E-commerce
June 28, 2026
Your store has 800 products, 14 filters per collection, and customers who abandon their cart after checking three incompatible boxes. E-commerce product filters work for expert buyers. For everyone else, they create analysis paralysis.
Fast Simon analyzed nearly 50,000 buyers in Q1 2026: interactive discovery dialogues boost conversion to 15-22% in fashion and footwear, nearly tripling the rate of keyword search alone (Fast Simon, AI shopper agents 2026).
This guide #110 approaches the AI catalog assistant as the primary navigation solution: intent dialogue → shortlist of 3-5 SKUs. It is distinct from the filters guide (#30), chatbot vs. search (#14), and the product quiz (fixed tree).
Summary
Why do wide catalogs paralyze buyers?
A large e-commerce catalog (500+ SKUs, multi-category) amplifies choice and friction simultaneously.
Symptoms of Catalog Paralysis
High collection bounce rate: PLP entry, exit without a product click
Long sessions, 0 ATC: infinite scrolling without a decision
Vague search terms: "cream", "gift", "black"
"I can't find it" tickets: support becomes the search engine
Empty combined filters: 0 frustrating results
Threshold and Business Cost
Generally 300+ active SKUs or 50+ products per major collection. Beyond this, filters alone are not enough without guidance. Roccai documents that too much choice reduces motivation and satisfaction: structure matters more than volume (Roccai, choice overload 2026).
Affected verticals: multi-brand fashion, spare parts, wide-range cosmetics, DIY, DTC marketplace. See searchandising, assisted customer journey.
What are the limitations of filters on large catalogs?
The product filter limits explain why they fail on their own with large catalogs.
Recurring UX Issues
Learning curve: the customer does not know your taxonomy
Dead combinations: filter A + B = 0 results
Mobile overload: 14 unreadable sidebar filters
Synonyms: searching for "sneakers", filter says "trainers"
Vague intent: "gift for mom" does not map to a filter
Filters vs Intent
Filters answer "which attributes". The customer thinks "what problem to solve". ScaleWise speaks of "filter fatigue": the menu designed to help costs the sale (ScaleWise, conversational product finder 2026).
When filters remain useful
Expert buyer, exact reference, re-purchase of the same SKU: quick filters. Optimal hybrid stack: assistant for vague intent, filters to refine the shortlist.
How to compare AI assistant, filters, and internal search?
Comparing AI assistant vs filters vs search clarifies the navigation stack.
Quick Decision Matrix
Vague intent: assistant wins, search is weak, filters are useless
Exact reference: search wins
Multi-faceted expert: filters win
Mobile: conversational dialogue wins
Zero results: assistant suggests alternatives
Distinction between quiz and gift finder
Quiz = fixed tree, same questions. Assistant = adaptive dialogue, follow-up, reformulation. Gift finder = gift-specific vertical. Catalog assistant = general navigation for any intent.
Zoovu: conversational search retains context and reduces the fear of missing out on a better option (Zoovu, conversational search 2026). See chatbot vs search (#14), gift finder chatbot.
What architecture for an AI catalog assistant?
The AI catalog assistant architecture connects dialogue, Shopify catalog, and recommendations.
Six components
Chat widget or dedicated "Find your product" page
Dialogue engine: progressive criteria collection
Catalog index: products, tags, metafields, stock, prices
Matching engine: multi-criteria relevance scoring
Presentation: 3-5 product cards + justification
Fallback: human handoff or broadening criteria
Shopify sync and guardrails
Products API: title, variants, inventory, collections
Metafields: usage, compatibility, level
Refresh: product/update webhook or hourly sync
No OOS recommend: except back-in-stock alert
Budget respected: customer range ±10%
Target latency < 3 s: Redis cache index. Generic LLM without catalog grounding = product hallucinations. See Shopify integration, train Shopify chatbot.
How to design the catalog navigation dialogs in catalogs?
Design the catalog navigation dialogue: progressive questions, not a 15-field interrogation.
4-6 step funnel
Open intent: "What are you looking for today?"
Usage or context: for whom, for what
Constraints: budget, deadline, size, compatibility
Optional preferences: color, brand, material
Refinement if too many matches: 1 discriminating question
Results + "Refine" or "See more"
UX dialogue rules
Quick replies: "< €50", "Sensitive skin", "Urgent this week"
Reformulation: bot confirms criteria before recommending
Early exit: customer mentions model XYZ → direct PDP
Exclusions: "no leather", "fragrance-free" first-class
Tone: store advisor, 1 question at a time
Roccai Framework: Ask (intent) → Filter (exclude) → Recommend (confident shortlist). See segmenting purchase intent.
How to score and recommend from the Shopify catalog?
The AI assistant product matching scores the catalog according to the collected criteria.
Scoring criteria
Match intent: usage tags, collection, product type
Budget: price varying within range
Stock: currently available is prioritized
Popularity: best-seller tie-break
Personalization: customer history if logged in
Results presentation
3-5 products max in the first response. Paralysis if 20 links. Each recommendation: 1 "Why" line. Alternative tier: premium + value + budget if trade-offs. Zero match: broaden criterion "No L, here is M available" or "Budget €80, here is €95 justified".
Business rules
Exclude outlet if customer is looking for new. Max 2 of the same brand unless explicitly requested. Variant explosion: assistant chooses the variant, not just the parent product.
See contextual recommendations, assistant vs recommendations.
Where should the catalog assistant be placed to maximize adoption?
The catalog assistant UX determines adoption and conversion.
Priority placements
Homepage hero: "Find the perfect product" CTA
Header collection: assistant above the product grid
Page /find: Indexable SEO "buying guide"
Search zero-result: "Describe it differently"
Menu: "Help me choose" entry
Mobile and measured proactivity
Mobile horizontal carousel, thumb-zone quick replies. Collection > 100 SKUs + 45s without click: "Need help finding?" chip, not a blocking pop-up. Progress indicator "Step 2/5" reduces drop-offs in long flows.
See optimize Shopify collection, mobile-first (#105), contextual help vs pop-ups (#107).
What are the use cases by e-commerce vertical?
Catalog assistant use case by vertical with adapted funnel criteria.
Six typical verticals
Fashion: gender, size, occasion, style, budget
Spare parts / tech: device model compatibility priority
Cosmetics: skin type, concern, excluded ingredients
Home / decor: space dimensions, style, color
B2B: volume, MCQ, pro use, qualification before recommendation
Multi-brand marketplace: intent-first, not brand-first
Specific funnels
Food: strict allergen exclusion, hard business rules. Sports: beginner / intermediate / pro level. Pet: species, breed, age. One funnel per mega-collection if logics differ (shoes vs electronics).
See compatibility questions, complex products, indecisive PDP assistant (variant choice on known product page, not global navigation).
How to integrate the assistant with existing filters and search?
Integrate the assistant with filters and search without cannibalization.
Bidirectional handoffs
Search → assistant: 0 or < 3 results → "Describe your need"
Assistant → filters: after shortlist → pre-filtered collection link
Deep link URL: `/collections/vestes?budget=100-150&size=L` generated by bot
Hybrid search bar and analytics
Placeholder: "Search or describe your need". NLP routes search vs dialogue. GA4 events: `catalog_assist_start`, `catalog_assist_complete`, `search_zero_result`. Fast Simon: the agent complements the search bar for long-tail zero-result queries, does not replace it.
See conversations merchandising (#108) for zero-match → tag enrichment.
What KPIs should be measured for an AI catalog assistant?
The AI catalog assistant KPIs measure navigation and conversion.
Engagement
Start rate: % of collection visitors launching the assistant
Completion rate: % completing the funnel
Drop-off step: problematic question
Conversion
Assist-to-ATC: % of sessions with add-to-cart post-assist
Assist-to-purchase: order within 24 hours
AOV assist vs non-assist: premium tier upsell
PLP conversion lift: A/B collection with assistant, 21 days minimum
Catalog Health
Zero-match rate = catalog gap. Long tail discovery: SKUs sold via assist that are never featured in ads. Deflection of "I can't find it" tickets. North star: revenue per session of large collections × completion rate. Assist-to-ATC benchmark: aim for 25-35% of assisted sessions.
See chatbot KPIs, chatbot ROI.
How does Qstomy help navigate large catalogs?
Qstomy deploys a Shopify catalog AI assistant connected to products, stock, and collections.
Key Capabilities
Catalog navigation flow: configurable dialog funnel
Shopify product sync: real-time inventory + metafields
Smart matching: multi-criteria scoring
In-chat product cards: image, price, ATC (Add to Cart)
Collection deep links: post-shortlist filter handoff
Zero-result rescue: internal search integration
DTC Case Study in Numbers
Outdoor store with 1,200 SKUs, jackets collection: 58% PLP bounce rate, 1.2% PLP conversion rate, 42 tickets / month of "I can't find my jacket".
Deployment of Qstomy collection header + zero-result hook. Customer: "3-day hiking jacket, autumn, €150 budget". 4 questions, 3 jackets recommended with waterproofing + weight justification, ATC in 2 min.
8-week A/B test results: assisted PLP conversion +38%, funnel completion 61%, assist-to-ATC 31%, "cannot find" tickets -47%, 12 long-tail SKUs discovered via assist.
Explore AI sales agent, AI support, cleaning bot corpus (#103), request a demo.
Which operational playbooks should be launched this week?
Playbook 1: top collection tag audit
On your #1 traffic collection, check completeness of usage / budget / fit tags on hero SKUs. Target > 90% before assistant launch. List 10 "I can't find" questions from 30-day tickets.
Playbook 2: 5-question funnel
Document 5-step funnel for a mega-collection. Test mobile mystery shop: find niche product in 2 min without assistant vs with. Time the delta.
Playbook 3: collection placement
Deploy "Find your product" CTA on header of poorest converting PLP collection. Quick replies for budget + usage. GA4 assist funnel events configured.
Playbook 4: zero-result search hook
Internal search 0 results → "Describe differently" chat invitation. Measure weekly zero-match rate → merchandising tags backlog (#108).
Playbook 5: 21-day A/B test
50% collection traffic with assistant CTA vs control with filters only. KPIs: PLP conversion, assist-to-ATC, revenue per session. Prove ROI on one collection before site rollout.
Useful links
A large catalog without a conversational guide externalizes navigation to the customer or support: the assistant transforms 800 SKUs into a dialogue, not a filter labyrinth.
Enzo
June 28, 2026
