The second step in the sportsbook value chain

In the previous article, we looked at how sports betting data feeds work and the data infrastructure that powers any operator. That was the first step in the chain. Now it’s time for the second: what happens to that data to turn it into competitive odds and, above all, into a profitable business.
Trading and risk management is where data is transformed into prices, where the operator decides how much risk to accept in each market, and where poor technological infrastructure shows up as lost margin. Understanding these mechanisms is essential for any platform provider that wants to operate seriously in regulated markets.

This article breaks down the four pillars of that function: what sports trading is, how odds calculation models are built, what live trading demands, and what role AI and Machine Learning play in an environment where speed is measured in milliseconds.

Sports Betting Trading

Sports trading is the discipline responsible for setting, adjusting, and managing a sportsbook’s odds in real time. It is what keeps the operator in financial balance over time, market by market and bet by bet. The challenge is bigger than it looks. A mid-sized sportsbook can simultaneously offer thousands of markets across dozens of sports and competitions. Each of those markets has its own risk profile, its own betting-flow dynamics, and its own rate of change. A trader’s work is built around three core responsibilities:

• Price setting: Converting a real probability estimate into a published price that already incorporates the operator’s margin. This is the highest-impact decision over the long term, because a systematic pricing error translates directly into structural losses.
• Exposure management: Continuously controlling the maximum risk the operator has assumed on each possible outcome. If exposure on a market exceeds acceptable thresholds, the trader can adjust the odds on the overexposed outcome, cap the maximum bet amount, or suspend the market temporarily until balance is restored.
• Line movement: Adjusting odds in response to incoming bets, relevant new information, or market movements at other bookmakers. When professional bettors place large, informed bets, the market moves quickly:

Odds Calculation Models

From probabilities to odds: the overround

The basis of any odds price is a probability estimate. Before publishing a price, the operator needs to determine the real probability of each possible outcome. Once the real probabilities are estimated, the operator converts them into odds by incorporating its commercial margin.

The mechanics work as follows: in a perfectly fair market, the sum of the implied probabilities of all outcomes would be exactly 100%. The bookmaker pushes that total above 100% to guarantee itself a profit regardless of the outcome.

The standard formula for calculating the margin in a three-outcome market is: Margin = (1/Home Odds + 1/Draw Odds + 1/Away Odds) × 100

If the result exceeds 100%, that excess is the operator’s margin. For example, odds of 2.56, 3.20, and 3.30 produce a total of 100.61%: the overround is 0.61%. An operator with less competitive exposure can apply margins of 5% or more on those same markets.

Margins are not uniform: they vary by sport, competition, and market type. In parlays, the overround compounds: four selections with an individual margin of 5% each generate a combined margin of more than 21%. This is one of the reasons parlays are, from the operator’s perspective, the highest-margin product per unit of risk.

Statistical pricing models

The models feeding pricing have evolved substantially. Early systems used Poisson distributions based on goals scored and conceded for soccer, or point-spread models for American sports. Today, leading operators combine multiple approaches:

• Historical regression models with variables for recent form, home performance, match load, and weather conditions.
• Neural networks trained on millions of historical matches, capable of capturing interactions between variables that linear models miss.
• External market signals: the price at which other bookmakers have opened the same market serves as a validation reference for the operator’s own model.The quality of the pricing model is an operator’s main sustainable competitive advantage. More accurate odds than the competition attract legitimate flow and reduce exposure to bettors with superior proprietary models.

Live Trading

Why in-play is the sector’s technical frontier

Live betting has gone from being an additional feature to becoming the main revenue engine of any modern sportsbook. In the US market, in-play betting is projected to account for 44% of the handle at the 2026 World Cup. In more mature markets such as Europe, that share already exceeds 70% in some sports.

The technical demands of in-play have no equivalent in any other iGaming product. Predictive models process historical match data, real-time data feeds, and contextual factors such as weather conditions or match statistics to forecast outcomes in fractions of a second.

Infrastructure required to operate live

Live trading is not just a mathematical problem: it is, above all, an infrastructure problem. Three requirements are non-negotiable:

• Ultra-low latency: odds must be generated and updated in under 500 milliseconds. Any delay beyond that threshold exposes the operator to bettors with faster information or technology exploiting outdated prices before the system corrects them.
• Automatic market suspension and resumption: events such as a goal, a red card, a visible injury, or the half-time whistle require markets to be suspended within tenths of a second. Delays in doing so can translate into substantial losses on a single event.
• Scalability under extreme traffic peaks: a penalty in the 90th minute of a tight match, or the closing moments of a high-profile event, can multiply betting volume tenfold within seconds. A system outage at that moment is not a technical problem: it is a business problem with an immediate impact on revenue and reputation.

According to industry estimates, operators that cannot scale their in-play infrastructure during extreme traffic peaks lose between 15% and 25% of potential handle in maximum-audience events.

How AI and Machine Learning Are Applied

Pricing automation and predictive models

The rise of artificial intelligence has radically transformed sports trading, accelerating processes that once required teams of sport-specific traders. The most revealing data point comes from Kambi, one of the leading B2B technology providers for sportsbooks: 48% of the bets managed across its network in 2025 were processed by AI, up from 28% in 2024 and just 4% in 2022. In three years, AI has gone from an experiment to handling practically half of the sector’s total volume.

While a human analyst can factor in a key player’s injury or a team’s recent performance, a Machine Learning system simultaneously calculates the impact of weather conditions, pitch conditions, individual player habits, head-to-head history, and the opposing team’s tactical setup. What’s more, these models self-refine after every match, adjusting their parameters to continuously reduce error through reinforcement learning techniques.

Fraud detection and player profiling

Risk management has a second dimension that is just as critical: identifying and neutralizing threats to the integrity of the book before they turn into losses. Machine Learning algorithms analyze bet timing, amounts, user behavior, and transaction history in real time, detecting anomalies with a speed and precision no human team could match.

The most common threats these systems need to detect include:

• Multi-accounting: creating multiple accounts to exploit bonuses or bypass prior restrictions.
• Betting bots: automated scripts that place thousands of bets in milliseconds to exploit price variations between bookmakers.
• Collusion and syndicates: groups of bettors who coordinate their bets to drain liquidity or manipulate specific markets.
• Sharp action: bettors with superior proprietary models who systematically exploit mispriced odds.

The hybrid model: AI + human trader

Automation has not eliminated the human trader: it has redefined their role. The model gaining traction among the most sophisticated operators is the hybrid one: AI sets prices, manages operational exposure, and filters betting flow autonomously, while the human trader oversees the process, reviews the highest-impact alerts, and makes strategic decisions in high-complexity or low-frequency situations that models don’t handle well.

This approach makes it possible to cover thousands of simultaneous markets with a precision and agility no human team could achieve, while preserving expert judgment for edge cases. Tools such as Genius Sports’ Edge platform have shown that this hybrid model can generate margin increases of 22% in soccer markets during the 2025/26 season.

Conclusion

Trading and risk management are not an optional technical layer in a sportsbook: they are the core that determines whether the operator is profitable. Setting a price correctly, managing exposure in real time, protecting the book against sophisticated bettors, and scaling without friction under extreme demand peaks is what separates an amateur operator from one capable of competing in mature markets.

AI and Machine Learning have exponentially accelerated this function, but they haven’t simplified it: they have made it more complex along a different dimension. The challenge now is not having enough traders to cover every market, but having the infrastructure to run pricing models with millisecond latency, integrate content providers in a modular way, and manage risk continuously without manual intervention in 95% of cases.

For platform providers looking to operate in markets such as the US — where in-play will account for close to half of the handle as early as 2026 — the question is not whether to integrate advanced trading capabilities, but which architecture to build them on without that integration slowing down operational agility. Systems with a clear separation between the data layer, the pricing engine, and risk management will be the ones that capture this opportunity with a real advantage.

In sports betting, speed and precision aren’t competitive advantages: they’re the minimum price of entry.

CRMPAM – The quiet power behind your wildest traffic