AI in the Cold Chain, A CSW–Datex-Endeavor Labs Case Study
Using AI and machine learning to produce effective predictive models for labor scheduling and throughput
About the Client
The largest independent cold storage warehouse provider in Wisconsin, Central Storage & Warehouse (CSW) provides a comprehensive array of cold storage solutions, including blast freezing, tempering, refrigerated and frozen storage as well as technology-enabled services
The Challenge
CSW needed a pro-active approach to forecast labor availability on an hourly basis, schedule loads to mitigate peaks, and prioritize work to maximize its workforce across daily shifts.
The Solution
Endeavor Labs utilized two years of data gathered from Datex Footprint WMS to develop machine learning models to present live predictions as part of an AI-driven labor optimization strategy that enhanced labor efficiency and responsiveness.
About CSW
A Legacy of Over 75 Years of Experience
As the largest independent cold storage warehouse provider in Wisconsin, Central Storage & Warehouse (CSW) has a legacy of over 75 years of experience serving food and beverage manufacturers and life sciences businesses. The premier provider of third party refrigerated warehousing services in the Midwest and a top 15 cold storage warehouse provider in North America, CSW has 5 warehouse locations throughout Wisconsin, each with frozen and refrigerated capabilities.
A growth-oriented and customer-centric organization, CSW focuses on continuously enhancing their capabilities. It continues to expand its reach and to ensure fulfillment of its longstanding commitment to its customers, community and employees. Personalized, exceptional service at each warehouse ensures that each customer receives consistent excellence and tailored solutions that it deserves.
With over 100,000 pallet positions, CSW provides a comprehensive array of cold storage solutions for diverse needs, including blast freezing, tempering, technology-enabled services, refrigerated and frozen storage to USDA compliance. CSW takes great pride in being able to understand and adapt to the unique business needs of each of its customers, from small-scale producers to industry giants.
“We pride ourselves on being flexible, reliable, and responsive to our customers” said Co-CEO Hill Hamrick. “We never want to tell our customers-no, we can’t. We can’t do that. And so especially when technology plays a role in our ability to deliver whatever services that that individual customer requires, we must be able to move fast and move collaboratively both on the outside and the IT side.”
Warehouse Locations
Pallet Positions
The Challenge
A Common Problem Among Most Warehouse Operations
Facing the need to scale operations to support growing customer demand, CSW leadership recognized that they needed to take a more proactive and data-driven approach to manage the inherent volatility of warehouse operations and scheduling.
CSW experienced challenges with labor, which is a common problem among most warehouse operations. This was especially true with respect to aligning labor with variable throughput needs on an hourly, daily, weekly basis. This led to manual adjustments which resulted in labor mismatches. Overall productivity and service levels were then impacted.
CSW identified several key needs:
- The ability to forecast labor availability and throughput on an hourly basis
- The need to schedule loads to mitigate peaks while ensuring fulfillment of customer service commitments
- The ability to prioritize work to maximize labor resources across daily shifts
To ensure that these needs were met, first CSW needed the ability to predict labor requirements on a per load basis to ensure hourly throughput expectations. Using real time data, CSW wanted to be able to optimize staffing, scheduling, and warehouse performance while also leveraging predictions to set accurate and reliable labor standards.
To solve this challenge, Hill Hamrick turned to Datex Footprint WMS for unified data within its warehouse management software as well as Endeavor Labs for collaboration with its AI and machine learning tools.
“Our goal was to use the accurate data we already collected within Footprint WMS to drive operational efficiency,” said Hamrick. “By partnering with Endeavor Labs and Datex, we were able to use machine learning to predict labor requirements and improve productivity across the board.”
CSW wanted to implement a machine learning-based predictive planning tool to automate labor forecasting down to the hour.
The Solution
Data-driven Solution with Footprint WMS
CSW recognized that it had access to a significant amount of data that could be used to develop a data-driven solution. Data was collected from Footprint WMS as well as through integrations. The information was then leveraged to optimize labor allocation based on real-time demand patterns to improve operational efficiency and manage costs more effectively.
Footprint WMS provides a comprehensive, immutable audit trail, often used for documenting regulatory compliance. This provided both real time and historical data from which each employee’s individual picking and loading time could be determined. Endeavor Labs developed a machine learning model and trained it on comprehensive CSW data.
“Using advanced analytics becomes pretty straightforward if your data is consistent and well structured. Datex provided a great foundation for our work,and helped us make an impact quickly” explained Nathan Gould, Founder/CEO of Endeavor Labs.
The project started out with over two years of data from Footprint WMS, specifically involving shipment details as well as timestamped warehouse activities such as picking and loading tasks to define model inputs and outputs. The goal was to be able to predict shipment labor requirements based on attributes known ahead of time. Because there is considerable variation involved with the way food is shipped, these variations and labor needs were incorporated in the data, and the machine learning model picked this up. Other factors also needed to be considered such as the fact that not all goods are on pallets and others needed packaging.
Driving forces that required the most consideration for how much labor effort would be required were determined to be the account and the specific destination for the main shipment. This data was crucial in developing the model which looks at upcoming shipments to estimate the labor and materials needed to perform that particular work. Next, the model projects the number of hours for which labor will be needed for loading.
The project found some aspects challenging. One of these challenges involved having to navigate inconsistent account names that had originated via EDI transactions. For example, the same account name for one specific warehouse location may have five or more different account names. This caused concern for how the Endeavor Labs’ algorithm would learn to disregard minor naming variations.
To solve this problem, Endeavor Labs used Byte Pair Encoding (BPE), a data compression technique widely used in natural language processing. THIS helped to create a fixed-size vocabulary to efficiently handle rare or out-of-vocabulary words. This eliminated the need to depend on a huge list of predefined words to create an efficient vocabulary.
Using BTE, Endeavor Labs was able to create a numeric representation of each account name to enable the machine learning algorithm to focus on key aspects of the account name.
The second challenge was the lack of direct measurements of historical labor requirements as the only available information was timestamped warehouse activity logs.
To resolve this issue, each employee’s individual picking and loading time were estimated, then summed and outliers were removed. The Endeavor Lab trained model was then able to explain over half the variation in shipment times.
After taking the initial CSW expectations, a modeling technique known as Gradient Boosted Decision Trees (GBDT) was applied. This machine learning technique combines the strength of decision trees (a model that splits data into branches based on specific features or attributes to make predictions) with the concept of boosting (building multiple models sequentially, so that each new model tries to correct the errors of the previous model). Using this technique provided clarity on feature importance with respect to that of other key metrics.
Live predictions were then delivered via Footprint WMS, initially as an hourly batch job. Subsequently, a weekly job was added to retrain the model using the latest data from Footprint WMS. Reports are now automatically emailed to subscribers and include the accuracy metrics from the previous week. CSW is now able to see the latest prediction for each appointment as well as the actual amount of labor for each shipment.
In addition, an algorithm can capture the difference of a shipment going to two different retailers for the same customer. Another looks at activities with a similar number of expected pallets and captures the difference between picking one SKU and multiple SKUs across multiple aisles.
Other detailed reports are provided to CSW including recaps of shift activity, pallets turned per labor resource, and 24-hour outbound shipment burndown.
“In this effort we have seen that when warehouse operators leverage their creativity and ambition and collaborate with knowledgeable technology providers, challenges can be resolved and remarkable results can be produced” said Frank Jewell, Chief Revenue Officer. “We look forward to more collaboration with these partners in the future.
Learn About Footprint WMS
The Conclusion and Results
Through their collaboration with Datex and Endeavor Labs, CSW successfully implemented an AI-driven labor optimization strategy that improved labor efficiency and responsiveness. This case illustrates the transformative potential of combining modern WMS platforms with advanced analytics, equipping cold storage providers with the insights they need to stay competitive in a technology-driven landscape.
This project specifically focused on picking and loading labor. It enables the CSW team to review upcoming shipments, examine their staffing and confirm that the correct number of people are planned to do the work. Datex collaborated with CSW and Endeavor Labs to produce easy-to-read visualizations in a hub for quick access by CSW team members. A process was created by which Datex connects to the data model and provides updates to refresh the model continually.
The effort continues to be a work in progress. CSW is actively using this to plan their labor. Rather than taking a static, reactive approach to labor scheduling, CSW is now better positioned to take a more dynamic, proactive approach and has found that the labor model has been most accurate when compared to the original model developed for this project.
With real time data, CSW is optimizing staffing, scheduling, and warehouse performance while also leveraging predictions to set accurate and reliable labor standards. With the support of this data solution, CSW continues to train the model to generate relevant and reliable predictions even as customers’ requirements change, and labor needs fluctuate.