March Madness betting: We had AI pick every first-round game of the men's NCAA tournament. Here's who won (and covered)

AI (ChatGPT) using a blended model of KenPom, Evan Miya, FTN Fantasy, ESPN BPI and Bart Torvik power ratings produced projected final scores and against-the-spread (ATS) covers for all 32 men’s NCAA first-round games. Examples include Duke 86–60 (Siena +28.5 ATS) and Michigan 89–59 (Howard +30.5 ATS); projections reference current BetMGM odds. Use for sportsbook exposure and odds-monitoring; negligible market impact.

Analysis

The rapid integration of algorithmic predictions into sports media and wagering workflows is a classic two-sided market shift: data/licensing owners capture recurring revenue while distribution platforms commoditize signal consumption. Expect consolidation of value with firms that control both real-time feeds and low-latency model deployment — they can charge premium fees for exclusivity and embed revenue share into handle growth, a dynamic that can lift revenue growth by +20–40% for incumbents that secure deals within 12 months.

A second-order technical effect is increased intra-event volatility in live markets as model-driven traders act off the same signals; bookmakers will respond by widening limits and increasing vig on high-correlation books, eroding operator margins by an incremental few hundred basis points unless they upgrade liability management. That creates a timing mismatch: data providers benefit quickly (weeks–months) while operators face margin pressure over quarters as risk controls and product changes propagate.

Tail risks center on model decay and adverse-feedback loops: when widely published signals move lines, they change the very distributions the models trained on, producing overfitting and a potential reputational shock if multi-event backtests fail in live conditions. Regulatory scrutiny and disclosure requirements around automated betting advice are plausible 6–18 month catalysts that could force licensing re-negotiations or introduce compliance costs, compressing multiples for exposed names.

Key Decisions for Investors

• Long GENI (Genius Sports) — 6–12 month horizon. Rationale: owns distribution/licensing leverage for real-time data and is first-order beneficiary of paid model feeds. Target upside 30–50% if it converts 2–3 mid-size sportsbook contracts; set a 30% trailing stop to limit model/contract risk.
• Long SRAD (Sportradar) — 9–18 months. Rationale: portfolio exposure to integrity services and data licensing; wins if operators outsource model hooks. Position size 3–5% capital, target 25–40% upside, downside protected by 35% stop-loss given execution/contract timing risk.
• Pair trade: Long GENI / Short PENN — 6 months. Rationale: data vendors monetize AI signals while domestically focused operators face margin compression from smarter limit-setting and higher commission; aim for 2:1 reward-to-risk. Size as market-neutral (dollar-match exposures), take profits at 20% pair move or cut at 12% adverse move.
• Buy DKNG 3–6 month calls (modest notional) to play short-term handle spikes around peak events, while keeping a 50% max loss per contract. This is a volatility play — if live-betting volumes increase, implied and realized revenues should lift; hedge by selling 1/2 position if implied vols rise >30% intraday.