Assuring generative AI in procurement decision support : a systematic mapping review and evidence-to-use framework

Abstract

Generative artificial intelligence (GenAI) is increasingly pro-posed for procurement decision support, but task performance alone does not establish decision readiness. This systematic mapping review searched Scopus, Web of Science and IEEE Xplore on 18 September 2026, supplemented by targeted searching and citation chasing. Of 181 identified records, 58 studies were included and assessed from full text, comprising 57 system studies and one study of practitioner acceptance. Task accuracy was measured in 46 systems, whereas grounding was measured in four and traceability in one. No measurement of calibration, confidentiality or bidder-level equal treatment was identified, and no study reported operational deployment with monitored outcomes. These findings indicate uneven coverage of reported assurance evidence, rather than demonstrated system failure. An evidence-touse framework translates the identified gaps into procurementspecific tests across eight dimensions, distinguishing four evidence states and three use classes. A conjunctive permission rule requires satisfaction of every assurance dimension applicable to the proposed use; operational value is assessed separately. The contribution is an evaluation-centred map and a structured basis for specifying procurement assurance requirements. The framework requires empirical validation and proportionate, use-specific acceptance thresholds.