How do I set up an on-site visual merchandising engine for shoppable collection pages and curated bundles?

Here’s a practical way to set up an on-site visual merchandising engine for shoppable collection pages and curated bundles.

1) Define the merchandising goals

Start by deciding what the engine should optimize for:

• Revenue: maximize conversion or AOV
• Discovery: expose new or long-tail products
• Margin: prioritize higher-margin items
• Inventory: clear overstock / aged stock
• Brand rules: preserve style, seasonality, or campaign intent

You usually want a ranking layer that balances these goals rather than relying on one static sort order.

---

2) Build the core data foundation

Your engine will only be as good as the product data feeding it.

Product catalog fields

Make sure each SKU/product has:

• Title, description, category, subcategory
• Price, compare-at price, discount
• Inventory on hand, sell-through, stockout risk
• Margin / profitability
• Attributes: color, size, material, style, season, occasion
• Ratings, reviews, return rate
• Historical performance: CTR, CVR, add-to-cart, revenue per impression
• Visual assets: primary image, gallery, video, model shots
• Compatibility signals for bundles: “pairs with,” “complete the look,” etc.

Event tracking

Track:

• Product impressions
• Clicks
• Add-to-cart
• Purchases
• Bundle interactions
• Collection page sort/filter usage
• Time on page and scroll depth

This lets you measure merchandising effectiveness and tune ranking rules.

---

3) Create a merchandising rules engine

A good engine usually combines:

A. Hard rules

These always override ranking logic.

Examples:

• Hide out-of-stock products
• Pin campaign products to the top
• Exclude products under legal or brand restrictions
• Only show matching gender/season/region products

B. Soft ranking rules

These adjust ordering.

Examples:

• Boost products with high conversion
• Penalize low stock
• Boost new arrivals for the first 48 hours
• Boost high-margin products
• Boost products with strong image engagement

C. Contextual rules

Change rankings by:

• Channel: homepage vs collection page vs search vs email landing
• Device: mobile vs desktop
• Season or weather
• Audience segment: new visitor, returning buyer, VIP, etc.
• Geography or store region

---

4) Use a scoring model for collection pages

For shoppable collection pages, create a composite score per product.

Example scoring inputs:

• Relevance to collection
• Conversion rate
• CTR
• Margin
• Inventory health
• Discount attractiveness
• Recency/newness
• Editorial priority

Example:

score =
  0.30 * relevance +
  0.20 * conversion +
  0.15 * CTR +
  0.10 * margin +
  0.10 * inventory_health +
  0.10 * editorial_priority +
  0.05 * newness

Then:

• Rank products by score
• Apply rules for pinning, blocking, and boosting
• Refresh rankings regularly

A/B test the weighting.

---

5) Make collection pages shoppable

For collection pages, the merchandising engine should support:

Product card enhancements

• Quick add to cart
• Quick view
• Variant selection
• “Complete the look” modules
• Badges: best seller, new, low stock, sale

Dynamic placement logic

• Top rows: highest-performing or editorially important items
• Mid-page: supporting products and variants
• Bottom: discovery items, long-tail inventory, and complementary products

Filters and sorts

Offer dynamic filters based on available inventory and shopper behavior:

• Sort by relevance, best sellers, new arrivals, price, margin, etc.
• Hide filters with too few results
• Surface high-intent filters based on browsing history

---

6) Build bundle generation logic

Curated bundles need a slightly different system than collections.

Bundle types

• Pre-built curated bundles: fixed combinations selected by merchandisers
• Dynamic bundles: generated from rules and compatibility data
• Personalized bundles: tailored to user behavior or cart contents

Bundle construction rules

Use compatibility logic such as:

• Same style family
• Complementary color palette
• Occasion pairing
• Cross-sell relationships
• “Frequently bought together”
• Size/variant availability compatibility

Bundle scoring

Score bundles by:

• Likelihood to convert
• Combined margin
• Inventory balancing
• AOV uplift potential
• Visual coherence
• Diversity of price points

Bundle presentation

Show:

• Bundle headline
• Total price and savings
• “Add all to cart”
• Individual item controls
• Visual grouping that clearly communicates the theme

---

7) Add a merchandising admin layer

Merchandisers need a UI to control the engine.

Key capabilities:

• Pin/unpin products
• Apply boosts and exclusions
• Build and edit bundles
• Schedule campaigns
• Set region/channel rules
• Preview storefront output
• Compare performance by rule or campaign
• Roll back changes

This is critical so teams can manage merchandising without engineering involvement every time.

---

8) Implement experimentation and optimization

To make the engine improve over time:

A/B tests

Test:

• Different ranking formulas
• Different bundle compositions
• Different page layouts
• Different badge treatments
• Different CTA wording

Metrics

Measure:

• CTR
• Add-to-cart rate
• Conversion rate
• AOV
• Revenue per visitor
• Gross margin per visitor
• Bundle attach rate
• Stock sell-through

Continuous optimization

Feed performance back into the engine so the system learns:

• Which products convert in which contexts
• Which bundles work best together
• Which merchandising rules help or hurt performance

---

9) Support personalization where it matters

Personalization can be powerful, especially for bundles and collection pages.

Examples:

• New visitors: show best sellers and broad appeal
• Returning visitors: surface items similar to prior purchases
• High-intent shoppers: show premium or complementary products
• Cart-aware bundles: recommend items that complete the current basket

Keep personalization bounded so it doesn’t conflict with campaign priorities.

---

10) Architecture recommendation

A simple architecture looks like this:

Data layer

• Product catalog
• Inventory/pricing service
• Analytics/event pipeline
• Customer/session data

Merchandising engine

• Rule evaluator
• Scoring/ranking service
• Bundle generator
• Experimentation framework

Frontend layer

• Collection page CMS/component
• Bundle module
• Product card quick actions
• Preview and fallback states

Operational layer

• Admin dashboard
• Scheduling
• Approval workflows
• Monitoring and alerts

---

11) Start with a phased rollout

A practical implementation plan:

Phase 1: Static merchandising

• Manual collections
• Basic bundles
• Simple pinning and exclusions

Phase 2: Rule-based engine

• Add scoring
• Stock-aware ranking
• Campaign scheduling
• Bundle rules

Phase 3: Optimization

• A/B testing
• Behavioral signals
• Personalization
• Automated bundle recommendations

Phase 4: AI-assisted merchandising

• Auto-generated collections
• Dynamic bundle suggestions
• Forecast-driven inventory balancing
• Predictive ranking adjustments

---

12) Common pitfalls to avoid

• Over-optimizing for conversion and ignoring margin or inventory
• Too many rules causing conflicts
• Poor image consistency breaking bundle presentation
• No fallback when inventory is low
• Lack of QA for region/device/variant availability
• Merchandisers not having direct control

---

Suggested MVP

If you want the shortest path to value, build this first:

1. Product feed with rich attributes
2. Rule-based ranking for collection pages
3. Pin/boost/exclude admin controls
4. Basic curated bundle builder
5. Performance tracking dashboard
6. A/B testing framework

That will get you a functional merchandising engine without overengineering it.

If you want, I can also provide:

• a reference architecture diagram
• a data model/schema
• or sample ranking logic/pseudocode for the engine.