Working in collaboration with a large online retailer, our goal is to develop an operational solution to the problem of order release control for its highest volume and most automated warehouse pick-to-ship process. This problem consists of dynamically varying the rate at which new picking orders are released into this process in order to achieve a high throughput while mitigating the risk of congestion-induced collapse (gridlock). We describe a queueing model of this complex process with validated predictive accuracy against actual historical data, and develop a numerical approximate solution method for an associated constrained dynamic program, which we implemented to compute our proposed policy.

Simulation experiments suggest that an implementation of our computed policy along with an increase in staffed packers by 25 percent could increase process throughput by 10 percent. They also shed light on why our proposed policy outperforms other simple release policies such as constant release rate and CONWIPin this setting, in terms of both steady-state throughput and robustness to transient disruptions..