Pricing something that is moving
Pre-match pricing estimates the probability of outcomes in an event that has not started. In-play pricing estimates them continuously as the event unfolds, with every phase of play changing the answer.
The conceptual difference is that an in-play model must incorporate the current state of the event, the time remaining, and the pattern of what has happened, then reprice everything affected. In football, a goal changes not only the match result market but the goals markets, the correct score market, the next goalscorer market, the half-time markets and every bet builder combination containing any of them.
The volume of recalculation is substantial, and it must happen in a very short time, because until it does the displayed prices are wrong in a direction anyone watching the match can identify.
Why in-play transformed the economics
The commercial significance is hard to overstate. A pre-match football fixture offers one betting opportunity: the customer decides before kick-off. The same fixture in-play offers continuous opportunities across ninety minutes and dozens of markets.
Three effects follow. Turnover per event rose substantially, because customers who would have placed one bet now place several. Margin per event rose, because in-play markets typically carry wider margin than headline pre-match markets, prices move quickly enough that comparison is impractical, and decisions are made under time pressure. And engagement rose, since a customer with a live position watches the whole match rather than checking the result.
In mature markets in-play now accounts for a large share of sportsbook turnover, and for some operators the majority. It is the single most important product development in the vertical's history.
The data supply chain
In-play pricing is only as good as the information feeding it, and the supply chain is worth understanding because it explains the latency problem.
At the fastest end sit scouts, people physically present at venues reporting occurrences to a data supplier through dedicated applications within a second or two of them happening. This is the primary source for in-play data on major events and is why data suppliers maintain large networks of on-site personnel.
Behind that sit official data feeds, supplied under agreement with leagues and governing bodies. These carry authority and, in several sports, exclusivity arrangements that make them the only permitted source for betting purposes.
Behind those sit broadcast feeds, which carry inherent delay from production and transmission, often several seconds and sometimes considerably more depending on the distribution path.
And behind everything sits social media and secondary reporting, which is fast for major moments and unreliable for detail.
The commercial structure matters here. Sports data supply is a concentrated market with significant pricing power, and the cost of live data is a substantial line item for any sportsbook offering meaningful in-play coverage. Exclusive data rights arrangements have also become strategically important, since an operator without access to the official feed for a competition is structurally slower than one with it.
Latency as a commercial variable
The core problem is straightforward to state. Between something happening on the pitch and the sportsbook's prices reflecting it, there is a window. Anyone who knows what happened during that window can bet at prices that are already wrong.
Latency accumulates at every step: the scout observing and reporting, the supplier processing and distributing, the operator's system ingesting, the model recalculating, and the new prices publishing. Each step is measured in fractions of a second and they sum.
The exploitation of this window is known as courtsiding, and it involves someone at the venue transmitting information faster than the official chain, allowing an associate to bet before prices update. Several sports have prohibited unauthorised data transmission from venues and have removed people for doing it, and some jurisdictions have prosecuted it.
More prosaically, the same window is exploited by anyone whose broadcast feed is faster than the operator's data path, which varies by territory and distribution method. This is not sophisticated and is entirely predictable, which is why operators build controls rather than relying on customers not noticing.
Suspension logic
The primary defence is suspension: closing markets so no bets can be placed while prices are recalculated.
Suspension triggers on significant events reported through the data feed, such as goals, cards, penalties and substitutions in football, and their equivalents in other sports. It also triggers on dangerous phases of play, such as an attack entering the penalty area, since a goal may be imminent and prices are about to change sharply. And it triggers on feed anomalies, where data stops arriving or contradicts itself, because operating blind is worse than operating not at all.
Two design tensions run through suspension logic.
The first is coverage against availability. Suspending aggressively protects the book and produces a product that is frequently unavailable, which frustrates customers and costs turnover. Suspending sparingly keeps markets open and exposes the book during precisely the moments that matter most. Operators tune this continuously and differ markedly in where they set it.
The second is automation against judgement. Suspension must be automatic, because the timescales are far shorter than human reaction, but automatic systems act on data that may be wrong. A false event report suspends markets unnecessarily; a missed one leaves them open when they should not be.
Reopening is equally consequential. Markets must reopen with correct prices, and reopening too quickly with prices that have not fully recalculated recreates the problem suspension was meant to solve.
Bet delay and acceptance controls
Even with suspension, a gap remains between a customer submitting a bet and the system processing it. Bet delay deliberately pauses acceptance for a short period, during which the system verifies that the price is still valid and that no suspension has triggered.
If the price has moved, the bet is either rejected or offered at the new price, depending on the operator's configuration and what the customer accepted. Many operators offer customers a choice about whether to accept price movements automatically, which reduces rejection frustration at the cost of accepting a different price than displayed.
Delay is a genuine customer experience cost. Customers dislike waiting and dislike rejection. It is accepted because the alternative is systematic loss to anyone with faster information, and the arithmetic on that is not close.
Alongside delay sit acceptance rules that vary by customer classification. A customer with no demonstrated edge may have bets accepted automatically up to substantial limits. A customer whose betting has consistently beaten closing prices may face lower automatic limits and referral for manual approval above them. This is the point at which in-play risk control and customer management intersect, which the next lesson takes up directly.
Building the in-play model
A brief note on what the pricing side actually requires, since it differs from pre-match modelling.
An in-play model needs to represent the state of the event, not merely its starting conditions. In football that means the score, the time elapsed, the players on the pitch, cards issued and, in more sophisticated models, some representation of territorial dominance and chance quality.
It needs to update quickly enough to be useful, which constrains model complexity. A model that produces excellent prices in thirty seconds is useless when prices must refresh continuously.
It needs to handle sport-specific structure. Football's low scoring makes individual goals enormously significant to prices. Tennis has a rigid point, game and set structure that makes state representation clean and probability calculation relatively tractable. Basketball's high scoring makes individual scores less significant but makes time and possession critical. Cricket has distinct phases with different scoring dynamics. Each requires its own approach rather than a general one.
And it needs calibration against reality, meaning that prices produced during matches should, over many matches, correspond to observed frequencies. A model that prices a scenario at 20% should see that scenario occur close to 20% of the time in similar situations.
Responsible gambling and regulation
In-play attracts specific regulatory attention, and the reasons are substantive rather than arbitrary.
The concern is that in-play compresses the cycle between decision, stake and outcome. A pre-match bet involves one decision and a wait of hours. An in-play market may resolve within minutes, and the next opportunity is immediately available. That pattern of rapid, repeated staking with short intervals between stake and outcome is associated with elevated risk, and it resembles continuous gambling products more closely than traditional sports betting does.
Regulatory responses vary and have included restricting which in-play markets may be offered, prohibiting in-play betting on certain competitions or on lower-level sport, requiring the same responsible gambling tooling applied to casino products, and in a small number of jurisdictions restricting in-play betting substantially.
Integrity considerations reinforce this. Markets on discrete micro-events within a match, such as the outcome of a specific point or the timing of a specific action, are more susceptible to manipulation by a single participant than markets on the overall result. Several sports and regulators have restricted such markets for that reason, and the topic is examined properly in the final lesson of this course.
The practical position for anyone working in this area is that in-play product decisions are not purely commercial. What may be offered, on which competitions and with what protections, is constrained by rules that differ by market and continue to tighten.
Sport-specific characteristics
In-play behaves differently across sports, and the differences drive both product design and risk.
Football is low-scoring, which makes each goal enormously significant to prices. A single goal can move a match odds market dramatically, which makes suspension around attacking phases essential and makes the goal itself the dominant risk event. The long periods between goals give traders time, but the moments of change are severe.
Tennis has a rigid structure of points, games and sets, which makes state representation clean and probability calculation relatively tractable. Prices update after every point, which produces very high volumes of small price movements rather than occasional large ones. It is also the sport where courtsiding has been most prominent, because the point-by-point structure creates frequent, discrete, exploitable moments.
Basketball is high-scoring, so individual scores matter less, but the clock and possession structure means the value of time remaining changes sharply in the closing minutes. Late-game situations require specific modelling rather than continuous extrapolation.
Cricket has distinct phases with different scoring dynamics, and match state involves several dimensions at once. It supports an unusually rich set of in-play markets and demands correspondingly sophisticated modelling.
Horse racing offers a short in-play window, which limits volume but concentrates it, and data latency is particularly consequential given how quickly races resolve.
The operational implication is that in-play capability is built sport by sport rather than generally. An operator with strong football in-play does not automatically have strong tennis in-play, and the investment required is separate for each.
Failure modes worth knowing
In-play systems fail in characteristic ways, and each has produced expensive incidents across the industry.
Stale prices after a feed interruption, where data stops arriving and markets remain open at prices that no longer reflect the event. The control is automatic suspension on feed loss rather than continuing on the last known state.
Incorrect event data, where a goal is reported that did not occur or the score is transmitted wrongly. Prices move sharply on false information, customers bet into the error, and the operator faces a settlement problem regardless of which way it resolves. Cross-checking against a second source is the standard mitigation.
Model divergence, where the in-play model produces prices inconsistent with the pre-match model or with the wider market, usually because of a state representation error. Reference price checks catch most of this.
Cascade errors in derived markets, where a mistake in a primary market propagates into every bet builder and combination that references it. This is where a single error becomes a large one.
Reopening too early, where markets return before repricing has completed, recreating exactly the window suspension existed to close.
The common thread is that in-play systems act faster than anyone can supervise, so the safeguards must be encoded rather than exercised. Every operator running meaningful in-play volume has a set of bounds, cross-checks and automatic halts, and the maturity of those controls is a reasonable proxy for how well the operation is run.
The economics of the data relationship
A closing commercial observation, because data is not merely an input.
Live sports data is supplied by a small number of specialist companies, several of which hold exclusive rights to distribute official data for major competitions. That concentration gives them substantial pricing power, and data costs represent a significant and growing expense for any operator with meaningful in-play coverage.
The strategic consequences are worth understanding. An operator without access to the official feed for a competition is structurally slower than one with it, which means it must suspend more aggressively, offer fewer in-play markets, or accept exposure during the latency gap. None of those is attractive, so access to competitive data is close to a precondition for competing in in-play at all.
This has fed into a broader pattern in the sector, where sports governing bodies have increasingly monetised their data rights and, in some cases, sought a share of betting turnover directly. The arguments run in familiar directions: sports contend that betting operators build products on their events and should contribute to the cost of staging and protecting them, while operators contend that data costs already represent a substantial payment and that additional levies simply raise the cost of licensed operation relative to unlicensed alternatives.
For a trading operation, the practical effect is that data is both a technical dependency and a commercial negotiation, and the terms of that negotiation shape what product can be offered at what margin.
What good in-play operation looks like
To close, a short description of the characteristics that distinguish a well-run in-play operation.
Data arrives from more than one source, with automatic cross-checking, so a single supplier error does not propagate unchallenged into prices.
Suspension is automatic, fast and tuned, with thresholds reviewed against actual incidents rather than set once at launch.
Reopening is gated on repricing completion rather than on elapsed time.
Bounds exist on how far automated systems may move prices without human confirmation, so a bad input cannot produce an unbounded response.
Bet acceptance applies customer-differentiated limits encoded in advance, since real-time judgement is impossible at these speeds.
Monitoring and alerting flag anomalies, including unusual acceptance patterns, prices diverging from reference sources and clusters of bets arriving immediately before a suspension.
And there is a post-incident process that examines what happened, identifies which control failed and adjusts it, rather than treating each loss as bad luck.
None of this is exotic. It is ordinary operational discipline applied to a system that moves faster than its operators, and the operators who take it seriously are the ones not appearing in industry conversation about expensive in-play incidents.