Every market starts with a model. How sportsbooks source data, build pricing algorithms, and move from opening to closing line — and where the blind spots are.
Teasers feel like free money. The math says they usually are not. The exceptions — Wong teasers crossing 3 and 7 — are real and rare.
AU exotics carry 18–30% takeout but pool overlays appear on Cup Day, Cox Plate, and any race with rec-money concentration. The maths is gettable.
Every betting market you see started as a number in a model. Before the odds appear on your screen, a sportsbook has sourced data from multiple feeds, run it through statistical models, calibrated the output against market prices, and applied a margin. The process is automated, quantitative, and — for major markets — highly efficient.
Understanding how sportsbooks build their numbers is not about replicating their models. It is about understanding where the models are strong (making those markets hard to beat) and where they are weak (creating the pockets of inefficiency that sharp punters exploit). This piece covers the data pipeline, the modelling approach, and the line movement mechanics — from opening to closing — that shape every price you see.
A sportsbook's data infrastructure has four layers:
Layer 1: Official league data feeds. The major professional leagues (AFL, NRL, NBA, NFL, EPL) sell official data feeds directly to sportsbooks. These feeds include:
These feeds are expensive. An official AFL data feed license for a sportsbook costs in the hundreds of thousands annually. The cost creates a barrier: smaller books cannot afford the full feed and rely on third-party aggregators or market-following pricing. The quality gap between the data-rich books (Bet365, Sportsbet) and the data-poor books (smaller corporate books, some AU domestics) is one reason the same market can show different prices across bookmakers — they are not all looking at the same information.
Layer 2: Third-party data aggregators. Companies like Sportradar, Genius Sports, and Stats Perform aggregate data from hundreds of leagues worldwide and sell it to sportsbooks. A bookmaker covering 50+ sports does not negotiate 50 separate league data deals — it buys from one or two aggregators. The aggregator feeds are comprehensive but slightly slower than direct league feeds. The speed difference (seconds to minutes) matters for in-play betting but is less relevant for pre-match markets.
Layer 3: Market data. Sportsbooks monitor each other's prices in real time. A trading desk screen shows their own odds alongside the odds of every major competitor. When Bet365 moves a line, every other book sees it within seconds. This market data is as important as the sports data — if a book's model price is significantly different from the market consensus, the book has to decide whether it knows something the market does not (unlikely) or whether its model is wrong (more likely). Most of the time, it adjusts toward the market.
Layer 4: Betting flow data. Sportsbooks track every bet placed: which customers bet on which side, at what price, for how much. This internal data — particularly the behaviour of identified sharp customers — is a powerful input. If the sharpest 5% of a book's customer base is consistently betting one side of a market, the trading desk treats that as information. The sharp customer flow is, in effect, a free research department.
A sportsbook's pricing model is not one model. It is a layered system:
Base forecast model. The core statistical engine. For an AFL game, this model estimates the probability of each team winning (and the margin distribution) based on:
The output is a set of probability distributions: win probability, margin distribution, total points distribution. From these, the model derives fair prices for every derivative market: head-to-head, line, total points, player disposals, first goal scorer. The derivatives are mathematically derived from the core game model, not independently modelled.
Market calibration layer. The model's raw output is then calibrated against the current market price. If the model says the fair price is $1.85 and Betfair Exchange is showing $1.95, the calibration layer adjusts the model weight downward relative to the market signal. The calibration recognises that the market (particularly the exchange price) contains information the model does not have — late team news, sharp money flows, information embedded in the price itself.
Margin application. Once the calibrated fair price is determined, the bookmaker applies the margin. On a two-outcome market, the margin is typically applied symmetrically: fair price $2.00 becomes $1.91 each side (4.5% margin). On multi-outcome markets, the margin is often applied asymmetrically — more on the favourite (where recreational money concentrates), less on the longshot. This is why underdog prices at corporate bookmakers occasionally beat the exchange price — the bookmaker is shading the favourite's price to capture recreational volume and offsetting on the underdog.
The opening line is the bookmaker's model output with margin applied. It represents the bookmaker's best estimate of the fair price before any market activity. The closing line is the price immediately before the event starts. The movement from open to close is the market becoming more efficient as information is incorporated.
What drives the movement:
The closing line is, on average, the most efficient price — it incorporates all available information, all sharp money flows, and all market adjustments. Beating the closing line consistently (getting a better price than where the market eventually settles) is the strongest statistical evidence of a genuine betting edge. See the CLV guide for the full framework.
Sportsbook models are excellent at major markets in major sports. They are less excellent — sometimes outright poor — in markets that are:
Low volume. If a market gets little betting volume, the bookmaker invests less modelling resources in it. Player props in smaller sports, exotic markets in lower-tier leagues, and niche bet types are priced with thinner models and wider margins. The model is less accurate, which means genuine mispricings are more common. The trade-off: lower liquidity means you cannot get as much money down.
Multi-dimensional. A head-to-head market has two outcomes. A player prop has dozens (points, rebounds, assists, three-pointers, each at multiple line levels). A same-game multi has hundreds of correlated outcomes. The more dimensions a market has, the harder it is to model accurately, and the more likely the bookmaker is to apply a larger margin rather than invest in better modelling. The margin is the bookmaker's defence against model uncertainty. Sharp punters who can model the multi-dimensional markets more accurately can find edges in the gap between the bookmaker's wide-margin price and the true probability.
News-sensitive. Markets where the price changes rapidly in response to news (player injuries in NBA, team selection in AFL, pitch conditions in cricket) create windows where slower books have not yet adjusted. The edge here is not analytical — it is operational. You are not out-modelling the bookmaker; you are out-speeding the slower bookmakers by placing bets at the old price before they update.
Correlated. Markets where outcomes are correlated in ways that are hard to model precisely. Same-game multis are the prime example: if Player A scores a try, Player B (who passes to A) is more likely to get a try assist. The bookmaker knows the correlation exists but the precise magnitude is hard to estimate. The bookmaker's response is to apply a correlation discount that is deliberately conservative (favouring the bookmaker). The discount is an opportunity for punters who can model the correlation more accurately — but the modelling is genuinely difficult and the bookmaker's conservative discount is usually, but not always, large enough to cover their model uncertainty.
The practical takeaways:
No, but the outputs converge. Bet365 runs its own proprietary models and is typically the market leader on price movement. Sportsbet runs independent models but monitors Bet365 closely. TAB and the smaller domestic books use a combination of third-party pricing services and market-following. The models are different but the inputs are similar (same data feeds, same news, same market prices), so the outputs converge quickly. Within 30-60 minutes of the opening line being posted, most books are within 1-2% of each other on major markets. The convergence is the market becoming efficient.
Continuously. The model ingests new data — game results, player stats, injury news, market movements — in real time and updates its output. The line you see on your screen is the current state of a continuously updating model, not a static price set once and left. The continuous update is why prices move throughout the week on pre-match markets: the model is incorporating new information as it arrives and adjusting the price accordingly.
Daniel writes about the maths underneath advantage betting — expected value, Kelly sizing, closing line value, bankroll theory. Translates the theoretical side into practical decisions AU punters can actually apply.
Every other major betting market in the world has online in-play. Australia doesn't, for a specific legal reason that dates back to 2001. Here's what actually exists and what you can do with it.
