Machine learning for Warehouse and Storage

Warehouse and storage operations are key component in most production as well as retail businesses. Optimal warehouse operation saves money on working capital, fixed assets, rents, etc. On the contrary, bad warehouse operations require high working capital, high volumes of storage and lead to low material turnover. There have been numerous systems created and applied to warehouse operations in order to achieve optimal performance. Perhaps the most famous one being just-in-time systems, which as hard as it is to believe has existed for over 50 years. The concept being that deliveries are made just when they are needed, so inventory can move as quickly as possible in the production process. The finished product then goes to its customer with the same speed as the material was handled. Customer orders are matched with supplier orders perfectly, so that maximum efficiency can be achieved. In businesses like car manufacturing (where just-in-time originates from), one car can have over 10 000 components so efficient order management and warehouse planning are essential. To aim at reducing costs and improve efficiency of operation, just-in-time techniques are widely employed.

Today’s interconnected software systems make it even easier to follow through the order management process from just-in-time manufacturing and apply it in other areas – e.g retail, wholesale, etc. A typical supermarket chain for example may not order its product in any other way but automatically (with minimum human interaction), practically directly from the software that manages their business (typically high-end ERP system). Such systems can easily manage over 50 000 items and their deliveries in multiple locations. The system can check for minimum quantities, plan purchase orders, calculate pricing, estimate delivery time, optimize delivery locations, but possibly even more importantly, communicate with supplier systems through a suitable interface. EDI systems are widely used to link suppliers and customers, so the supermarket chain can export its purchase orders and send them automatically to their suppliers. The supplier on the other hand would process the order in similar way in their ERP system. Then the supplier will confirm products deliveries, lot numbers, expiration dates, delivery time, etc. – again through EDI system. The supermarkets will know when to expect the delivery and when arrived – will check it to the warehouse typically through a highly automated process. The result is fast deliveries, optimal warehouse and efficient product management.

So after reading this, you may ask: what is there to improve in this process? Why do we need machine learning algorithms when the currently employed techniques, refined for over half a century, have proven to be quite effective as is. Here are some areas where AI techniques can expand beyond what we currently can do and improve further warehouse and storage performance:

  • Demand forecasting. While ERP systems are great at managing orders throughout the entire order lifecyle and the entire supply chain, what they are not as good at is planning for the orders that will come. Demand planning requires a different set of techniques and machine learning offers exciting new possibilities for that. With their ability to incorporate a vast amount of inputs, analyze them with a variety of techniques and arrive to meaningful conclusions, machine learning can help demand planning substantially. Importantly you may incorporate parameters and external data sources in the analysis which cannot be included in typical statistical or mathematical solutions.
  • Design optimization. While sizing a warehouse may not always be that difficult, optimal internal design and optimal routing inside the warehouse are much more complex tasks. Having a product in the warehouse, knowing where it is and being able to handle it is enough to get the job done, but the speed at which it is done and the costs involved are quite a different story. Various techniques, including machine learning techniques, could be used to address both organization and the picking process, so that more orders can be processed with less people and at a lower cost.
  • Automated warehouses. While warehouses have existed for a long time, machine learning backed robots are now revolutionizing how we handle them entirely. You can get a completely robotized warehouse where robots are responsible for picking, moving and storing all goods inside. No human presence is typically required or needed inside the warehouse area and humans are only responsible for moving goods to and from the warehouse. Alternatively, robots can be used selectively for handling specific tasks – for example palletizing. In such cases, robots add on top of existing human workforce inside the warehouse, augmenting and replacing tedious and repetitive tasks. In both cases machine learning techniques could be used to improve performance and efficiency, which is good news for an area, that has been focused on cost reduction for a long time.