Harsh Vora, Lead Data Scientist
Zachary Thorell, Business Data Analyst
Sandeep Punjari, Data Analyst 3
Irwin Castellino, Director of Data and Analytics
Deepak Arora, Vice President Corporate Strategy
Javier Diaz, Senior Analyst Quality Assurance Ops
Federico Gervasio, Industrial Engineer
Shannon Sarkees, Sustainability, Strategy & Digitization Manager
Tishlee Rivera, Business Intelligence and Analytics Director
Sudip Roy, Big Data Solutions Architect
Sanjay Khobragade, MLOps Architect
Country: United States
Crowley, founded in 1892, is a privately-held, U.S.-owned and operated logistics, government, marine, and energy solutions company headquartered in Jacksonville, Florida. Services are provided worldwide by four primary business units – Crowley Logistics, Crowley (Government) Solutions, Crowley Shipping and Crowley Fuels. Crowley owns, operates, and/or manages a fleet of more than 200 vessels, consisting of RO/RO (roll-on-roll-off) vessels, LO/LO (lift-on-lift-off) vessels, articulated tug-barges (ATBs), LNG-powered container/roll-on, roll-off ships (ConRos) and multipurpose tugboats and barges. Land-based facilities and equipment include port terminals, warehouses, tank farms, gas stations, office buildings, trucks, trailers, containers, chassis, cranes, and other specialized vehicles.
Crowley did not have a centralized platform to utilize data and machine learning for decision-making in our logistics business unit, where we face several fundamental issues:
A. Missed revenue from dummy bookings – Customers book extra slots for containers on container ships and eventually show up at port with fewer containers since no cancellation fees are enforced (industry standard).
B. Lack of demand forecast for each node – The availability of empty containers at the right nodes/ports in the supply chain is the key to meeting our customers’ demand. We did not have a historical and forecasted view of the demand for each container type, and between each origin and discharge node, which is key to enabling decision-making for empty container repositioning.
C. Late customs documentation – Improper or late customs documentation provided by customers resulted in offloaded containers residing at the port, costing the port time and space, and incremental efforts for planning fulfillment.
D. Unknown container weights – Each container ship has a maximum weight capacity. However, the weights of containers booked on a ship were only known once they are weighed at the port, resulting in last-minute planning for stowage (placement of containers on the ship) and accommodating weight constraints.
E. Lack of Carbon Footprint estimation – Our customers seek to estimate the carbon footprint of their supply chain. We did not have the technology and tools to automate and expose the calculations of the carbon footprint from container shipments.
F. Lack of Predictive Maintenance – Port equipment to load containers onto ships, trains and trucks are prone to failure due to extreme loads. Unplanned and immediate maintenance requests are disruptive and expensive.
G. Non-targeted Promotions – The marketing methods for logistics customers were a manual and subjective process. A data-driven methodology to predict customer churn can improve the targeting of marketing efforts, especially for non-contract customers.
As a 130-year-old company undergoing digital transformation, we seek to utilize predictive and prescriptive analytics with our business leadership to boost our revenue, customer experience, employee experience and sustainability efforts. We pioneer digital transformation in the supply chain industry through (1) centralization of data from our operational, commercial and sustainability data into a data warehouse, (2) utilization of singular platform (Dataiku) to develop predictive and prescriptive analytics that enables all personas through no-code, low-code, and full-code capabilities, and (3) democratization of data engineering and machine learning activities through employee upskilling programs.
Through Dataiku, we developed/are developing solutions to our focus areas:
A. Container sail/rollover model [in production] – We developed a classification model to predict the probability of show/no-show for each container booked on our ships, providing visibility of at-risk containers to our voyage planning team for improved decision making.
B. Demand Forecasting [in production] – We utilized Dataiku’s AutoML capabilities to forecast demand associated with each container type, between each load and discharge node, enabling strategic decisions around empty container repositioning on a weekly basis.
C. Customs documentation classification [in production] – We developed a classification model to predict the probability of improper/late documentation for each container, reducing manual work for our claims and customs department.
D. Predict container weights [in development] – We are developing a regression model to predict the weight of containers booked on our voyages prior to them arriving at the port, enabling improved voyage planning with constrained weights of booked containers.
E. Estimate Carbon Footprint [in development] – We are developing a methodology to dynamically calculate and serve the estimated carbon footprint as a service using Dataiku’s API capabilities.
F. Predictive Fleet Maintenance [in development] – We are developing an anomaly detection model to identify concerning signatures from sensors on port equipment to implement a recommender system for inspection, reducing unplanned maintenance.
G. Predict Customer Churn [in development] – We are developing a customer churn classification model to improve the targeting of our marketing and promotional efforts in our logistics business.
Classification model to predict Container sail/rollover – Expected revenue gain of $5000-$10000/week, and 10-15 hours of employee time saved per week.
Demand Forecasting model – Expected revenue gain of approximately $10000-$20000/week, and approximately 5 hours of employee time saved per week.
Customs documentation classification model – Expected cost savings of approximately $5000-$10000/week, and approximately 5 hours of employee time saved per week.
Regression model to predict container weights – Expected revenue gain of approximately $5000-$10000/week, and 5-10 hours of employee time saved per week.
Estimate Carbon Footprint – This will be rolled out in a new product offering that enables optimization of supply chains based on carbon footprint and will position Crowley as a sustainability leader in the supply chain industry. Two potential customers have been identified, potentially generating revenue within the first year.
Predictive Fleet Maintenance – Expected reduction in unplanned maintenance and last-minute planning of port equipment, and potential reduction in scheduled maintenance time. Potential cost saving tens to hundreds of thousands of dollars per year.
Predict Customer Churn – Improved promotional targeting, reduction in manual hours for marketing, and data-driven identification of at-risk customers are expected to enable superior customer service to at-risk customers, increasing customer retention.
In addition, Dataiku has also generated additional value at Crowley through the democratization of data analytics through upskilling and enablement of Crowley’s business analysts.
Prior to Dataiku, each department worked in a silo utilizing disparate ETL, analysis, and reporting tools that did not integrate well. Dataiku provides a centralized, end-to-end platform for business analysts, data engineers, and data scientists to work together on analytics use cases.
Another significant value addition comes from the interactive visual interface and a great suite of AutoML models provided by Dataiku, enabling data analysts to design predictive and prescriptive models. For the MLOps team, Dataiku provides a seamless manner of registering and deploying models to production. The deployer enables the necessary governance checkpoints and the inbuilt drift monitoring, metrics and checks enable the development of appropriate post-production alert systems.
Finally, Dataiku simplifies the infrastructure needs of a maturing company. Our compute needs are always changing/increasing, and Dataiku’s Fleet Manager enables seamless scaling of servers and Kubernetes clusters. Due to the popularity of data science workflows developed in Dataiku at Crowley, the tool has got increasing interest from data engineering teams that are exploring the ETL functionalities that Dataiku provides, especially through seamless integration with Snowflake.