Agentic AI Is Reshaping Sports Business Economics

The standard arc of a technology conversation in professional sports runs like this: enthusiastic vendor, cautiously curious team executive, moderator steering toward "what does this mean for fans." What made a recent SBJ Live panel worth pulling apart is that the executives doing the talking weren't speculating about potential. They were reporting back from deployment.

Iwao Fusillo, who holds the unusual title of Chief Data and Analytics Officer at the New York Jets — a role that spans both football analytics and business analytics, apparently unique across all 32 NFL clubs — described compressing what used to be an eight-month, nearly 100-contractor data foundation rebuild at General Motors down to eight weeks at the Jets, using AI. That's not a forecast. That's a completed project.

The structural observation embedded in that comparison deserves more weight than it typically gets in these conversations. The bottleneck in sports analytics has never been the lack of sophisticated people. It's been the cost and timeline of making raw data usable. When that compression becomes routine rather than exceptional, the competitive calculus across front offices changes — not just for teams with Fusillo's mandate and resources, but for organizations that have been structurally priced out of the analytics arms race.

Fusillo frames the Jets' AI rollout through a three-horizon model he developed during his PepsiCo tenure. Horizon one is adoption — getting the workforce to actually use AI tools daily, which he describes as moving from a handful of Microsoft Copilot users to 91% of the front office in 100 days. Horizon two is workflow automation, where he claims double-digit revenue and productivity gains. Horizon three is what he calls AI as decision architecture: a hybrid workforce of humans and AI agents where, in his telling, the returns become multiplicative rather than incremental.

The important operational note, which often gets skipped in the abstraction of these frameworks, is that Fusillo argues the horizons run simultaneously rather than sequentially. The Jets logged more than 20 Horizon 2 deployments and three Horizon 3 deployments concurrent with the cultural adoption push. Each Horizon 2 deployment runs roughly four weeks with 10 to 20 iterations. That's a meaningfully different operating cadence than the traditional software development lifecycle, where a dashboard UI change might take months.

The mechanics matter here. Fusillo described a coaching-facing scouting application the Jets call Titan. Asked to update the interface, his team used Claude Code, fed it a photograph of a preferred dashboard from a completely different context, and had the application rebuilt in 30 minutes. The analogy he reaches for is card-counting: the football acumen in the building stays constant, but better data and faster tool iteration changes the odds.

On the sponsorship side, the structural shift being described is a move away from what Shripal Shah of Next League calls the retrospective valuation model — where a sponsor buys a promise and waits, sometimes months, for confirmation data from the team or a syndicated source. The emerging alternative is real-time activation tied to in-venue intelligence.

Fusillo walked through a live example that crystallized this well: the Jets, using three years of concessions transaction history — 2.4 million transactions, roughly $80 million in value, five million items sold — built a mobile-ready analytics application for their in-game commerce team in 24 hours. The insight that emerged wasn't the top-selling item (four-piece chicken tenders and fries, which Fusillo cheerfully admitted wasn't the revelation). It was the co-purchase behavior: food paired more with water than with beer, an anomaly explained by the grab-and-go structure of in-stadium beer portables. That kind of friction diagnosis, at that speed, is what makes AI useful in an operations context rather than a research context.

Shah's point about what this means for the commercial structure of sponsorship is worth sitting with. The shift from retrospective reporting to real-time activation changes the product a team is actually selling. A sponsor isn't buying impressions to be measured later — it becomes a participant in a live data loop where activation triggers are set against observable in-venue behavior.

Josh Gwyther of Owl AI occupies a different part of this conversation — he's building the vision layer, the intelligence that sits behind broadcast cameras rather than in databases. The core argument is that leagues already own the hardware (high-resolution broadcast cameras), and what's been missing is an AI layer capable of interpreting what those cameras see. Owl AI is deploying this specifically on officiating and judging, turning the highest-stakes, highest-engagement moments in a broadcast into both an accuracy tool and — this is the commercial angle — a sponsorship inventory that didn't exist before.

"Just in the last couple weeks, we've created multiple seven-figure activation sponsorship inventories across multiple sports that just didn't exist before," Gwyther said during the session.

That framing is deliberate. The AI layer isn't replacing a human-facing product; it's creating a new sponsorship surface in an environment that already has maximum audience attention. Shah extended this toward a structural observation about tech sponsorship categories: what used to be a single "powered by" arrangement can now be an ecosystem activation — Nvidia, AWS, OpenAI, Anthropic, Google Cloud — where each vendor has aligned incentives because their commercial success is interconnected. A CRO can, in theory, sell into that entire stack through a single activation.

Whether that ecosystem argument holds under real commercial scrutiny — whether tech giants will actually co-brand in ways that benefit a league's revenue rather than primarily benefiting their own brand relationships — is an open question the panel didn't fully press. But the directional logic is coherent.

The data infrastructure question for emerging leagues surfaced late in the conversation and got the most honest treatment. The minimum viable answer from Shah: a data warehouse, a customer data platform, an email/SMS solution, a CRM. The unsexy foundation. What's changed, he noted, is that 18 months ago you'd stand that up and then tackle AI applications. Now you can integrate AI into the initial rollout, compressing the timeline from foundation to deployment.

Gwyther's contribution here was the most practically urgent: save your camera feeds. Not just the broadcast cut — the raw feeds. Cloud storage is cheap, he argued, and unstructured video becomes structured, actionable data quickly with the right platform. The heartache he described, working with a new league that had deleted last season's camera footage, is the kind of institutional decision that looks reasonable under old economics and catastrophic under new ones.

The counterweight to all of this enthusiasm, raised with appropriate understatement by Shah, is cost. As AI query volume scales across an organization, so does the infrastructure bill. Shah pointed to Uber as an example of an organization that burned through its AI budget for the year ahead of schedule. For sports organizations operating on tighter margins than Big Tech, the productivity gains need to outpace the consumption costs — and the accounting for that isn't always in place before deployment begins.

On the three-to-five year horizon, Gwyther made what strikes me as the most consequential structural argument in the entire discussion: AI-produced broadcasts could reduce production costs to roughly 5% of current levels, which would lower the revenue threshold a sport needs to cross to justify professional infrastructure. The implication isn't just efficiency for existing leagues — it's an expansion of what qualifies as a viable professional league in the first place.

Fusillo's endpoint is different but complementary. His target isn't 2-3x productivity — it's 20-30x, achievable only when every workflow is reimagined around AI agents rather than augmented by them. Coding agent, testing agent, an agent that argues with the other agents. "Then as we've taken the human out of the loop," he said, "we could actually get to a 20 or 30x." He was careful to note that governance becomes the dominant concern at that multiplier — which is either a reassuring acknowledgment or a problem deferred, depending on how seriously organizations are building governance infrastructure today versus planning to address it after the productivity gains are already locked in.

The sports industry has absorbed many rounds of technology promises. What distinguishes this conversation is the specificity of the deployed examples and the seniority of the people describing them. The question worth holding is whether the organizations that move fastest on deployment are also the ones building the oversight architecture to manage what they're deploying.

— Marcus Tate, Sports Desk Editor