Optimizing the picking process using an AI-based slotting engine

About client

Our client is a US-based company offering a group of tech implementation services for logistics, supply chain, and portfolio management. Being an Oracle Cloud WMS partner, they help businesses strategize inventory, order management, mobility, and supply chain processes by setting up the right software solutions.

Challenges

The client wanted to solve one of the biggest issues warehousing units face - higher costs and physical labor involved in the picking process. This is specifically a bigger challenge for large warehouse units handling multiple products and storage slots. 

If you consider the out-cycle, they involve a large number of workers for the picking, staging, and packing processes, where employees have to move around a lot. 

The client wanted to solve this challenge and reduce their in-warehouse movement by designing a slotting algorithm. This algorithm will discover a near-optimal placement for each product while loading, making it easier for the picking staff to collect them for order processing with the least physical effort.

This is highly useful for warehouse managers, considering their two important goals—optimizing product accessibility and labor productivity. 

But here are certain challenges they face while solving this problem by traditional means.

Too many products, Too many slots: Most large warehouses have the highest inventory capacities, handling n number of products which has to be stored in n number of slots across their huge unit. Every day, they receive products from vendors to be arranged in these slots. Similarly, they receive out-cycle orders on these products, which they have to move to the staging, pack, and send them. 

The dynamic change in the number and volume of products made it nearly impossible to identify the optimal placement strategy without knowing what orders they would receive for the day.

Picking policies changes every time: The picking policy is how the batched or individual orders are collected from different slotting units. Every warehouse has a different process for this and even within the same warehouse, the process is never constant. Some use technical equipment for picking, some have different employees to cover the vertical and horizontal distances involved, and so on. This varying factor is a challenge while optimizing the distance covered by the employee ideally.

Deciding what product can fit in a typical slot and how many: Every slot differs in size to accommodate products of varying dimensions. This means that the warehouse manager has to identify what products can fit in which slot and how many of them can be stored. Doing this at scale took long hours for them. This delayed loading also affected the picking process.

Other constraints: There are products that require cold inventory storage. Some products were brittle in nature and required special care and had to be stored in slots of lowered heights. Some product combinations couldn’t be stored together. Every warehouse has to deal with constraints like this which makes it difficult to find out the right slotting strategy for the product arrangement.

Unable to predict the order book ahead: Without knowing what order they might receive for shipping, they cannot place the products in the right slots or shift pre-existing products closer to slots near the staging area. 

Predicting the order book can also help them coordinate with vendors and re-slot products on time, so they can meet end-consumer demands.  

Solutions

Our goal is to minimize the physical load of workers involved in the order-picking process without compromising the delivery throughput of the warehouse. We have to identify a ‘configuration’ of SKU arrangements (placements, re-slotting) based on a predicted order book. 

With this huge number of products and slots, the placing and slotting combination can explode as there are exponential ‘configurations’ (this problem is proven to be NP-Hard). 

This slotting engine is designed assuming that the warehouse is rectangular or a combination of rectangles with partitions and vertical slots.

We have used combinatorial optimization techniques inspired by AI to solve this problem without exclusively relying on classical mathematical techniques.

Inputs given to the slotting engine

1. Warehouse landscape geography

2. Slot geography and geometry

3. Available inventory (empty slots) at the given point of time

4. Picking policies of the warehouse

5. Order book (or) historical data (in case this needs to be predicted).

Inferences generated by the slotting engine

1. Forecasting the future order based on the historical order data

2. If any products already in the slots can be re-grouped or moved closer to the slots near the staging area (profitable re-slots).

3. Discovering product affinities and combinations which customers usually order together. 

4. Generating a ‘configuration’ that’s considerate of the inventory constraints (not grouping products that cannot be placed together, etc).

Our slotting engine generates these suggestions on time to optimize the out cycle, reducing the physical distance covered by the picking staff. 

Conclusion

Modern warehouses are swamped with data that they can use to extract useful inferences and run AI-inspired applications. Slotting algorithms are one of them that tackle a complex problem, helping warehouse workers to achieve more with less physical movement in the warehouse. This will be time and money-saving for the company too and prevent worker fatigue. 

With vendors paying increased importance to on-time-order fulfillment, this slotting engine can help achieve high order throughput without worker overload.