IKEA – Smart allocation of online orders #SWI2025

IKEA produces and distributes home furniture in 31 countries from more than 482 stores worldwide, and it does so with a shared and inspiring vision: to create a better everyday life for many people. Therefore, IKEA continuously focuses on improving its services and becoming more accessible to many people. To meet their expectations, we are investing in new and existing channels, innovative store formats, and digital platforms.

Problem description
The advent of e-commerce greatly accelerated during the COVID-19 pandemic, has posed novel challenges in how IKEA manages its logistic processes, particularly order fulfillment. As often in logistics, low costs and sustainability go hand in hand, making the order allocation problem particularly important.

Consumers continuously place orders at IKEA’s websites, to which IKEA’s servers respond with a decision on the best location to pick all items in the order, the earliest available delivery date, and the price. This decision is highly complex because it depends on the available stock levels at different sites, the order picking costs, and transportation costs, which depend on the total volume, total weight, and number of items in the order. IKEA’s data scientists have found useful solutions for this problem based on mathematical modeling of all costs in terms of mixed integer programming.

Nevertheless, an important limitation of the current solution is that it considers every incoming order sequentially, one at a time (i.e., without any consideration for future orders), probably resulting in sub-optimal performance across time and across sites. For example, the current approach could result in the allocation of a big order to a distant picking site (because picking capacity is exhausted in nearby sites), leading to high transportation costs that could perhaps have been avoided when a small order processed earlier would have been allocated differently.

The central challenge of the case we are presenting at the SWI is to develop a strategy to assign orders to picking sites in such a way that upcoming (unknown) orders are anticipated, and the expected total costs are minimized. To address this challenge, you will be given a sample of orders with a realistic distribution, a simplified cost structure, and a picking capacity for each site. Not all articles will be present on each site, but if they are, you can assume sufficient inventory. You can also assume all orders can be transported.

Expected results
We would consider the case successful if, by the end of the workshop, we have gained new insights into the best way to anticipate future orders in the assignment problem and how performance across time can be optimized. How should the distribution of future order best be modeled and incorporated into the assignment problem? What are the key indicators of sub-optimal allocation when considering one order at a time without future orders? How do we deal with the large variety of articles and article combinations in the orders? It would be ideal if these insights were illustrated with some simulations where different approaches are compared.