Reducing Order Picking Time by 36% with Machine Learning-Powered Warehouse Optimization

Customer

A U.S.-based distribution center supplying a network of retail stores managed approximately 11,000 active products and processed 4.2 million transactions over 18 months.

As purchasing patterns changed, its manually maintained warehouse layout became increasingly inefficient. Frequently ordered products were often stored in suboptimal locations, increasing picker travel distance and slowing order fulfillment.

Outcome

Thanks to the collaboration with HQSoftware, the distribution center transformed its warehouse operations significantly. The system replaced manual layout planning with an automated optimization process that is continuously updating.

Product placement is now dynamically adjusted based on demand forecasts and purchasing behavior patterns. Products are automatically organized into 47 stable clusters, and zone assignments are recalculated as demand shifts, without requiring manual intervention from analysts or warehouse planners.

The approach was validated through a 6-week live A/B test comparing a control warehouse with a test warehouse operating under the ML-driven layout system.

As a result, the warehouse evolved from a manually managed operation into a self-optimizing environment that continuously adapts product placement to changing demand patterns. The solution can easily be scaled over the long term and applied across facilities with large product catalogs and seasonal demand fluctuations.

Solution

The HQSoftware team built a ML-driven system that determines optimal physical placement of each product in the warehouse based on two core signals:

• Demand-based zoning — frequently requested items are placed closer to packing and shipping areas to minimize walking distance, while low-frequency items are placed in more remote zones.
• Co-location of frequently co-purchased items — products that often appear together in the same order are assigned to neighboring zones to speed up multi-item picking.

These signals are combined in an optimization layer that produces a valid warehouse layout while respecting real operational constraints such as storage capacity, product storage type, and zone limitations. The system fully replaces manual layout planning.

How It Works

1. Data collection

The system aggregates 18 months of transaction history (~4.2M orders), a catalog of ~11,000 SKUs, detailed order composition data, warehouse layout metadata (zones, storage cells, coordinates, and storage types), seasonal and calendar effects, and historical picker route data.

2. Demand prediction and zoning

For each SKU, the system forecasts weekly demand over a 4-week horizon. These forecasts are converted into a ranking that determines product proximity to high-traffic warehouse zones. If an SKU changes its relative ranking by two or more zones, the system triggers a relocation update, ensuring that the layout continuously reflects current demand dynamics.

Looking for similar solutions or something unique to your needs?

Contact us today! We’re happy to explore your needs!

Victoria Rokash
Business Development Manager

Get in touch

3. Product affinity modeling

The system analyzes purchasing behavior to identify products that are frequently ordered together. Based on these patterns, related items are grouped into stable product categories and assigned to neighboring warehouse zones.

This approach reduces the time required to assemble multi-item orders and helps maintain efficient product placement as customer preferences evolve. Product groups are reviewed and updated regularly to reflect changing purchasing trends.

4. Warehouse layout optimization

The final stage combines demand forecasts and product groupings to generate an optimal warehouse layout. When assigning products to storage locations, the system takes into account operational constraints such as available space, storage requirements, and zone capacity. As a result, warehouse staff receive an automatically generated placement plan that maximizes picking efficiency while requiring no manual layout planning

Team

• Project manager / Business analyst
• ML engineer / Data engineer
• Full-stack developer

Learn more about our Artificial Intelligence development services.