Transforming Machine Learning with Databricks Optimization: Pacers Achieve Unprecedented Cost Reductions

For Pacers Sports and Entertainment (PS&E), understanding fan behavior is just as crucial as analyzing basketball statistics. The organization oversees the Indianapolis Pacers (NBA), Indiana Fever (WNBA), and Indiana Mad Ants (NBA G League). Initially, PS&E invested $100,000 each year in a machine learning (ML) platform to generate predictive models for ticket pricing and demand. Unfortunately, this approach did not yield timely insights, causing frustration among the team.

To resolve this issue, Jared Chavez, the manager of data engineering and strategy, made a significant shift to Databricks Optimization on Salesforce just over a year ago. Remarkably, his team now executes the same predictive projects at a minimal cost of just $8 annually. The primary factor behind this drastic change is their ability to reduce ML compute requirements significantly.

Operational Cost Reductions: A 98% Decrease for PS&E

Besides managing basketball teams, PS&E operates an esports franchise known as Pacers Gaming, organizes March Madness events, and oversees an extensive event schedule at the Gainbridge Fieldhouse arena. The company also plans to build a $78 million Indiana Fever Sports Performance Center, slated for completion in 2027. This ambitious project generates enormous amounts of data and poses various data management challenges.

Previously, PS&E used two separate data warehouses within Microsoft Azure Synapse Analytics. Different teams applied various analytics tools, leading to inefficiencies. While Azure Synapse could connect to external platforms, it was too costly for PS&E’s operations. The integration between the ML platform and Microsoft Azure Data Studio only intensified the issue.

To address these hurdles, Chavez transitioned to Databricks AutoML and the Databricks Machine Learning Workspace in August 2023. Their goal was to configure, train, and deploy models for ticket pricing and game demand more effectively.

Cost Efficiency and Enhanced Usability through Databricks

Both technical and non-technical users have found the Databricks platforms to be significantly beneficial. This transition has accelerated the ML processes while dramatically cutting costs. As Chavez points out, “It dramatically improves response times for my marketing team because they don’t have to know how to code.” By creating an intuitive user experience, Databricks ensures seamless access to extensive data, empowering all team members.

Additionally, Chavez’s team organized over 60 systems into Salesforce Data Cloud. This reorganization resulted in a staggering 440-fold increase in data storage and eight times the number of data sources in production. Now, PS&E operates with less than 2% of its original annual operational expenses, saving hundreds of thousands each year. These savings have been redirected towards enhancing customer data and providing improved tools for their analytics teams.

Optimizing Data Use for Superior Insights

But how did PS&E achieve such an incredible decrease in compute costs? Chavez explains that Databricks has consistently refined cluster configurations and improved schema connectivity. The sophisticated ML engine frequently enriches, combines, refines, and predicts customer data across multiple systems.

This continual refinement enhances prediction accuracy with every iteration. Intriguingly, some AutoML models can reach production without further adjustments. Chavez mentioned, “It’s just knowing the size of the data going in, but also roughly how long it is going to take to train.” His team’s focus lies in optimizing data storage and retrieval for improved efficiency.

Uncovering Customer Preferences

One innovative technique that Chavez’s team uses is propensity scoring for season ticket packages. Their aim is to analyze customer traits that influence seating choices. By geo-locating customer addresses, they correlate demographics like income levels and travel distances. Additionally, they examine purchase histories across food and beverage sales, mobile app interactions, and various events hosted on PS&E’s campus.

Furthermore, they leverage external data from ticket vendors such as Stubhub and Seat Geek to assess pricing strategies and inventory flux. This comprehensive analysis allows them to upsell customers effectively by identifying fitting seating options, thereby boosting revenue opportunities.

Enhancing Sponsorship Opportunities through Data

Data also plays a vital role in elevating sponsorship agreements, which are crucial for sports organizations. Chavez emphasizes the importance of partnering with companies that share similar audience demographics. By enriching data, creating custom segments, and boosting prediction capabilities, PS&E delivers actionable insights.

Ideally, they aspire to develop an interface that enables users to easily request and obtain relevant data, such as identifying fans in their mid-20s with disposable income. This capability allows partnership teams to access critical information independently, speeding up the deal-closing process.

To foster these endeavors, Chavez’s team is working on establishing a secure data clean room. This environment will facilitate the sharing of sensitive information, which can be particularly beneficial for sponsors and collaborative projects with other organizations.

Enhancing Customer Experience with Location Data

Another area of focus for Chavez’s team involves examining visitor patterns across PS&E’s campus, which includes various facilities, such as a three-tier arena with an outdoor plaza. By utilizing data capture methods via WiFi access points, they can track visitor movements without compromising user identity.

This capability will eventually help guide patrons throughout the arena by assisting with navigation towards concession stands or merchandise kiosks. Additionally, location data can optimize signage placement and measure impression counts.

For instance, by monitoring foot traffic around signage at typical viewing heights, they can report impactful visibility metrics to sponsors. This data-driven approach also allows the team to evaluate advertisement exposure based on fan location and visibility during events.

Ultimately, PS&E aspires to model its entire campus in partnership with Indiana University’s VR lab, creating an innovative environment to explore and analyze customer behavior and data utilization.