BAIA Configuration
Info
The BAIA configuration described here, enables the following AI features in PIM:
For enabling AI Product Placement Suggestions please refer to the developer tutorial.
BAIA text generation is configured in the following ways:
- Relevant attributes: Configure which attributes are relevant for BAIA text generation.
- Attribute field configuration: Configure AI text generation for attribute fields. An attribute field must have a configuration to be able to generate text for it.
- Product category configuration: Configure AI text generation for product categories, when they are set as a primary category for a product.
When a product has a primary category that has been configured for BAIA text generation, the category configuration takes precedence over attribute field configuration. If a product has multiple primary categories with configurations, all category configurations are used simultaneously.
Relevant attributes
Relevant attributes are configured globally as a whitelist that applies to all BAIA text generation configurations. These attributes serve as input for the AI when generating text or suggestions. The relevant attributes you select should be present on your products to get the best possible results.
Note
Any attribute that does not use predefined values can be selected as a relevant attribute.
Relevant attributes for BAIA text generation can be configured on the Relevant Attributes page, which can be found under the BAIA section in PIM.
Configure attribute fields
Attribute configurations allow you to set up AI text generation for specific attribute fields. An attribute configuration is required to generate text for that specific attribute field.
A new BAIA attribute configuration for an attribute field can be created from the attribute details page, as seen in the screenshot below:
All existing attribute configurations are listed on the BAIA Configuration page, which can be found under the BAIA section in PIM. From here, existing configurations can be modified or deleted.
The following input parameters must be configured per attribute field:
- Best practice products: Products that set the standard of well-enriched content
- Use externally owned value: Option to use the externally owned value for the attribute field as input
- Instruction prompt: Context and guidance for the AI feature
Best practice products
The products configured as best practice products act as example products for the AI feature. Best practice products are considered well-enriched products that BAIA uses to understand what characteristics the generated value or suggestion should have. The AI engine learns how to interpret attributes and master data properties from these best practice products.
Use externally owned value for field as input
When this option is enabled, the externally owned value for the attribute field will be used as input for text generation if the externally owned value is present. This is useful for generating a text based on an external source, such as a product description from a supplier or manufacturer.
Note
Only the externally owned value for the attribute field will be used as input for text generation, the relevant attributes' values will be resolved based on the Override rule on the attribute.
Instruction prompt
Providing context to the AI feature in the form of an instruction prompt allows users to customize its result. The instruction prompt can contain any type of instruction, from text length restrictions to instructions regarding tone of voice and target audience.
The instruction prompt is segmented for segmented attribute field types. This means that a specific instruction prompt can be set for each segment, in case the data type of the attribute field which the AI feature is configured for is a segmented type.
Example
A custom instruction per segment allows for targeting different audiences, for example, when working with channel segmented product descriptions. A good example of this is creating a specific prompt for the B2B and B2C segments.
This prompt can instruct the text generation for the B2B segment to include specific technical details about a product, which could be interesting for the business customer reading it:
Generate a product description with targeting other businesses as customers. The description should include a summary of the technical specifications of the product.
The B2C text on the other hand, can be generated with specific use cases in mind which are relevant for a regular customer:
Generate a product description with a friendly, approachable tone. The description should explain the product's main uses and benefits, appealing to a broad audience. Keep the language simple and relatable.
How to Write a Good Instruction
- Avoid negatives: Instead of telling BAIA what not to do, state what it should do.
- Be precise and consistent: Use clear, direct language and ensure your examples match your instructions.
- Specify the expected output: Clearly define what you want BAIA to generate, including length, tone, and style.
Configure product categories
Product category configuration allows you to set up AI text generation at the category level. When a product has a primary category (or one of the primary category's ancestor categories) with a BAIA configuration, that category configuration is used for text generation instead of the attribute configuration.
A new BAIA product category configuration can be created/edited from the category's details page.
Existing BAIA product category configurations are listed in the CATEGORIES-tab on the BAIA Configuration page under the BAIA section in PIM.
The following input parameters can be configured per product category:
- Best practice products: Same as for attribute configurations. (See Best practice products)
- Instruction prompt: Same as for attribute configurations, except they are not segmented. (See Instruction prompt)
Note
The relevant attributes used as input are configured globally (see Relevant attributes section above) and apply to all product category configurations.
Configuration inheritance
Product categories can inherit BAIA configuration from their ancestor categories. If a category doesn't have its own configuration, BAIA will look up the category hierarchy to find the nearest ancestor with a configuration.
Example
If you have a category hierarchy like Furniture > Chairs > Office Chairs, you can configure BAIA at the Furniture level, and all subcategories will inherit that configuration unless they have their own specific configuration.
Configuration resolution
BAIA uses the following priority order when determining which configuration to apply:
-
Product category configuration (highest priority)
- If the product has one or more primary categories with BAIA configuration, those configurations are used
- Multiple primary category configurations are applied simultaneously
- Configuration can be inherited from ancestor categories
-
Attribute field configuration (fallback)
- If no primary category has BAIA configuration, the attribute field configuration is used
- This applies when the product has no primary categories or when primary categories (or ancestors) lack BAIA configuration
Configuration resolution examples
Example 1: Product with configured primary category
- Product: "Office Chair Pro"
- Primary category: "Office Chairs" (has BAIA configuration)
- Attribute field "Description" (has BAIA configuration)
- Result: "Office Chairs" category configuration is used
Example 2: Product with multiple primary categories
- Product: "Gaming Office Chair"
- Primary categories: "Office Chairs" (configured), "Gaming Furniture" (configured)
- Attribute field "Description" (has BAIA configuration)
- Result: Both category configurations are used simultaneously
Example 3: Product without category configuration
- Product: "Standard Chair"
- Primary category: "Chairs" (no BAIA configuration)
- Attribute field "Description" (has BAIA configuration)
- Result: Attribute configuration for the "Description" field is used
Example 4: Product with inherited category configuration
- Product: "Executive Office Chair"
- Primary category: "Office Chairs" (no BAIA configuration)
- Parent category: "Chairs" (has BAIA configuration)
- Result: "Chairs" category configuration is inherited and used
Dos and don'ts
When configuring BAIA for product text generation, follow these guidelines to ensure high-quality, reliable results.
Best Practice Products and Inputs
Ensure Best Practice Products Only Contain Information Present in the Inputs
Example: If your input attributes are Name, Material, and Color, your best practice product output should only reference these.
Input: Name: Classic Chair, Material: Oak, Color: Brown
Output: "The Classic Chair is crafted from oak and comes in a brown finish."
Why: Including only information from the inputs prevents BAIA from hallucinating or inventing details.
Don't Include Details in Best Practice Products That Aren't in the Inputs
Example: If your input attributes are Name and Color, but your best practice product output says "Made from oak wood", BAIA may start hallucinating materials.
Input: Name: Classic Chair, Color: Brown
Output: "The Classic Chair comes in a brown finish and is made from oak."
Why: This teaches BAIA to invent information, which can lead to unreliable product texts.
Instructions and Examples Consistency
Keep Instructions and Examples Consistent
Example: If your instruction says "Do not mention the word 'Pineapple'", do not use 'Pineapple' in your examples.
Why: Contradictory instructions and examples confuse the AI and reduce output quality.
Don't Mix Contradictory Instructions and Examples
Example: Instruction: "Never mention 'Pineapple'." Example: "This pineapple chair is..."
Why: Inconsistencies confuse the AI and degrade output quality.
Don't Provide Examples That Contradict Length Instructions
Example: If your instruction says "Write a product description in 40-50 words", your example should not only contain 20 words.
Why: Examples that do not match the specified length confuse the AI and reduce the quality and consistency of the generated output.
Clarity and Precision in Instructions
Be Precise and Consistent in Instructions
Example: "Always use metric units. Write in a formal tone."
Why: Clear, unambiguous instructions help BAIA generate the desired output.
Don't Use Negatives in Instructions
Example: Avoid: "Do not use technical jargon." Prefer: "Use simple, everyday language."
Why: Negatives can be misinterpreted by the AI, leading to unexpected results.