The Problem: Bundles That Nobody Buys
Most merchants create bundles manually based on gut feeling. They pair products that seem related, slap a 10% discount on them, and hope for the best. The result: bundles that don't match real purchase patterns and discounts that either leave money on the table or aren't compelling enough to convert.
SellerZoom's AI Bundle Generator takes a data-first approach. It analyzes your actual co-purchase patterns using graph clustering (Louvain community detection), identifies product groups that customers naturally buy together, and generates ready-to-sell bundles complete with AI-written names, descriptions, and optimized discount strategies.
How It Works Under the Hood
The bundle generation pipeline has five stages. First, SellerZoom builds a co-purchase graph — a network where products are nodes and edges represent how often they're bought together. Only product pairs with at least 3 co-purchases qualify.
Second, Louvain community detection identifies natural clusters in this graph. These clusters represent products that real customers treat as a set — not just products that look similar to a merchandiser.
Third, each cluster is validated: SellerZoom checks that the products make sense together (different categories, compatible attributes, in-stock). Fourth, an LLM generates names and descriptions for each bundle — compelling, marketable copy like "Weekend Camping Essentials" or "Home Barista Starter Kit."
Finally, the system prices each bundle using a discount optimization algorithm that maximizes total bundle revenue (not just conversion rate).
Setting It Up: Step by Step
Ensure Sufficient Order History
The bundle generator needs co-purchase data to work. You'll need at least 200 orders with multi-item purchases. Go to Products in your dashboard and check the "Relationships" count — you need at least 5 product relationships with 3+ co-purchases each.
Navigate to Bundles
Click Bundles in the sidebar. If you have enough data, you'll see a "Generate Bundles" button. If not, the page will tell you how many more co-purchase relationships are needed.
Generate & Review
Click "Generate Bundles." The AI will process your co-purchase graph, detect clusters, and produce bundle candidates. Review each bundle — you can rename them, remove products, adjust the discount, or reject bundles that don't fit your merchandising strategy.
Publish to Your Store
Approved bundles are pushed to your storefront. They appear in the recommendation widget as "Bundle Deals" and can also be displayed on product pages for items included in a bundle.
Test Discount Levels
Use A/B Testing to run bundle discount experiments. Test 10% vs. 15% vs. 20% off — the right discount maximizes total revenue, not just conversion rate. A 10% discount that sells 3x more bundles can beat a 20% discount.
Why AI-Generated Bundles Outperform Manual Ones
Data beats intuition. Your merchandising team might think "running shoes + sports socks" is a natural bundle. But the co-purchase data might reveal that running shoes are most frequently bought with foam rollers and energy gels. Data-driven bundles match actual shopping behavior, not assumed behavior.
LLM naming converts better. A bundle called "Product A + Product B + Product C — 15% Off" doesn't sell. A bundle called "Trail Runner Recovery Kit" tells a story and creates desire. The LLM generates names that frame the bundle as a solution, not just a discount.
Bundles increase AOV mechanically. Every bundle sold is, by definition, a multi-item purchase. Even with a discount, the total order value is higher than a single-product purchase. This makes bundles one of the most reliable AOV levers in ecommerce.
Regenerate bundles quarterly. As purchase patterns shift with seasons, new product launches, and trends, the co-purchase graph changes. Fresh bundles capture current buying behavior instead of stale patterns.
Case Study: Pet Supplies Store
PawPath Pet Supply
Background: PawPath sells pet food, toys, grooming supplies, and accessories on WooCommerce (1,100 SKUs). They had manually created 8 bundles over the past year — none were selling well. Average order value was $47.
Implementation: Ran the AI Bundle Generator, which produced 14 bundles from their co-purchase data. The top performer: "New Puppy Starter Pack" (food + training treats + chew toy + cleanup bags) — a cluster the team hadn't considered as a bundle.
Key insight: The 12% bundle discount A/B tested against 18% showed nearly identical conversion rates but significantly higher total revenue. Shoppers were buying the bundles for convenience and curation, not just the discount — proving that the AI naming and product selection were the real conversion drivers.
Generate Bundles That Actually Sell
Let your co-purchase data and AI do the merchandising. Bundles in minutes, not weeks.
Get Started Free