By industry
Customer stories
See how our clients are using Zipchat and the awesome benefits it's bringing to their businesses!
Zipchat AI
Your AI Agent live in under 1 hour
No code. Trained on your catalog. Converts on every channel.
Create free Agent Book a demoLearn Agentic Commerce
Earn certification and merch rewards. No credit card needed.
Start now →Summary: Product discovery is the process by which shoppers find the right product. Ecommerce stores have 5 discovery patterns available in 2026: keyword, semantic, conversational, guided, and agentic. Each solves a different class of shopper intent. This guide maps each pattern to its best use case, failure modes, and conversion profile so you can pick the right stack for your catalog.
Most ecommerce CRO work targets the conversion funnel: landing page, product page, checkout. But 30% of sessions end before the shopper ever reaches a product page. They browse, search, find nothing relevant, and leave. This is a discovery failure, not a conversion failure.
The Baymard Institute 2024 site search benchmark puts the average zero-results rate at 12% to 18%. For category navigation, bounce from category pages without a product click averages 40% to 55%. Combined, this means more than half of ecommerce sessions end in discovery failure.
Fixing discovery captures more revenue than optimizing the checkout page of an already-engaged shopper.
For the cluster overview this fits into, see product discovery for ecommerce.
How it works: The search index stores product titles, descriptions, and tags as text tokens. When a shopper submits a query, the engine matches tokens in the query to tokens in the index. Exact matches rank highest; partial matches rank lower.
Best for: Simple catalogs (under 100 SKUs) where product names are unambiguous and shoppers know what to search for (“Nike Air Max 90,” “iPhone 15 Pro Case”).
Fails when: Shoppers describe needs in natural language (“something for dry, itchy skin”), use different terminology than the catalog (“trainers” vs. “sneakers”), or search for attribute combinations (“waterproof hiking boot under $150 for wide feet”).
Conversion profile: 2% to 4% for search sessions. Zero-results rate 12% to 18%.
How it works: Product data is converted to vector embeddings (mathematical representations of meaning). Query embeddings are matched to product embeddings by semantic similarity, not exact text. “Moisturizer” matches “hydrating cream” because they share semantic proximity.
Best for: Mid-size catalogs (50 to 2,000 SKUs) where synonym density is high and shoppers use varied terminology. Fashion, beauty, and home goods benefit most.
Fails when: Queries require AND logic across multiple attributes (“fragrance-free AND for rosacea AND SPF”). Semantic search returns items that are close on one attribute but not all three.
Conversion profile: 4% to 7% for search sessions. Zero-results rate 4% to 7%.
How it works: Shoppers submit natural language queries to a chat interface. AI interprets intent, queries the catalog semantically, and returns relevant products with explanations. Follow-up questions maintain context. “Which of these is fragrance-free?” understands “these” refers to the previously presented products.
Best for: Complex catalogs where shoppers need to describe a problem or need that does not map to a product category (“something to help my daughter sleep better without medication”).
Fails when: Response latency exceeds 3 seconds (shopper impatience threshold) or when the AI hallucates product attributes that are not in the catalog.
Conversion profile: 8% to 15% for sessions with AI interaction. Zero-results rate under 3%.
How it works: A structured question flow narrows the catalog based on shopper answers. Each answer filters the product set. The flow terminates when 3 to 5 relevant products remain.
Best for: High-consideration categories where the right product depends on personal attributes: skincare (skin type, concerns), supplements (health goals, dietary restrictions), electronics (use case, compatibility requirements).
Fails when: Shoppers are in a hurry or do not want to answer questions before seeing products. Drop-off in guided flows exceeds 40% if the first question is too broad or the flow has more than 4 steps.
Conversion profile: 10% to 18% for completed guided flows. Completion rate 40% to 60% depending on category and flow design.
How it works: AI acts as a personal shopper with autonomous initiative. It detects when a shopper is stalling (60 seconds on a category page without a product click) and proactively engages. It reasons about the shopper’s stated need, asks clarifying questions, retrieves relevant products, explains the recommendation, and handles follow-up questions. The search and chat distinction collapses: the same AI handles a direct search query and an exploratory “I’m not sure what I need” conversation.
Best for: Any catalog where shoppers need help finding the right product and the cost of a lost sale exceeds the cost of engagement. Highest ROI in beauty, supplements, specialty electronics, and high-ticket items.
Fails when: The AI is poorly constrained and makes product claims not supported by catalog data. Requires ongoing catalog data maintenance to prevent accuracy degradation.
Conversion profile: 10% to 20% for sessions with agentic engagement. Zero-results rate under 2%.
| Catalog profile | Recommended pattern | Expected zero-results rate |
|---|---|---|
| Under 50 SKUs, simple names | Keyword search | 5% to 10% |
| 50 to 200 SKUs, varied terminology | Semantic search | 3% to 7% |
| 200 to 2,000 SKUs, complex attributes | Conversational search | 1% to 3% |
| Any size, high-consideration purchase | Guided + conversational | 0.5% to 2% |
| Any size, proactive engagement goal | Agentic discovery | Under 2% |
Navlas SK deploys Zipchat for guided haircare discovery. Shoppers describe their hair type, damage level, and treatment goals. The AI surfaces the relevant treatment products and explains why each matches. See how Navlas SK uses agentic discovery for expert haircare guidance.
Collezione Casa uses conversational discovery for home furnishings. Shoppers describe a room’s existing style and get product recommendations filtered by aesthetic compatibility. How Collezione Casa handles product discovery with Zipchat.
Vaonis uses Zipchat for telescope product discovery across an international customer base. Technical queries about telescope specifications, compatibility with observing conditions, and suitability for different skill levels are answered in 20+ languages. See the Vaonis case.
The right discovery stack is the simplest one that eliminates your specific failure pattern.
Run this diagnostic:
If your zero-results rate is above 8%: upgrade to semantic or AI search first. If more than 20% of queries are intent-based: add conversational search. If “which product is right for me?” is a top support query: add guided discovery or agentic engagement.
Agentic commerce explained: what it is, how it differs from conversational commerce, the 4 agent capabilities, 10 use cases, ROI benchmarks, and the 2026 platform shift.
Learn what agentic search is, how it differs from keyword and semantic search, and why it cuts zero-results rate to under 2% for ecommerce stores.
Should you build or buy an AI chatbot for ecommerce? True cost comparison, timeline reality, maintenance burden, and when building makes sense vs. when buying wins.
Everything you need to know about AI chatbots for Shopify. WISMO automation, cart recovery, product recommendations, Shopify Plus support. Compare the top apps for 2026.
