Best Practices for Using Power Pilot

Modeling

The semantic model is at the heart of the interaction between your Power BI data and AI. It translates the technical structure of your data into understandable business concepts, allowing AI to understand questions such as:

“What was the total sales amount by category in the last quarter?”

For this interpretation to be accurate, the model must reflect clear relationships between the data contained in the dataset.

The Star Schema is the recommended architecture for analytical models in general and is especially important for Power BI models, particularly when integrated with an AI environment. It provides a leaner and more cohesive model, enabling better interpretation of relationships between tables.

• Fact:

Concentrates business events or transactions. Examples: FactSales, FactOrders, FactSupportTickets

• Dimension:

Describes contextual attributes related to fact tables. Examples: DimCustomer, DimProduct, DimDate, DimRegion

Key Characteristics:

1. 1:N relationships;

2. No complex joins or multiple ambiguous paths;

3. Simple and intuitive structure for semantic interpretation.

Naming:

The assistant uses NLP (Natural Language Processing), which means it analyzes the Power BI model linguistically. In other words, names and descriptions are extremely important to how AI understands the data.

• Avoid abbreviations: Use complete and clear names. Example: QuantitySold instead of QtySold.

• Use descriptive and semantic names: Make the meaning explicit. Example: SalesAmount, OrderDate, ProductCategory instead of SaleAmt, OrderDt.

• Avoid special characters and accents: Use only letters and numbers. Example: UnitPrice instead of Preço_Unitário.

Each table and column in Power BI has a description field that can—and should—be filled in with information about:

1. The meaning of the data;

2. The business context;

3. How the information is calculated or where it comes from.

These descriptions are directly consumed by AI, expanding its ability to reason about the context and generate more accurate and contextualized responses.

WARNING: Descriptions should be clear, concise, and objective. The amount of content included in a description impacts token usage. Therefore, be mindful when writing descriptions in the model.

Example:

Table: FactSales Description: Contains records of completed sales. Each row represents an individual transaction. The associated measures reflect the sales amount and quantity sold.

Column: SalesAmount Description: Total monetary value of the sale, including applied discounts.

The same principle applies to measures. Define clear DAX measures with semantic names and appropriate documentation.

1. Use meaningful names: TotalSales, GrossProfit, ProfitMargin;

2. Avoid duplicated measures or redundant logic;

3. Document measure descriptions.