27 October 2026 | iGaming Industry News |
For most football followers, a match ends when the final whistle blows. For Tony Bloom’s analysts, however, the score is just one piece of a much larger puzzle. How well did each side play? What did the result reveal about their underlying strength? And did the betting market reach the right conclusion when setting its odds?
Brighton & Hove Albion chairman and professional gambler Tony Bloom built Starlizard around the business of researching sporting probabilities. Established in 2006, the company applies sophisticated analysis to a deceptively simple challenge: identifying when the odds available in the betting market offer value.
Starlizard’s own description provides a useful outline of the business. It gathers information from events both on and off the pitch, develops predictive models and uses technology to identify potential betting opportunities. Its integrity division describes the core business as turning millions of data points into betting probabilities. The detailed algorithms and their weightings remain closely guarded.
A major part of this approach involves understanding performance rather than focusing solely on the final result.
A 2016 Business Insider investigation, republished by The Journal, described researchers watching matches, recording chances and collecting information about team news, playing conditions and even team morale. It outlined a four-stage process consisting of data collection, statistical modelling, bet selection and, finally, execution.
That reporting provides a historical perspective on the operation rather than a verified account of its current internal structure.
Consider a hypothetical 1–0 defeat. One side might have created several excellent chances while conceding a goal from their opponent’s only dangerous attack. Treating that match simply as a loss overlooks valuable information. Analysing the quality of the chances created and conceded provides a more complete picture of the performance and potentially a stronger basis for predicting the team’s upcoming fixtures.
The principle will be familiar to anyone who follows expected goals (xG), a metric that estimates the probability of a scoring chance resulting in a goal. However, Starlizard’s private system should not be assumed to be identical to publicly available statistical models.
This distinction matters because football is a relatively low-scoring sport. A single deflection, penalty or goalkeeping error can decide a match without revealing much about either team’s longer-term quality.
A useful forecasting model must separate consistent ability from short-term fluctuations. Otherwise, it risks overvaluing a fortunate winning side while undervaluing a stronger team whose recent results have been disappointing.
The next challenge for Bloom’s operation is determining how much weight to give each piece of information.
The absence of a key striker, for example, could reduce a team’s attacking threat. However, the extent of that impact depends on the replacement, the opposition and the team’s playing style. A tactical adjustment might even compensate for the missing player.
More information is only valuable if it is interpreted accurately.
The 2016 investigation described quantitative analysts combining information into models that estimated the probabilities of different scorelines. Those estimates then helped inform assessments of potential betting opportunities.
Starlizard also states that it continually adapts its models using large volumes of real-time data.
There is an important difference between fitting a model to yesterday’s results and successfully forecasting tomorrow’s matches.
A model can appear impressive when tested against the same matches used to develop it, yet perform poorly when applied to new fixtures. The broader statistical challenge is determining whether identified patterns remain useful beyond the original sample, while accounting for changes in teams, managers and playing styles.
Crucially, predicting the most likely winner and identifying a good bet are two different tasks.
Imagine a model gives a team a 50 per cent chance of winning. Its fair decimal odds would be 2.00.
If a bookmaker offers odds of 2.20, a successful £10 bet returns £22, including the original stake.
Assuming the model’s probability estimate is accurate, the expected return across a large number of equivalent bets would be £11 for every £10 staked. That represents an expected profit of £1 per bet before costs.
This illustrates the concept of value betting, but it does not guarantee a return.
The team could still lose, and a sequence of losing bets would remain entirely possible. If the same team were available at odds of just 1.80, the model would instead imply an expected loss, despite assigning the team exactly the same probability of winning.
The opportunity therefore lies in the relationship between probability and price.
A strong favourite can represent poor value, while an outsider can sometimes offer good value.
The betting market may already know that a club is experiencing an injury crisis or facing a particularly difficult fixture schedule. The important question is how accurately the available odds reflect that information.
Simply identifying negative news about a team does not automatically create an advantage.
If other bettors have already reacted, the odds may have adjusted sufficiently to eliminate the opportunity. In some circumstances, the market may even have overreacted, pushing the price further than the available evidence justifies.
Turning these assessments into a potentially profitable operation also requires precise execution.
Starlizard’s website explicitly identifies bet execution as part of its business, alongside data collection and analytics.
Its integrity division identifies Asian betting markets as a focus of its football market analysis. It compares market prices with its own predictions, including assessments based on expected goals and the relative strengths of competing teams.
This offers a public glimpse into the company’s analytical framework, although its integrity service serves a different purpose: investigating potentially suspicious betting activity.
In any betting market, an attractive price has limited practical value if only a small amount can be wagered at those odds.
Prices can change as new information becomes available or as bets are accepted. Transaction costs, available liquidity and the impact of a bettor’s own activity can all influence whether a theoretical advantage survives in practice.
Risk management is equally important.
Even a well-calibrated probability estimate leaves considerable room for an unfavourable outcome. Several bets involving the same team or competition may also share underlying risks, making them less independent than they initially appear.
Maintaining a statistical edge therefore requires sufficient capital, careful risk management and the discipline to withstand periods of uncertainty.
The probability estimates themselves can also be wrong.
An apparent advantage may disappear if a model systematically underestimates an opponent or places too much importance on an individual player.
Evaluating forecasts therefore requires more than simply counting winning bets.
It also involves examining whether events assigned a particular probability actually occur at approximately that frequency across a sufficiently large and relevant sample.
Bloom’s association with football analytics also extends into player recruitment through Jamestown Analytics.
Hearts’ announcement of its partnership with the firm in November 2024 identified existing relationships with Brighton, Union Saint-Gilloise and Como. It also stated that the data would support both recruitment and opposition analysis.
The conceptual link between betting and recruitment is valuation.
A betting operation asks whether the available odds accurately reflect a team’s chances of winning. A recruitment department asks whether the transfer market has correctly assessed a player’s potential contribution.
Both seek to identify useful information that the market price may not fully reflect. In each case, success depends on sound judgement and effective implementation.
This helps explain the appeal of building an entire enterprise around analysis.
Skilled researchers, reliable information, sophisticated software and accumulated experience are difficult to replicate with a spreadsheet and a weekend’s worth of results.
The potential advantage comes from the combined system rather than any single statistic.
Its value depends on how consistently people can turn evidence into informed decisions and, just as importantly, learn when those decisions prove incorrect.
The broader lesson of Bloom’s approach is that football’s unpredictability can be examined more closely through systematic analysis.
Better information can improve probability estimates, while more accurate estimates may reveal opportunities that the betting market has overlooked.
Neither, however, removes uncertainty.
The potential edge comes from the entire process: observing carefully, testing assumptions, assigning probabilities and acting only when the available price justifies the risk.
Every result then provides another opportunity to examine whether the model accurately understood the fixture and what might be improved before the next one.
