Useful product bundles contain items that work together, are available and match the shopper need. Semantic similarity can find related descriptions, but two similar products may be alternatives rather than complementary purchases. Zubby can also use the products bought together in the connected store order history.
Read a bundle recommendation as a candidate
The current bundle tool first looks for shared collections, then product type or vendor; if none exist, it returns no suggestions. Those relationships are candidates, not compatibility proof. Suppose a sample store sells a camera body, compatible battery and several lenses. Products appearing in the same historical orders can provide a starting point, but co-occurrence does not prove that a particular lens fits the current camera. Review product compatibility, selected variants, availability and merchant rules before showing a confident recommendation.
Test the recommendation and the cart separately
- Choose an anchor product in a test store and inspect the recommended product IDs and explanations.
- Include an unavailable companion item and a blocked product to verify they are not presented as ready to buy.
- Check any compatibility claim against an explicit catalog attribute or approved knowledge entry. If it is missing, ask for clarification.
- Add each recommended item to the test cart and verify variant, quantity, price and currency. A suggested bundle is not automatically a special bundle price.
Handle sparse history honestly
A new store may not have enough relevant completed orders to support co-occurrence suggestions. Returning no bundle is preferable to implying a measured pattern from insufficient data. Keep the fallback clear and do not label an invented pairing as frequently bought together.
Measure commercial value without inventing a result
Review bundle impressions, additions, completed orders and returns over a stated period. Higher average order value alone can reflect changes in traffic or promotions. Check gross margin after discounts and avoid crediting every item to the assistant merely because it appeared in an order. Use the published attribution rule and a controlled comparison where the integration supports it.
Before enabling it for shoppers
Sync a representative catalog, review product rules and test a few successful and unsuccessful pairings. Merchant oversight remains important for regulated products and technical compatibility. This article describes product behavior and a sample test plan; it does not publish a customer conversion result.