Using Unsupervised Learning for Data-driven Procurement Demand Aggregation

Procurement is an essential operation of every organization regardless its size or domain. As such, aggregating the demands could lead to better value-for-money due to: (1) lower bulk prices; (2) larger vendor tendering; (3) shipping and handling fees; (4) reduced legal administration overheads. This paper describes our experience in developing AI solution for demand aggregation deploying it A*STAR, a large governmental research Singapore with procurement expenditure scale hundreds millions dollars annually. We formulate problem using bipartite graph model depicting relationship between procured items target vendors, show that identifying maximal edge bicliques within would reveal potential patterns. propose unsupervised learning methodology efficiently mining such novel Monte Carlo subspace clustering approach. Based on this, proof-of-concept prototype was developed tested end users during 2017, later trialed iteratively refined, before being rolled out 2019. The final performance 71% past cases transformed into tenders correctly detected by engine; new opportunities pointed engine 81% were deemed useful tender contracts future. Additionally, per each valid pattern identified, achieved 100% precision (all aggregated purchase orders correct), 79% recall (the identified should have been put tenders). Overall, cost savings from true positive spotted so far are estimated be S$7 million