The real world challenges of incomplete information access and bounded computational resources in supply chain management motivate us to propose a bottom up approach to supply chain intelligence, built over a widely used reactive card-based replenishments system (Kanban). The rationale is to use agent technology to improve the performance of the traditional Kanban system while maintaining its recognized usability. Instead of optimizing a system utility function, we encode the system goal in desired behaviours of individual agents that reason about their own behaviours in the local context. This paper discusses a rigorous framework for evaluation of the proposal based on the concept of benchmarking. Preliminary results from these simulations show remarkable improvements over the traditional system. Furthermore, use of the benchmarking framework gives confidence that these results will translate into real performance gains in practical implementations.