What a trading desk actually does, how it is organised, and the decisions it makes on the operator's behalf.
In this lesson:
- Describe the responsibilities of a modern trading function and how they are distributed across roles
- Explain the shift from manual compiling to model-driven pricing and what remains genuinely human
- Identify the trade-offs between building pricing capability and buying it from suppliers
- Understand how trading performance is measured and why the obvious metrics mislead
What the desk is actually for
A sportsbook has to answer two questions continuously, for thousands of markets at once. What price should we offer? And now that money has arrived, what do we do about the position we hold?
The trading function exists to answer both. Everything else in this course, the mathematics of margin, the handling of correlation, the management of in-play risk, the treatment of winning customers, follows from those two questions.
It is worth being clear at the outset that these are commercial decisions rather than purely technical ones. A price that is too generous attracts volume and loses money. A price that is too tight retains margin and drives customers to competitors, since betting prices are trivially comparable and customers do compare them. The desk sits permanently between those failures.
The core responsibilities
Pricing covers setting the opening odds on a market and maintaining them until it settles. This includes deciding which markets to offer at all, since every market carries a cost to price, monitor and settle.
Risk management covers the exposure created once customers bet. As money arrives unevenly, the sportsbook's liability across outcomes shifts, and the desk decides whether to accept the resulting position, adjust prices to attract money elsewhere, reduce the stakes it will accept, or lay off risk.
Customer management covers how individual customers are treated: what stakes they may place, whether their bets are accepted automatically or referred for approval, and whether their account is restricted. This is the most contested part of the function and is dealt with fully in a later lesson.
Settlement covers resolving markets correctly once events conclude, which sounds administrative and is not. Disputed results, abandoned matches, rule changes and ambiguous market definitions all generate settlement problems, and errors here produce customer complaints, regulatory attention and direct financial loss.
Market design covers what products exist: which bet types are offered, how they are structured, what limits apply and how they behave. This overlaps with product management and is where a good deal of competitive differentiation actually happens.
How the function is organised
Structures vary with scale, but the roles are recognisable across operators.
Compilers set opening prices. Historically this was a role built on deep sport-specific expertise, with individual compilers responsible for particular leagues or sports and pricing largely from knowledge. That description is now partly historical, since opening prices for most events come from models or purchased feeds, and the compiler's role has shifted towards reviewing, adjusting and overriding model output where circumstances warrant.
Traders manage live markets. They watch money flow, adjust prices, respond to news, monitor competitor movement and manage liability. In-play traders do this under considerable time pressure, since a market may need repricing several times a minute.
Risk analysts work on the quantitative side: building and validating pricing models, analysing customer profitability, assessing whether margin is being achieved and investigating where it is not.
Sport specialists provide depth in particular areas, most commonly in sports where models perform poorly or where local knowledge matters disproportionately.
Settlement and operations handle resolution, disputes and the operational infrastructure the desk runs on.
In smaller operations these roles collapse into fewer people, and in the smallest the entire function may be a thin layer of oversight on top of a supplied feed. In the largest, trading is a substantial department with its own technology, its own analytics capability and its own profit and loss.
From expertise to models
The most significant change in trading over the past two decades has been the shift from expert judgement to quantitative pricing, and understanding what did and did not change is useful.
Models now produce the great majority of prices. They ingest historical results, team and player data, current form, market prices from other sources, and event-specific information, and they output probability estimates that are converted into prices with margin applied. For high-volume sports with abundant structured data, models are considerably more consistent than human compilers and vastly more scalable.
What remains human is the handling of situations models handle badly. Team news arriving shortly before an event. Conditions the model has no variable for. Sports with thin data. Novel competitions without history. Circumstances where something has changed structurally, so that historical data is misleading rather than informative. And judgement about how much to trust the model in any given case, which is itself a skill.
There is also a category of decision that models do not make at all. Whether to offer a market, how much exposure to accept, when to close a market, how to treat a particular customer, and how aggressively to compete on price in a given period are commercial decisions informed by data rather than determined by it.
The practical consequence is that the skill profile of the function has changed rather than diminished. A modern trader needs to understand what the model is doing, recognise when it is likely to be wrong, and act on that recognition, which is a different capability from pricing a football match from scratch but not a lesser one.
Build or buy
Almost every sportsbook faces a decision about how much pricing capability to own, and the answer is usually a mixture.
Buying feeds provides prices for a very large number of events at predictable cost. Specialist suppliers price enormous catalogues, and for the long tail of minor competitions this is the only economically sensible approach, since the margin those events generate would never cover the cost of pricing them independently.
The disadvantage is that a sportsbook running purely on supplied prices offers essentially the same odds as every competitor doing the same, which removes price as a point of differentiation and means it holds no edge anywhere. It is also exposed to the supplier's errors, and to the fact that sharp customers know exactly which operators run on which feeds.
Building capability allows an operator to price differently where it believes it can price better, to react faster than the feed, and to offer markets competitors do not. It requires quantitative staff, data infrastructure and continuous model maintenance, which is a substantial permanent cost.
The common resolution is to buy the tail and build for the core: proprietary pricing on the highest-volume competitions and bet types where volume justifies the investment, supplied pricing everywhere else, and human oversight concentrated where errors would be most expensive.
What the desk watches
The information flow into a trading operation is worth describing, because it explains the rhythm of the work.
Money taken by market and by outcome, which drives liability. Customer identity behind that money, since a given stake means something different depending on who placed it. Competitor prices, monitored continuously, because a price materially out of line with the market attracts volume for reasons that are usually bad. Exchange prices where available, since betting exchanges provide a market-derived probability estimate that many desks treat as a reference. News and information, particularly team news, injuries, weather and anything affecting participation. Model output and its confidence, and event data for anything in play.
The interpretive skill lies in weighing these against one another. Heavy money on one outcome might mean the price is wrong, or might mean a popular team is attracting recreational support, and the two require opposite responses. Distinguishing them is largely a matter of who is betting, which is why customer classification sits so close to the centre of the function.
Measuring the desk
Trading performance is unusually easy to misjudge, and the reason is variance.
The obvious measure is realised margin: what proportion of turnover the book actually held. Over a full year, across a large volume of bets, this converges towards the theoretical margin and is informative. Over a month, it is dominated by results. A desk that priced everything correctly can hold well below theoretical margin because favourites won, and a desk that priced badly can hold above it because they did not.
Judging trading on short-term profit therefore rewards luck and punishes correct decisions that happened to lose, and it creates an incentive to price conservatively and accept less volume, which is not what the business wants.
Better measures separate decision quality from outcome. Expected margin against realised margin over meaningful periods shows whether pricing is systematically off. Closing line comparison, assessing whether the desk's prices at the point a market closed were accurate relative to the eventual result across many events, is a standard method for evaluating pricing quality independent of individual results. Volume at expected margin captures whether the book is attracting business at acceptable prices rather than simply protecting itself. Error rates in pricing and settlement measure operational quality directly.
The general principle is one worth carrying beyond this function: in any activity where outcomes are heavily influenced by chance, measuring decisions rather than results is the only way to assess skill over any short horizon.
The commercial tensions
Three tensions run through the trading function permanently, and recognising them explains most of the disagreements the desk has with the rest of the business.
Margin against volume. Marketing wants competitive prices to attract customers. Trading wants margin. Both are right, and the resolution is a commercial judgement about where the operator wants to sit, which should be made explicitly rather than fought out market by market.
Risk against acceptance. Restricting stakes protects margin and frustrates customers. Accepting everything maximises volume and exposes the book to customers who are consistently better at this than it is. Where the line sits is a strategic choice with reputational consequences.
Speed against control. In-play trading rewards fast reaction and punishes stale prices, but automated systems reacting fast can also propagate errors at speed. Every control that prevents a costly mistake also slows the response that generates margin.
None of these has a permanent answer. They are managed rather than solved, and how an operator manages them defines what kind of sportsbook it is.
How a market reaches a customer
It is worth following a single market through its life, because the sequence explains where each part of the function contributes.
Event creation. The fixture enters the system, usually from a data supplier feed, with its participants, start time, competition and status. Errors here propagate everywhere downstream, and mismatched event identifiers between suppliers are a persistent operational nuisance.
Market generation. The sportsbook decides which markets to offer on this event. For a major football fixture that might be several hundred, from match odds through to individual player props. For a minor fixture it might be three. This is largely templated by competition tier, with manual additions for events warranting them.
Opening prices. Models or supplied feeds produce initial prices. A compiler reviews those on events where the stakes justify attention, adjusting for information the model lacks.
Publication. Prices go live across the website, apps and any partner channels. Margin is applied at this stage according to configured rules that vary by sport, competition and market type.
Live management. Money arrives, liability accumulates, competitor prices move, news breaks. Traders and automated systems adjust. Some markets are suspended and reopened repeatedly.
Suspension and closure. Markets close at kick-off or, for in-play markets, at defined points during the event.
Settlement. Results arrive, markets are resolved, winnings are paid. Disputes and voids are handled here.
Review. Realised margin is compared against expected, errors are logged, and anything that went wrong feeds back into pricing rules or model adjustments.
Most operational problems in a sportsbook can be located precisely in one of these stages, and diagnosing them starts with asking which stage failed rather than treating the symptom.
Automation and where humans remain
The proportion of trading decisions made automatically has risen enormously, and the direction is one way. Understanding the division of labour matters for anyone working in or alongside the function.
Fully automated in most operations: opening prices for the long tail, routine margin application, price movement in response to accumulating liability within defined bounds, suspension triggered by event data, settlement of unambiguous results, and acceptance or rejection of bets against configured limits.
Automated with human oversight: pricing on major events, in-play price movement on high-volume markets, and customer limit adjustments driven by profiling models.
Predominantly human: pricing where data is thin or circumstances are unusual, decisions about which markets to offer on novel events, handling of disputed settlements, judgement about whether a suspicious pattern warrants action, and any decision with reputational implications.
The risk that comes with automation is speed of error propagation. An incorrectly configured rule can misprice thousands of markets in seconds, and a system reacting automatically to bad data can move prices dramatically before anyone notices. This is why control layers matter: bounds beyond which automated movement will not proceed without approval, sanity checks against reference prices, and alerting on anomalous patterns.
The general lesson is that automation shifts the failure mode rather than removing it. Manual operations fail slowly and locally; automated operations fail fast and everywhere, which makes the controls around them a genuine part of the trading discipline rather than an IT concern.
Where trading sits in the business
Trading does not operate in isolation, and its relationships with adjacent functions determine how effective it can be.
Product decides what the sportsbook looks like, how markets are presented, how bet builders are constructed and how quickly a customer can place a bet. Trading determines what can safely be offered within that. When the two are disconnected, the usual symptom is a product team launching a feature the desk cannot price or control.
Marketing wants competitive prices, generous promotions and headline offers. Every price boost is margin deliberately surrendered to attract attention, and every free bet creates liability. Well-run operators treat promotional pricing as a joint decision with an agreed budget rather than as marketing spending trading's money.
Compliance constrains which markets may be offered, which events may be priced and which customers may be accepted. Integrity obligations and market restrictions land on the desk operationally.
Customer operations deals with the consequences of trading decisions, particularly restrictions and settlement disputes. A desk that restricts accounts without the support team understanding why creates avoidable friction.
Finance needs forecasts, and forecasting sportsbook revenue requires understanding expected margin, seasonality and the variance around both. Trading is the source of that understanding.
The recurring failure across operators is treating trading as a technical black box that produces prices. It is making commercial decisions continuously, about competitiveness, about risk appetite and about which customers the business wants, and those decisions belong in commercial conversations rather than being made by default inside a function nobody else engages with.
Key takeaways
- Trading exists to set prices that attract volume while retaining margin, and to manage the risk those prices create once money arrives.
- The function has shifted from expert judgement to model-driven pricing with human oversight, but judgement still governs the situations models handle worst.
- Most sportsbooks buy pricing for the long tail of events and apply their own capability selectively where they believe they hold an edge.
- Trading performance measured over short periods is dominated by results rather than skill, which makes it one of the easiest functions in the industry to misjudge.
- The desk's decisions about which customers to restrict shape the operator's public reputation as much as its margin.