AI shopping assistants increasingly employ agent-based retrieval, combining lexical search, structured filtering, and LLM-mediated selection. However, effective retrieval often requires knowledge beyond textual matching. This exploratory single-case study of an Indonesian grocery e-commerce assistant triangulates 279 observations, 52 failure traces, 74 practitioner-reported defects, a schema audit, and four practitioner interviews. The analysis identifies six retrieval-critical product knowledge dimensions and shows that knowledge externalization fails at two distinct representational layers. At the product data-model layer, essential fields—such as allergens, dietary constraints, and age suitability—were absent and remained missing despite architectural changes. At the retrieval-schema layer, available knowledge could not reach candidate sets: structured filters were present in only 29% of calls, and even correctly invoked filters frequently returned empty sets from non-empty pools. Addressing these layer-specific failures, the study proposes a structured knowledge framework based on the knowledge management process cycle. It positions GraphRAG as a future direction to enable hybrid structured retrieval for candidate formation and improve eligibility, substitution, and context-aware ranking.