Data-driven warehouse optimization: deploying skills of order pickers

Batching orders and routing order pickers is a commonly studied problem in many picker-to-parts warehouses. The impact of individual differences in picking skills on performance has received little attention. In this paper, we show that taking into account differences in the skills of individual pickers when assigning work has a substantial effect on total batch execution time and picker productivity. We demonstrate this for the case of a Finnish retailer. First, using time-stamped picking data, multilevel modeling is used to forecast batch execution times for individual pickers by modeling individual skills of pickers. Next, these forecasts are used to minimize total batch execution time, by assigning the right picker to the right order batch. We formulate the problem as a joint order batching and generalized assignment model, and solve it with an Adaptive Large Neighborhood Search algorithm. For the sample company, we are able to improve state-of-the-art batching and routing methods by almost 10% taking skill differences among pickers into account and minimizing the sum of total order processing time. Compared to assigning order batches to pickers only based on individual picker productivity, savings of 6% in total time are achieved. Keywords— Logistics, Order picking, Analytics, Combinatorial optimization, Data driven modeling