Shopware AI Cross-Selling: LLM-Determined Product Rules
AI-generated cross-selling for a German plant nursery's Shopware 6 shop. An LLM proposes accessory and companion-plant rules per category, verified live.
The Challenge
The ERP export driving this Shopware 6 plant nursery shop carries pure product and category data. No accessory relationships, no "goes well with" field, nothing to import. The nursery wanted cross-sell suggestions anyway: fertilizer and soil aids next to the plants that need them, companion plants next to the ones they pair with in a garden bed.
Critical Pain Points
- No source data existed anywhere; every accessory and companion pairing had to be inferred, not imported
- An existing order-history "customers also bought" slider only surfaces pairs that have already sold together, useless for a newly imported category or an accessory nobody has bought alongside a plant yet
- The catalog runs into the thousands of parent products; judging each one individually for cross-sell would produce more one-off pairings than a human could realistically review before anything went live
- A wrong pairing is not cosmetic: the wrong fertilizer grade linked to an acid-loving plant is bad horticultural advice sold to a paying customer
- Whatever generated the suggestions had to be provably incapable of recommending a product that doesn't exist in the catalog
Cross-sell rules had to be generated from nothing, at a scale a human reviewer could actually get through, with a hard guarantee the system could never point a customer at a product that isn't real.
The Solution
I built an AI pipeline that reasons about cross-sell at the category level rather than product-by-product, with guardrails that make it structurally impossible for the AI to suggest a product outside the real catalog.
Reasoning scoped to stay reviewable
The AI judges cross-sell fit one category at a time instead of one product at a time, turning what would be thousands of one-off pairings into a small, human-reviewable rule set that expands to every product in an approved category automatically.
Guaranteed to stay inside the real catalog
The pipeline is built so the AI can only ever point at a product that genuinely exists in the shop, or suggest nothing, closing off the single biggest risk of putting AI-generated suggestions in front of paying customers.
Two cross-sell types, one efficient pass
Physical accessories and companion plants get proposed together per category, keeping the AI cost proportional to the size of the category list, not the size of the catalog.
Applies and re-applies without breaking anything
Approved rules go live as native Shopware cross-selling groups that can be safely regenerated after every catalog update, with no duplicates and nothing created by hand ever overwritten.
Results & Business Impact
Verified against a live 210-product category
The pipeline generated, applied, and re-applied rules against a real 210-product plant category and its accessory options, with zero duplicate groups across repeated runs.
Zero hallucinated products
Every suggestion the AI produced pointed at a product that actually exists in the catalog, with no exceptions across the verification run.
Companion-plant reasoning holds up horticulturally
The system correctly proposed only shade- and acid-soil-appropriate companions for acid-loving plant categories, rejecting weaker candidates instead of forcing a pick just to fill a slot.
Native storefront integration, not a bolt-on widget
Approved rules render through Shopware's own cross-selling groups, inheriting the same variant quick-view dropdown and quick-buy behavior as every other slider on the product and cart pages.
Technologies Used
- Shopware 6
- PHP 8
- Symfony
- LLM API
- MySQL
- DAL
Cross-sell data your ERP doesn't have?
If your catalog has no source data for accessories or complementary products, an AI pipeline with the right guardrails can infer defensible rules at a scale a human can actually review. Let's talk through what that looks like for your shop.
Book a consultationRelated Case Studies
Dropshipping automation
Shopware 6 dropshipping automation for a German shop. Multi-supplier routing, real-time stock sync, and a supplier portal. No human in the order loop.
Read Case Study → AutomationAI reorder prediction
Shopware 6 plugin that predicts B2B reorder timing from order history and emails customers before they run out, with a pluggable AI layer and built-in licensing.
Read Case Study → AutomationShipping bin-packing engine
Shopware 6 cart processor that solves shipping cost as a bin-packing problem: mixed carts pack into the cheapest real combination of parcels and pallets, live at checkout.
Read Case Study →