Attribute Extraction allows AI Workflows to automatically detect and fill structured product attributes based on existing product content.

Instead of manually completing product data, AI can analyze descriptions, specifications and other content to generate missing attribute values automatically.

This is especially useful when supplier data is incomplete or unstructured.

Attribute Extraction reads existing product information and converts it into structured attribute data.

Examples:

The generated values are written into selected attributes inside Elovate.

Many suppliers only provide:

without proper attribute structures.

This creates problems for:

Attribute Extraction helps automate this process at scale.

Attribute Extraction is commonly used for:

A supplier provides only this description:

"Premium salmon cat food for adult cats with sensitive digestion."

AI can automatically extract:

These attributes can then be used for:

An Attribute Extraction action consists of:

The workflow processes products individually and attempts to fill the configured attributes automatically.

Inside a workflow:

You will then enter the Attribute Extraction configuration screen.

The configuration screen allows you to customize how extraction behaves.

Common settings include:

You can select:

Example:

Different models may affect:

You must choose which attributes the AI should fill.

Examples:

Only selected attributes are processed.

The prompt determines how the AI should extract information.

Prompts can range from:

Example:
"Extract the flavor and lifecycle stage from the product description."

Advanced prompts may:

Prompt quality strongly affects extraction accuracy.

The scope determines which products the action can process.

This is often linked to:

Well defined scopes improve:

Before running the action on large product groups, you can test extraction on selected products.

Testing helps validate:

This is strongly recommended before full execution.

Test results display:

You can:

Reasoning explains why the AI selected certain attribute values.

Example:
"The description references adult salmon cat food, therefore Flavor was set to Salmon and Lifecycle to Adult."

Reasoning improves transparency and moderation control.

Confidence scoring indicates how certain the AI is about the generated result.

Higher scores usually indicate:

Lower scores may require:

After testing:

Products are then processed individually through the extraction pipeline.

Results become visible inside the Results tab.

The Results tab displays:

This provides visibility into extraction progress.

Depending on the workflow configuration:

Once approved, extracted attributes synchronize back into the product data.

Begin with simple high confidence attributes such as:

before attempting more complex extraction logic.

Different categories often require different extraction logic.

Example:

This improves extraction accuracy.

Always test:

before running large workflows.

Low confidence outputs may require:

Confidence scoring should help prioritize manual review.

Example:
A webshop imports pet food products with incomplete attribute data.

Workflow:

This entire process runs automatically inside one AI Workflow.