Bottino, Ferrero, Dosio and Beneventano’s Retrieval Is Not Enough argues that the central limitation of organisational AI is not retrieval fidelity but epistemic fidelity: the capacity to distinguish decisions from hypotheses, evidence from observations, contradictions from settled claims, and unresolved questions from usable knowledge. The paper challenges the prevailing assumption that better embeddings, longer contexts or denser retrieval pipelines can solve organisational reasoning failures; such systems may retrieve relevant documents while remaining unable to interpret their epistemic status. Its proposed framework, OIDA, restructures organisational memory into typed Knowledge Objects with epistemic classes, importance scores, class-specific decay and signed contradiction edges. The most original case-study mechanism is QUESTION-as-modelled-ignorance, whereby unresolved questions do not decay into irrelevance but gain urgency over time, making organisational ignorance computable and operationally visible. The Knowledge Gravity Engine further introduces deterministic score maintenance, contradiction suppression and memory-zone allocation, transforming knowledge from a flat archive into a dynamic epistemic substrate. Although the authors report a pilot comparison in which their OIDA RAG condition trails a full-context baseline on composite quality, they identify token-budget disparity as the decisive confound and isolate a cleaner result: explicit ignorance declarations appear consistently in OIDA outputs. The paper’s importance lies in its conceptual inversion of current AI practice: organisations should not merely improve what agents retrieve, but redesign what knowledge is before retrieval begins.