However, customers rarely search by part number; instead, they phrase their enquiries in everyday language – for example: “I need a filling system for viscous media for use in a cleanroom.” Traditional shop filters fail to cope with such free-text queries. Yet even a pure AI vector search often falls short, because in B2B, precise technical and commercial specifications must be adhered to.
The solution: hybrid RAG with structured metadata
A Retrieval-Augmented Generation (RAG) system links unstructured documents (such as manuals) with structured product data from ERP or PIM systems. To ensure this works reliably, the product data is divided into two parts:
- Free-text chunks: Continuous text and descriptions are fed into the vector search so that the AI can understand the general query.
- Metadata as hard filters: Technical specifications, prices, ISO certifications or delivery times are stored as structured metadata and serve as pre-filters.
How the system caters to different target groups
An upstream analysis identifies who is searching based on the wording of the query – and provides tailored responses for different roles within the company:
1. The engineer is looking for technical parameters
Customer: “We are filling high-viscosity cream (approx. 12 Pa·s) in ISO Class 7 cleanrooms. Is the Fill-Master V3 suitable?”
System: Yes. Rated up to 15 Pa·s, certified to ISO Class 7 and equipped with a modern standard PLC. Qualification reports are available as PDFs.
2. The procurement manager checks commercial details
Customer: “Second-hand equipment carries risks. What arguments are there against buying new?”
System: 12-month warranty, €45,000 net instead of the new price of €110,000 (59% saving) and a delivery time of just 5 working days, compared with waiting times of several months for new systems.
Conclusion
In B2B e-commerce, text generation alone is not enough. Only when unstructured descriptions are combined with hard metadata does a reliable digital advisory process emerge, providing precise answers to both technicians and buyers.
Download the detailed white paper here: rag_metadata_b2b_ecommerce_v3