momox

Pickers walk 40% less in warehouse

Momox is one of the biggest re-commerce player and the world's largest retailer of used goods on Amazon and third on eBay. Momox buys used books, media articles and clothing from private individuals at a fixed price. After a comprehensive quality check, the articles are resold via online shops as well as online marketplaces.

Challenge

momox has a huge capacity for storing bought books and media in their 90000 m^2 warehouse. The warehouse consists of 9 floors and can hold about 2.5M items at the same time. With an annual volume of up to 50M items sold, pickers were sent consolidated picklists printed from the terminal. These lists were still unsorted and it was up to the picker to make the best combinations and the optimal route. Given the complexity of TSP, this is an impossible challenge.

Approach

After a detailed analysis of the picking process, Solvice identified 2 key areas where algorithms could optimise the processes. First, all incoming orders for the next day are grouped into waves of orders that have similar delivery time and similar location & zone. During these waves, picklists are generated. We discovered that the load of a picklist had too much variability. Distributing this load optimally ensured that warehouse managers were able to have a much smoother operation.

Solution

After the proof of concept of warehouse picking optimisation, momox development team started integrating the OnPick API. In order for OnPick to be able to know where every item is all the time, it should be aware of the layouts of every floor in the warehouse. That means uploading (and maintaining) 9 graphs to the API.

Excel visualisation of the aisles in a zone in the warehouse

There is an excel file that keeps this information on aisle length up-to-date at momox side (left).

From this file, a graph with actual paths can be extracted and be used when calculating distances from any location to any other location in the warehouse.

Based on the layout, momox sends 100 solve requests to the OnPick API in parallel in order to obtain the optimised routing for 2500 picks per request.

Compared to their homegrown solution, these picklists result now in up to 40% less travel time. This means that the cost per pick decreased with up to 7%.

Get more insights