The conventional narrative about prediction markets is that they have no house. Prices emerge from the crowd, representing the aggregate wisdom of thousands of traders. That story was mostly true in 2022. It is no longer true in 2026.
Today, the dominant price-setters on Kalshi and Polymarket are not retail punters or even sophisticated sports bettors. They are Susquehanna International Group, Jump Trading, Galaxy Digital, and a cohort of quantitative funds that have imported Wall Street market-making infrastructure directly into sports event pricing. Approximately 55% of prediction market funding now comes from institutional capital, according to industry analysis compiled through early 2026. The sports bettors show up for the liquidity. The institutions set the price.
For traditional sportsbook operators, this creates an urgent strategic question: are you a price-setter or a price-taker in the fastest-growing segment of sports wagering? The answer hinges on three things—your calibration methodology, your re-pricing infrastructure, and whether you own the exchange layer or merely rent access to it.
Market StructureNo House Edge: How Prediction Market Prices Actually Form
Prediction markets operate on peer-to-peer central limit order book (CLOB) mechanics. There is no house setting the line and no spread baked in by an operator. Contracts trade at prices between $0.01 and $0.99, representing implied probability. A contract priced at $0.62 means the market collectively assigns a 62% probability to that outcome. Winning contracts settle at $1.00. The probability framing is intuitive for users who think in percentages rather than fractional odds—a UX insight with real CRM implications we will return to.
In a functioning liquid market, prices are emergent. Informed traders buy underpriced contracts; selling pressure corrects overpriced ones; equilibrium forms. But in thin books—new event categories, niche sports, early-week markets before volume builds—emergent pricing breaks down. A single $1,000 order can move a prediction market price by 10 percentage points when the book is thin. Into those gaps, institutional market makers step in to provide liquidity and, in doing so, effectively anchor the price.
The scale of the market these institutions are operating in has become impossible to ignore:
Kalshi alone generated $263.5 million in fee revenue in 2025, the majority derived from sports event contracts, and closed a Series E at an $11 billion valuation in December 2025. Monthly active users grew 8.5x during the year, from 600,000 to 5.1 million. These are not experimental numbers. This is a mainstream category, and institutional capital recognized that before most sportsbook operators did.
The Real OddsmakersWall Street Moved In: The Institutional Takeover of Sports Pricing
The import of financial market microstructure into sports prediction pricing is now complete. Susquehanna and Jump Trading operate as primary market makers on Kalshi, deploying the same infrastructure they use in equities and options markets: co-location for latency advantages, automated arbitrage desks running cross-platform strategies, and algorithmic book management that adjusts positions within milliseconds of any new information entering the market.
Cross-platform arbitrage between Kalshi and Polymarket yields 0.5–3% returns on a given contract, closing within seconds. This sounds small until you consider that these strategies run continuously, at scale, with institutional capital. An operator without sub-10ms re-pricing infrastructure is systematically disadvantaged every time a sharp bettor or algo fund spots an arbitrage window between their book and the prediction market price. The adverse selection is structural, not occasional.
The oddsmaking battle playing out on Kalshi and Polymarket is described as “underappreciated” by industry insiders who have watched it develop. Who anchors lines on these platforms will determine whether traditional sportsbooks or financial institutions extract the majority of operator margin from sports prediction markets as the category matures. Right now, the answer leans toward Wall Street—but the structural advantage is not permanent. Sportsbook operators hold something institutional market makers do not: proprietary sports data.
Calibration Is the Right Target—Accuracy Will Lose You Money
Most operators running pricing models today are optimizing for the wrong objective. The default assumption in sports analytics is that better predictive accuracy—picking more winners correctly—produces better pricing. That assumption is wrong, and the financial cost of operating under it is measurable.
A rigorous NBA study comparing calibration-optimized machine learning models against accuracy-optimized models found a 69.86% performance gap in realized returns: calibration-optimized models returned +34.69%, while accuracy-optimized models returned -35.17% over the same period. This is not a marginal difference. It is the difference between a profitable pricing operation and one that systematically gifts margin to sharp bettors and institutional arbitrage desks.
The distinction matters because calibration and accuracy measure different things. An accuracy-optimized model asks: does this model pick the winner more often than not? A calibration-optimized model asks: when this model assigns 60% probability to an outcome, does that outcome occur 60% of the time? Accuracy is about being right. Calibration is about pricing risk correctly.
In a peer-to-peer prediction market where you are competing as a market maker, miscalibration is directly extractable by any sophisticated counterparty. A contract you price at $0.52 that should be $0.58 will be bought immediately by any algorithm running a calibration-adjusted model. The institutional firms already have this infrastructure. Traditional sportsbook operators, many of whom inherited pricing models built for house-edge environments rather than competitive market-making, are catching up.
The good news: AI-driven pricing models are measurably better at calibration than traditional statistical approaches. Top AI models beat closing line value (CLV) by 3–7% on average across tested markets. Generative AI models reach 75–85% game-winner accuracy compared to 50–60% for traditional statistical models—but more importantly, they tend to produce better-calibrated probability distributions, not just better binary predictions. The upgrade path from accuracy-first to calibration-first pricing is tractable for operators with the right data infrastructure.
Speed as WeaponSub-Second Re-Pricing Is Now Table Stakes
Calibration alone is not sufficient. In a live prediction market, prices are not static. Breaking news—an injury confirmed in warmups, a weather change affecting an outdoor game, a lineup surprise—can shift market prices 40–50 percentage points instantly. The question is not whether prices will move; it is whether your book moves first or whether sharp bettors and algorithmic funds move first against your stale price.
The latency arms race in prediction markets is now equivalent to what occurred in equity markets a decade ago. Sharp bettors and algorithmic funds react to breaking news within milliseconds. In a thin order book, a $1,000 order moves the price 10 points. An operator running batch pricing updates—even on a five-minute cycle—faces a window during which every trade against their book is adversely selected by someone with faster information processing.
Sub-second re-pricing requires event-driven architecture, not scheduled jobs. Triggers must fire on news feed updates, not on a clock. The re-pricing model must be pre-loaded and callable in memory, not fetching parameters from a database on each request. This is an engineering constraint as much as a data science one—and it is the reason that financial market infrastructure firms have an early advantage over traditional sportsbook technology stacks.
A new category of tooling is emerging to help operators profile the other side of the market. Precog, launched in Q1 2026 beta, assigns every prediction market wallet an “alpha score” based on historical hit rate, timing precision, and conviction sizing. This is the prediction market equivalent of sharp bettor profiling in traditional sportsbooks—operators who have built internal sharp-money identification systems for their core books can understand the structural parallel immediately. Knowing which counterparties are systematically informed changes how you manage book exposure and at what odds you are willing to make markets.
Operator AdvantageWhere Sportsbooks Beat Wall Street: Proprietary Data Moats
The institutional market makers dominating prediction market pricing today are extraordinary at financial market microstructure. They are less extraordinary at sports. Susquehanna and Jump Trading can run latency arbitrage and manage cross-platform book exposure at scale. What they do not have—and cannot easily acquire—is the proprietary real-time sports data that large sportsbook operators have built over years.
Flutter Entertainment CEO (FanDuel parent company) articulated this advantage directly: “The ability to price complex correlated outcomes accurately is something that we do every day in our core business.” This is a genuine structural edge. Sportsbook operators with access to real-time injury feeds, internal sharp money flow data, player tracking signals, and officiating pattern databases can construct pricing models for sports event contracts that financial market makers simply cannot replicate from publicly available data alone.
The operators who recognize this are not waiting. Underdog acquired Aristotle Exchange, a CFTC-regulated prediction market operator, directly acquiring the exchange infrastructure layer. DraftKings launched its Predictions product in 38 US states via CME Group, with plans to integrate Railbird for additional exchange functionality. DraftKings CEO Jason Robins has explicitly identified prediction markets as the primary growth lever in approximately 12 US states where regulated sports betting is not yet legal. FanDuel has expanded its prediction market product nationwide, including California. This is not experimentation—these are strategic infrastructure acquisitions.
The critical distinction is between owning the exchange layer and merely accessing it. Operators relying on vendor-supplied odds piped into a prediction market front-end are permanently downstream from whoever is setting those odds. Owning exchange infrastructure means setting the initial market, managing the book directly, and retaining fee revenue that would otherwise accrue to a third-party platform. Kalshi’s $263.5 million in 2025 fee revenue demonstrates what that revenue line looks like at scale.
CRM ConnectionPricing Gets You In. Personalization Keeps You Winning.
Getting prediction market pricing right is necessary but not sufficient. The operators who will extract durable advantage from this category are those who connect pricing infrastructure to CRM personalization at the individual bettor level.
The probability framing of prediction market contracts—$0.01 to $0.99—is a CRM asset that most operators are underusing. Users who think in percentages rather than fractional odds find prediction market contracts more intuitive than traditional bet slips. That lower cognitive barrier translates to higher engagement rates and more CRM touchpoint opportunities: a user who checks a prediction market price daily is a user you can communicate with around that contract, its progress, and related markets.
The personalization economics are well-documented in adjacent sportsbook CRM work. AI-driven personalized offers generate 20–30% higher revenue compared to generic campaigns. AI personalization drives a measurable +21% increase in bet frequency. For prediction markets specifically, the personalization opportunity extends beyond content—it includes market recommendations calibrated to a user’s demonstrated probability sensitivity, preferred contract duration, and historical position sizing.
The market is also bifurcating in a way that creates distinct operator strategies. Content-driven platforms compete on novelty, engagement, and new market creation. Market-driven platforms compete on liquidity depth, fee structures, and re-pricing speed. Operators with sportsbook heritage have the data and customer relationships for both dimensions—which is precisely why financial market makers are not the inevitable long-term winners here, despite their current infrastructure advantage.
The global AI sports betting market was valued at approximately $9 billion in 2024 and is projected to reach $28 billion by 2030, a 21.1% CAGR. Prediction markets are the fastest-growing segment within that trajectory. Operators who invest in both the pricing infrastructure and the personalization layer now are positioning for compounding retention advantages that will be structurally difficult to close later.
PlaybookThe Operator Checklist: From Price-Taker to Price-Setter
The path from price-taker to price-setter in prediction markets is a five-step infrastructure build. Each step is tractable independently, but the compounding value comes from executing all five in sequence.
Step 1: Audit Your Pricing Stack
Are your models optimized for calibration or accuracy? Most operators cannot answer this question because they have never explicitly framed the distinction. Run your existing model’s probability outputs against historical outcomes across a representative sample of markets. If your 60% contracts are winning 64% of the time, you are systematically underpricing. If they are winning 56% of the time, you are overpricing and gifting value to informed bettors. The 69.86% performance gap between calibration-first and accuracy-first models is not a theoretical concern—it is a margin extraction problem you can quantify against your own book.
Step 2: Build Sub-Second Re-Pricing Infrastructure
Move from batch pricing updates to event-driven triggers. News feed integrations, officiating alerts, injury confirmation feeds, and lineup release systems should all fire re-pricing events in real time. The model must be callable in under 100 milliseconds. This is an engineering project, not a data science project, and it should be scoped as one.
Step 3: Layer Proprietary Data on Vendor Odds
Even if you are not yet operating your own exchange, you can differentiate pricing by layering proprietary signals on top of vendor-supplied odds. Internal sharp money flow data, player tracking feeds, and real-time injury information are not available to Susquehanna or Jump Trading unless they are purchasing the same commercial data services you are. Your internal data is your moat—use it in the pricing layer, not just in the CRM layer.
Step 4: Consider Exchange Infrastructure Acquisition or Partnership
The Underdog/Aristotle and DraftKings/CME models point toward the same conclusion: front-end-only prediction market participation is a shrinking competitive position. CFTC-regulated exchange infrastructure is acquirable. The operators who own the exchange layer retain the fee revenue that would otherwise accrue to a third-party platform and set the initial market rather than reacting to it.
Step 5: Connect Prediction Market Engagement to CRM Flows
AI personalization should drive repeat handle and cross-sell into regulated sports betting markets in states where both products are available. The 20–30% revenue uplift from personalized offers documented in traditional sportsbook CRM is directly applicable to prediction market engagement flows. Users who engage with prediction market contracts are high-intent sports bettors—exactly the segment your CRM should be communicating with around upcoming fixtures, related markets, and personalized offers calibrated to their demonstrated preferences.