Case Study · Sports & Sponsorship
Upload a race. Get a verifiable sponsor exposure report. No manual review.
01
Sponsorship in cycling has long relied on manual estimation of broadcast exposure. It is a process that is both unreliable and expensive, and it leaves teams unable to give sponsors a defensible answer to the question that matters most: what did we actually get?
Together with Lucky Sport Cycling Team, we set out to replace that estimate with measurement. The solution is a serverless platform that turns raw broadcast footage into a verifiable exposure report. Users upload a race, and the system handles the rest.
02
Lucky Sport Cycling Team is a competitive cycling team with a portfolio of commercial sponsors. Like most teams, their sponsors invest significant budgets with little visibility into actual media exposure. The team had no reliable way to give sponsors a defensible answer to the question that matters most: what did we actually get?
03
Sports sponsors invest significant budgets with little visibility into actual media exposure. Lucky Sport needed a way to prove approximately how much visibility their sponsors receive, with hard data rather than estimates. Existing approaches relied on manual review, which is both unreliable and expensive.
04
A race goes in. No prep, no manual review.
Custom-trained object detection identifies the team's riders frame by frame across the entire broadcast.
AWS Transcribe converts race commentary into a searchable text stream.
Anthropic's Claude on AWS Bedrock surfaces narrative mentions of the team and individual riders.
A finished PDF, per race or aggregated across a season.
Technologies used
AWS Batch · AWS Lambda · AWS Transcribe · AWS Bedrock + Anthropic Claude · Custom object detection · Claude Code (development)
05
Each report delivers concrete figures: screen share, estimated exposure time, virtual exposure time, verbal mentions, and a shot-type timeline showing how the team appeared throughout the broadcast.
Building this meant solving the real-world messiness of broadcast footage along the way: corrupted video containers, multi-hour races, and an asynchronous pipeline that has to coordinate ML inference, transcription, and language model analysis across a single job.
For Lucky Sport and their partners, this turns a sales conversation built on estimates into one built on evidence.
Sponsors receive verifiable exposure data rather than estimates
The team can generate a full report for any race without manual review
Reports aggregate across a season for trend-level analysis
Generative AI at the core enables fast iteration as the sport and sponsors' demands evolve
Let's explore how AI and computer vision can turn your data into evidence.
Get in touch