Each attribute in your schema has an AI strictness setting that controls how aggressive AI suggestions are when mapping values. This gives you fine-grained control: tight constraints for attributes with controlled vocabularies, more freedom for attributes where specificity matters.
The Three Modes
Strict
AI will always map to existing approved values. It will never suggest creating a new value. If no good match exists among your approved values, the AI will leave the value unmapped rather than invent something.
Best for: Controlled vocabularies where your list of valid values is complete and should not grow.
- Basic colors (red, blue, green, black, white, etc.)
- Standard sizes (XS, S, M, L, XL, XXL)
- Boolean attributes (yes/no, true/false)
- Gender (men, women, unisex)
Moderate (Default)
AI will prefer mapping to existing approved values, but will suggest creating a new value when the incoming value is genuinely distinct and doesn't have a reasonable match in your current list.
Best for: Most attributes. This is the default mode and works well when your approved value list is mostly complete but may need occasional additions.
- Brand names
- Categories and subcategories
- Common materials (wood, metal, plastic, fabric)
- Standard product types
Flexible (Lenient)
AI is more willing to suggest new values rather than collapsing everything into existing ones. It still maps obvious matches (abbreviations, translations) but preserves specificity when the incoming value carries meaningful detail.
Best for: Attributes where specificity matters and your value list should grow to capture nuance.
- Detailed materials ("Brushed Stainless Steel" vs. just "Steel")
- Specific finishes ("Matte Black Powder Coat" vs. just "Black")
- Technical specifications
- Style descriptors
How to Set It
Strictness is configured per attribute on the Value Mapping tab when viewing a feed. Select an attribute, and you'll see the strictness toggle (Strict / Moderate / Flexible) next to the AI Suggest button. Change it before running AI suggestions to control how the AI behaves for that attribute. The setting is saved on the attribute and persists across feeds.
What It Does Not Affect
Strictness modes apply only to AI suggestions. They do not affect:
- Manual mapping — you can always manually map any value to any approved value, or create new approved values, regardless of the strictness setting.
- Deterministic normalization — rules like unit conversion, HTML stripping, whitespace cleanup, and transform rules run independently of the AI strictness mode.
Example: Color Attribute
Imagine your approved color values are: black, white, red, blue, green, yellow, grey, brown, navy, pink.
A supplier sends the value "Dusty Rose".
- Strict: Maps to "pink" (closest existing value) or leaves unmapped if the match isn't confident enough.
- Moderate: Maps to "pink" since it's a reasonable match and the value isn't radically different.
- Flexible: Suggests creating "dusty rose" as a new approved value, preserving the specificity.
The right choice depends on your use case. If your store only shows basic color filters, Strict is appropriate. If you sell fashion and "dusty rose" is a meaningful distinction for your customers, Flexible makes more sense.