When Buyers Describe a Need but Do Not Know the SKU: How Should AI Product Discovery Work?

Taxonomy

Product families, attributes and parent-child relations must be consistent.

Synonyms

Connect buyer language, technical terms, old codes and language variants.

Constraint graph

Represent mandatory conditions, preferences, conflicts and substitutions.

Evidence

Link datasheets, certificates, effective dates and applicable markets.

• Avoid forced answers: Keep multiple options when evidence cannot distinguish them.
• Require sources: Do not use marketing copy as a technical criterion.
• Expose assumptions: Show customers and Sales what remains unconfirmed.
• Block high-risk advice: Escalate safety and compliance decisions.
• Regression testing: Retest after taxonomy, document or rule changes.

Must SKU data be perfect?

No, but the pilot line needs reliable core attributes and sources.

Can AI suggest substitutes?

Yes, when substitution relations and differences are explicitly modeled.

How many questions should it ask?

Only until the shortlist is useful or expert review is required.

What if the buyer does not know a specification?

Ask through application examples or escalate; do not fill it in.

Which KPIs fit?

Shortlist accuracy, questions to brief completion and Sales acceptance.

Good discovery removes the burden of remembering codes and gives Sales richer context. The value comes not merely from natural conversation, but from taxonomy, constraints, evidence and an expert feedback loop.