A fast solver can return a proven optimum for a mathematical model, while the policy built around that model makes disappointing decisions in operation. A policy that succeeds in isolation can also underperform when other parts of the organization change how they act. In a new long-form article, Adam DeJans Jr. examines this gap between mathematical optimization performance, policy performance, and the performance of a complete decision system.
The article develops the argument through a manufacturer-to-warehouse-to-marketplace supply chain, where ordering, allocation, inventory positioning, and replenishment decisions interact across stages and across time. Lead times delay the effects of upstream actions, while temporal correlations in demand and operational feedback change the states future policies encounter. These dependencies mean that evaluating each component independently is necessary for diagnosis, but insufficient for judging the deployed system. The essay explains why controlled policy experiments, joint simulation, consistent economic accounting, and out-of-sample validation all belong in the same scientific workflow.
Read the full article: The Next Frontier in Operations Research Is Decision Systems, Not Solvers. The complete essay includes a practical framework for system-level evaluation and closes with information about consulting through Bit Bros LLC.
Join the discussion
Sign in with GitHub to leave a comment.