How SūmerSports Uses Databricks to Deliver Real-Time NFL Insights in Seconds

SūmerSports was founded in 2022 with a single ambition: turn the explosion of NFL sensor data into decisions coaches, scouts, and executives can actually trust. Modern player tracking systems capture movement data multiple times per second, generating an entirely new class of intelligence that had previously gone unused at scale. The startup’s job is to translate raw tracking streams, scouting evaluations, financial contract data, and web-scraped sources — 14 in total, totaling over 200 terabytes per season — into actionable insight before the next draft pick, trade deadline, or roster cut.

Before Databricks, the company’s infrastructure was, in the words of Director of Data Engineering Matthew Coffey, “disorganized.” Projects ran on siloed pipelines, each with its own tooling and conventions. There was no unified lineage, no centralized orchestration, and no reliable way to validate that data was current or accurate. Data scientists prototyped on personal laptops; when team members moved on, institutional knowledge left with them. VP of Data Science and AI Adam Vonder Haar described the monitoring situation bluntly: “Visibility was limited. We only surfaced issues after they had already impacted downstream users.” Post-game intelligence — the core product — took up to four days to deliver after a Sunday game.

SūmerSports adopted the Databricks Data Intelligence Platform as its unified foundation for ingestion, transformation, ML development, and customer-facing analytics. Unity Catalog brought governance and lineage to all 14 data sources, allowing curated, tagged datasets to be safely discovered and reused across teams without duplicating work. MLflow replaced laptop-based experimentation with a persistent, searchable record of every model run. Databricks Asset Bundles enabled data scientists to ship infrastructure-as-code, moving models to production without a dedicated DevOps team. Databricks Assistant and AI/BI Genie gave analysts natural-language access to data without requiring SQL fluency. SūmerBrain, the conversational football chatbot, was prototyped on Databricks before routing production inference to AWS Bedrock.

The performance improvements were measurable and immediate. A team-level Expected Points Added model that previously required a three-month development cycle reached customers in under four weeks. Post-game data delivery, once a four-day pipeline, was first compressed to overnight delivery by early Monday morning, and then again to sub-minute delivery — meaning Sunday night games are available to professional users in near real-time. Data processing volume has tripled compared to two years prior, and 95% of workflows now run on serverless compute, eliminating cluster management overhead for most jobs.

With that foundation stable, SūmerSports is pushing into higher-resolution analytics. The company is combining structured player tracking with unstructured video to give SūmerBrain richer context, and is moving from play-level analysis toward within-play granularity — understanding not just what happened, but why. As Vonder Haar summarized, the infrastructure now in place has “enabled the company to grow, move faster and explore opportunities that simply were not possible before.”