Analytics Agent Answer Accuracy Improvement - Best Practices

In response to the most frequently asked question: What methods are available to improve answer accuracy? Which scenarios are suitable for each method? How to manage them?

1. What Capabilities Improve Accuracy?

Analytics Agent (formerly known as DataGPT) provides 4 core capabilities to give the model more business information input and improve answer accuracy:

1. Table/Column Schema Annotations

2. Knowledge (Knowledge Base / Analysis Documents)

Business background knowledge provided in file form, injected into conversation context via RAG retrieval.

Content suitable for the knowledge base:

• Business rule descriptions
• Data dictionary / code mapping tables
• Industry terminology explanations
• Analysis methodology / calculation formula specifications
• FAQ frequently asked questions
• Content recalled only in specific scenarios

Core Value: Helps the model understand the "business logic behind the data," not just the data structure.

3. Metrics / Answer Builder

Predefined calculation formulas supporting alias matching, ensuring consistent metric definitions. Specifically:

• Metrics: Used to define aggregation calculations (e.g., SUM, AVG)
• Answer Builder: Used to define multi-table JOINs and complex filter conditions

Element	Description	Screenshot
Metric Name	e.g., "Average Transaction Value"
Calculation	SUM(revenue) / COUNT(DISTINCT order_id)
Alias	Helps the model understand different names for the same concept
Auto-Generation	Basic aggregate metrics (e.g., SUM, COUNT) can now be auto-generated for selected tables. Context-aware metric insights and auto-generation will be available in future releases, enabling smarter metric recommendations

Core Value: Ensures key business metrics are calculated consistently and accurately, preventing different SQL from being generated each time

4. Domain Prompt

System prompt at the analysis domain level, equivalent to giving the model a "role setting" and "behavioral guidelines."

• Role setting: "You are a data analyst for the Hong Kong food & beverage industry"
• Answer guidelines: "Monetary amounts default to HKD, date format uses YYYY-MM-DD"
• Business constraints: "When calculating same-store growth, only include stores that have been open for at least 12 months"
• Output preferences: "For trend-related questions, prioritize using line charts"

Core Value: Set global rules to avoid repeating prerequisites in every conversation

2. Which Capability for Which Scenario? (Scenario → Capability Mapping)

Decision Flowchart

User question is inaccurate

↓

Is it "wrong table/column selection"? → Yes → Improve Schema Annotations + Column Value Index

↓ No

Is it "wrong calculation logic"? → Yes → Create Metrics or Answer Builder

↓ No

Is it "not understanding business concepts"? → Yes → Add Knowledge Base documents

↓ No

Is it "need to repeat prerequisites every time"? → Yes → Configure Domain Prompt

↓ No

Investigate using a combination of the above methods

3. How to Use? (Knowledge Base Management)

3.1 Knowledge Base Organization

The new knowledge base uses a hierarchical folder structure (similar to Lark/Feishu knowledge base):

3.2 Content Creation Methods

3.3 Associate with Analysis Domains (Critical!)

Knowledge base content must be associated with an analysis domain for conversations under that domain to retrieve it:

• Can associate at the folder level → child files automatically inherit
• A single file can be associated with multiple analysis domains
• Unassociated files will not be retrieved by any conversation

Best Practices:

• Create folders by business theme, associate at the folder level
• General knowledge (e.g., company glossary) associates with all analysis domains
• Specific business knowledge associates only with the corresponding domain

3.4 What Content is Suitable for Knowledge Base?

✅ Content suitable for knowledge base:

❌ Content NOT suitable for knowledge base:

3.5 Management Recommendations

4. Summary: Synergy of the 4 Capabilities

In one sentence:

• Schema Annotations help the model "find the right data"
• Metrics help the model "calculate the right numbers"
• Knowledge Base helps the model "understand the business"
• Domain Prompt helps the model "follow the rules"

Use all four together for the highest accuracy.