown in the following figure.

Attribution Analysis

When too many dimensions are disassembled during analysis, the analysis loses focus. Either the cause of the problem cannot be found, or every dimension may be problematic. In this case, you can use Attribution Analysis to mine key influencing factors and disassemble dimensions for targeted analysis.

After you right-click the chart and select Attribution Analysis, the system will perform attribution analysis on the selected data and provide an analysis report, as shown in the following figure.

Note:

Attribution analysis can only be used when the entire table has at least two dimensions, and the question involves only one group and one indicator (where the indicator is neither a record count nor a calculation field).

Prediction Analysis

Method one: After obtaining the line chart, click the Trend Prediction on the right, as shown in the following figure.

Method two: After obtaining the trend of the line chart, you can continue to ask questions. For example, you can ask "Predict sales amount trends next year.", as shown in the following figure.

Generating a Dashboard

Click Generate Dashboard. In the pop-up box, select the components to be saved, set Name and Storage Location, and click Save. After the dashboard is successfully saved, click View to jump to the place where the dashboard is located.

The generated analysis subject includes data used during Q&A, data models, components, and the dashboard generated by components.

You can perform further analysis in the analysis subject.

Exporting as an Excel File/Image

Export the generated component as an Excel file or image.

• Export as Image: Export the component as an image.

• Export as Excel: Export the analyzed data as an Excel file.