FedEx Bolsters Financial Services with AI and Azure Synapse Analytics

One of the biggest challenges FedEx has faced in recent years has been keeping up with fast growth. This acceleration brought new challenges and needs for the organization, especially within Revenue Services. Large amounts of data were becoming difficult to manage, and it was clear that Revenue Services within FedEx was becoming overwhelmed. The Enterprise Business Strategies team knew they could easily leverage technology to create more efficient solutions. To do so, they would need to retire their current legacy systems and upgrade to a new infrastructure.

Investing in Azure Synapse Analytics to match company growth

The Enterprise Business Strategies team wanted to ensure that the new tooling was not only cost-effective but offered scalability and performance when handling large data sets. As a package and freight delivery company, FedEx processes huge volumes of data to run effectively, making an agile and reliable solution critical. The flexibility and sustainability of a new enterprise solution would be key for an effective and successful outcome.

Jack Wong, Senior Manager of Enterprise Business Strategies at FedEx, leads four teams to help tackle this revitalization of the Financial Services infrastructure: Application Development, Dashboard and Report Development, Prediction Model Development, and Infrastructure.

The Enterprise Business Strategies team wanted to lean into data modelling to better understand their breadth of data and make more informed decisions across the Revenue Services division. After running a POC and evaluating different competitors, Azure Synapse Analytics was selected to support their analytics and data warehousing needs. The scalability and affordability of the tool made Synapse a clear standout.

“Evaluating the functions and the features and the price, as well as the value you get out of [Azure Synapse Analytics] and the integration that is in place using the Microsoft technology stack, Synapse just won out,” explains Ronald Nelson, Manager of the Infrastructure team.

Because of FedEx’s longstanding relationship with Microsoft, the Enterprise Business Strategies team also wanted to be within the same ecosystem as other bigger enterprise solutions within FedEx, especially regarding data storage. Plus, with the heavy use of Azure in other parts of FedEx, the transition to this new solution would be seamless.

Creating a foundation around Azure Synapse Analytics

After confirming the decision to move forward with Synapse, the team set out to build out their new infrastructure. The team wanted to dive into building predictive models and reap the benefits of better understanding their data, and wanted a foundation that could support this endeavor.

With the new infrastructure, a Linux server pulls data from numerous data sources and converts them into CSV files (flat files). These CSV files are controlled and populated into SQL databases via Rundeck, a third-party automation tool. Azure Synapse Analytics then applies data transformations via pipelines and Spark notebooks. Once data has been prepped, it’s pushed into the consumption layer where it’s used for visualization and building specific data models.  

FedEx is currently using Power BI to produce reports and dashboards from their data models and has been standardized on the tool since 2017. Visualization is mostly orchestrated by the enterprise business strategies team via requests, but there is a small amount of self-service occurring within FedEx.

With this new foundation, prepping data and executing transformation has never been easier, and the team can shift their focus away from their concerns around data volume issues. With Azure Synapse, FedEx can feel confident in having increased capacity for data and data management; the ability to process and leverage data is now faster than ever. “Everybody took to [Azure Synapse] right away. We’re more productive in it and have had good results,” explains Tom Heroux, Manager of the Prediction Models team.

Diving into data with prediction models

The biggest use case for prepped data has been the ability to build prediction models. The prediction model team builds problem-specific models to help further support visualization and pinpoint the right data for decision making. Models pull raw data into Synapse pipelines to execute transformations and aggregations on massive tables to make them more consumable. From there, the model is trained, tested, and undergoes daily and weekly batch predictions. Data is then written out into Synapse SQL Databases and the data is pushed back into on-prem systems for production access and visualization. The team has created a wide range of models to fit the growing needs of the Financial Services organization.

The Customer Payment Segmentation model is a cluster model used to classify customers by payment behavior to target collection strategies and understand customers that don’t always pay timely. The CPS models add an additional layer of payment behavior classification. The 12-month model indicates long-range behavior patterns. By comparing those patterns with a 3-month model, we can identify short term behavior changes and react to them sooner.

The Credit Risk Score model was created to assist decision making on whether FedEx should extend credit to new and/or existing customers. With the help of TATA consulting, FedEx leveraged batch prediction to help reduce the accumulation of bad debt and not extend debt to risky customers.

The Freight and Anomaly detection model (underway) aims to manage invoicing, agent workload, and customer experience. Leveraging AutoML capabilities from the Azure Machine Learning directly in the Synapse Workspace, this model attempts to predict if a freight customer will be invoiced correctly or if the customer will request any adjustments to the invoice. If the model is successful, the financial services team can reduce incorrect invoices and improve the overall customer experience.

Azure Synapse Analytics has opened the possibilities for the data models that are being built, and the volumes of data that the models can process. “If we must do data transformations or if we must prep data for modeling, we can do a huge file all at once and [Synapse] will ramp up and crunch the data. This would have taken hours to do on prem and impact a lot of users because you’re stealing so much CPU,” Heroux explains.

Leaning into the cloud vision

Since the adoption of Synapse and Azure, paired with the use of Power BI, the Enterprise Business Strategies team has been impressed with the changes and impact they’ve seen with their new solution. The scalability and resilience of having the foundation of Azure has been key to keeping up with the demand and growth FedEx has seen.

“Learning the Azure platform requires rethinking and reevaluating the way you think through everything, but once you’ve adapted to the process, utilizing it, and even spinning up new functions and features becomes very rapid. I think it’s one of the extremely large strengths that Azure has,” says Nelson. “The speed and scalability of provisioning resources has been amazing.”

Moving forward, FedEx wants to continue to migrate more of their services to Azure and leverage the cloud to its fullest extent, including an expansion of the integration between Power BI and Azure Synapse. The team wants to keep building more models with Azure Synapse Analytics and Power BI, to help take data modelling and visualization to the next level.