Product Knowledge Co-Pilot: Source-Grounded Product Data for Customer Service and Operations

A ~$300M global distributor whose catalog, inventory, pricing, and supplier records span NopCommerce and NetSuite, consolidated into an Azure data lake used by Customer Service and Operations.

A readiness assessment across seven stakeholders named product information management the highest-priority opportunity. The catalog was consolidated but not directly queryable, so answer speed tracked individual familiarity with the data.

Compose built the Product Knowledge Co-Pilot, a retrieval-augmented conversational interface deployed inside the client’s own Azure tenant: the data lake indexed into Azure AI Search, queries resolved through Azure OpenAI against retrieved context, source-grounding that refuses ungrounded answers, and a response audit trail.

• Product Knowledge Co-Pilot deployed to the production Azure tenant
• RAG index of the product catalog in Azure AI Search
• Technical Architecture Document with data lake schema mapping
• Process documentation of in-scope product data lookup workflows
• Functional specification: query taxonomy and persona-based use cases
• Hands-on user training for Customer Service and Operations

Average time to answer a product question fell from ~15 minutes to under 1 minute across Customer Service and Operations.

The Co-Pilot met every acceptance criterion on a jointly signed evaluation suite of 80–100 query pairs: at least 90% task completion, at least 95% correct refusal of out-of-scope queries, and a critical error rate under 2%.