Conversational Search

What is Conversational Search?

Conversational search is an interactive search experience that maintains context across turns. The system parses intent, asks clarifying questions, and returns grounded answers or product suggestions instead of a static list only.

How it Works (quick)

• NLU: Identify intent, entities, constraints (brand, size, budget).
• Grounded retrieval: Hybrid keyword + vector search with filters/ACLs.
• Clarification: Ask for missing constraints when confidence is low.
• Answering: Summarize with links/cards; cite sources; keep state.
• Safety & analytics: Guardrails, PII filters, and metrics per turn.

Why it Matters in E-commerce

• Faster path to the right item: “Find a waterproof hiking jacket under €150 in M.”
• Lower zero results: System negotiates constraints before searching.
• Higher conversion & support deflection: Mix PDPs, guides, and FAQs.

Best Practices

• Start structured: facets first, then natural chat.
• Ground every answer in retrieved items/pages (no free-form hallucination).
• Show clarifying chips (price, size, brand) and let users tap.
• Cache session context; log intents and missing facets for merchandising.
• Track success: solved in ≤3 turns, CTR, conversion, reformulations.

Challenges

• Latency budgets with multiple model calls.
• Ambiguity and over-confident answers.
• Privacy/PII and compliance logging.

Examples

• “Running shoes for flat feet, size 45, budget 100–150” → guided chips; filtered results.
• “How to wash merino base layers?” → answer + link to care article + relevant PDPs.

Summary

Conversational search combines intent understanding with grounded retrieval to guide shoppers quickly to the right products and answers—under strict latency and safety controls.