Saving Millions in Government Procurement Through Data Science and Market Design *

Abstract

Framework agreements (FAs) are procurement mechanisms used in private and public organizations by which a central procurement agency selects an assortment of products, typically through auctions, and then affiliated organizations can purchase from the selected assortment as needs arise. In Chile’s central procurement agency (ChileCompra), FAs accounted for 23% of the procurement expenditures during 2018-19. However, descriptive analysis of purchase transaction data suggests that some FAs exhibited low levels of competition in the auctions used to select the suppliers, which could potentially result in larger government expenditures. We collaborated with ChileCompra to redesign FAs to enhance competition introducing two important changes: (i) standardize the product catalogue using natural language processing algorithms; (ii) use this product standardization to induce more competition in the auctions to select suppliers. These changes were implemented through an experimental approach in a redesigned Food FA to measure its impact, showing that inducing more intense competition in the auction stage reduced transaction prices by 8%. This pilot study ultimately led ChileCompra to implement a similar design in all of its FAs, and many of the improvements in the design of the FAs were included in the new regulation on government purchases. If we were to extrapolate the savings from our pilot re-design to all of these FAs, the total savings would amount to around $74 million in 2022.