Buyer behavior for procuring large, complex, customized items consists of initial phases where firms first specify the needed item(s) and pre-qualify vendors, following which an auction is used to choose the contractor. Despite the rigid rules and detailed specifications accompanying auctions, it is striking that procurement auctions in industries like information technology (IT) display very large differences between the initially agreed-to payment and the actual payments because of revisions negotiated during the execution phase. However, these revisions are ignored in the theoretical and empirical auction literature. After developing a mathematical model to specify how bidders accommodate post-auction modifications, I develop a method to take my model to a comprehensive dataset of IT procurement auctions. I find that the prospects of modifications lower bidders' latent costs, leading to more aggressive bids, especially by bidders without a previous contract with the buyer. To fix the magnitude of this effect, I consider a buyer who credibly commits to a no-modification policy, and find that such a commitment would increase bids by 27%, all else is equal. I also find that the size of this shadow of the future is larger for lump-sum bids. To fix this magnitude, I consider a buyer who switches the auction bid format from a lump-sum bid to a more flexible bid format (e.g., time and materials), and find that bids are lower by 16%, all else being equal. These large effects form the basis of my recommendations for improving procurement auctions. They also contribute to long-standing theory concerns studied in transaction cost economics. My findings support the Williamsonian critique of procurement auctions as a solution for the ex-post monopoly problem; my estimates demonstrate that ex post modifications remain an intrinsic aspect of procurement auctions. However, auctions remain a valuable procurement device for customized goods in complex, fast-moving environments, particularly when used with more flexible payment formats.