Risk, Process and Outcomes: What Football Can Teach Risk Teams
Updated: 5 days ago
Stories etch grooves deep enough for people to follow in the same way that water follows certain paths down a mountainside. And every time fresh actors tread the path of the story, the groove runs deeper. Terry Pratchett, Witches Abroad
TL:DR; Football analysis is terrible. It is riddled with outcome bias, causal bias, and narrative fallacy. Football is particularly vulnerable to these biases because of its low scoring nature and high level of randomness in the goals that are scored. This creates a disconnect between team performance and team results.
By examining where football analysis goes wrong we can learn a lot about how organisational analysis – and risk analysis – can falter. The lesson is that good organisational analysis recognises the difference between good performance and good outcomes, it creates and tests theories on what improves performance, and it seeks out new data where existing information is insufficient.
I love football. I loved playing it, until my body told me it had had enough. I love watching it. What I hate is the punditry.
A big team's late winner is proof of “character” or a “winning mentality”, even when the underdog has repeatedly hit the woodwork and had goals dubiously ruled out by the referee. These "narrative" explanations are emotionally rewarding, but they have little predictive value because they confuse performance and outcome.
Persuasive narrative explanations can also make it easier to believe things that aren't true. For instance, so-called "Fergie time" - Manchester United's perceived habit of winning games deep into injury time - is largely a work of fiction. During Sir Alex Ferguson's twenty-year tenure at Man U, they scored just sixteen 90+ minute winners in the premier league. Four teams scored more over that period.
By examining where football analysis goes wrong we can learn a lot about how organisational analysis – and risk analysis – can falter.
The views and opinions expressed on this account are my own and do not reflect the official policy or position of my employer. Any content provided is for informational purposes only and should not be considered or relied upon as professional advice.
Football contains a lot of randomness
“Football is a low-scoring game, the lowest-scoring of all the popular team sports. This lack of goals increases the role of luck, makes the underdog win more often and tends to become more pronounced when teams of equal standing meet in a high-pressure situation such as the latter stages of an international competition[i].”
A study of over 7,000 Premier League goals found that random events were relevant to almost half of them[ii]. It examined deflections, rebounds, goalkeeper errors, defensive errors, referee errors and own goals. Randomness was more pronounced for weaker teams and when the score was level.
The study probably understates the uncertainty. It excludes goal-bound shots blocked accidentally, awkward bounces, injuries, dismissals, suspensions, weather, and cramp[iii]. It also doesn’t consider secondary sources of randomness that crystallise before the game is even played, such as player suspensions, transfers, bereavements and game schedules. An Oxford academic calculated that for a season to be totally “fair” – i.e. negating these secondary sources – a season would have to be 35 years long.
The prevalence of randomness in the scoring of goals does not erase skill - the best teams are much more likely to beat the worst teams. But they do make the link between performance and result far less certain, and make explanations that seek to explain the result far less useful for predicting future outcomes.
Luck is emotionally unsatisfying, so commentators seek to minimise in their explanations. When they can’t ignore it they seek to rationalise it as part of their broader narrative: “you make your own luck” or “it evens out over a season”.
Pundits’ three favourite biases
Outcome bias
Outcome bias arises from judging a decision by its result rather than by the quality of the processes and decisions that sit behind it. In football, goals, league position and recent results are highly visible results, but they can be poor indicators of quality, or predictors of future performance. Chance creation, shot quality and long-term capability are often more informative.
For instance, league positions are not statistically relevant for a number of weeks, and early season performance contains a lot of noise. Nevertheless, commentators often construct sweeping narratives after only a few games[iv].
Causal bias
Causal bias appears when an action is assumed to result in its intended outcome, even where the evidence for that assumption is weak or non-existent. If a football manager brings on a striker, and they score, the substitution is declared inspired, and the manager congratulated. We notice the success, but have no counterfactual to prove the effectiveness of the decision.
Models that seek to isolate the individual contribution of a football manager find that they can influence only between 5% and 10% of the team’s performance efficiency, equating to roughly 5 to 10 points a season[v]. This is not immaterial, but indicates how the amount of attention manager decisions get is disproportionate relative to their impact. Football analysis focuses on the visible actions of the manager (such as squad selection, formation, and substitutions), whilst ignoring the many invisible actions of the management staff (defensive coaches, offensive coaches, goalkeeper coaches, fitness coaches...).
Narrative fallacy
The narrative fallacy reverses the proper order of scientific analysis: observe the result, construct a coherent story, then present that story as the cause. Because the explanation emerges after the event, it is often difficult to falsify, but has limited predictive value.
Football is especially vulnerable to the narrative fallacy:
Results are highly visible
Randomness matters
Audiences want simple explanations,
Audiences have often not watched the full match,
Pundits and audiences have little access to the intended strategy or decision process.
Explanations linked to “destiny”, “character” and “knowing how to win” make better television than a discussion of probability. Narratives are entertaining, statistics less so. “Nobody buys a season ticket to watch a pair of balance sheets represented through the medium of 22 players kicking a ball[vi].”
Brentford FC: a case study in better analysis
Useful analysis should be falsifiable, forward-looking, and have predictive value. It should rely on metrics that can prove or disprove theories for achieving better outcomes. The metrics commonly used in football analysis, such as goals scored, goals conceded, or home form do not meet this standard.
Brentford FC[vii] offers a practical example for how to construct better analysis. Under owner Matthew Benham, the club treated football as a decision-under-uncertainty problem. It used data and probability theory to challenge conventional wisdom, identify undervalued players, and judge performance through measures such as expected goals, pressing intensity, progression and chance creation. These measures were linked to an underlying theory of good process and their effectiveness was statistically proven.
Brentford’s analysis found that orthodox football management strategy tended to overvalue players based on reputation and recent performance, judged managers by short-term results, assumed former players made good coaches, and judged performance based purely on outcomes. Does any of that sound familiar?!
Leicester City: a case study in how narratives take over
Eventual winners Leicester City began the 2015/16 Premier League season as 5,000-1 outsiders, before pulling off one of sport’s greatest upsets. Following their victory, pages upon pages of gleeful commentary appeared “analysing” how the betting companies could have got it so wrong. Explanations centred on tactics, management, recruitment, and team spirit. Much of the commentary described what happened, rather than offering a model that could have predicted future performance.
Looking at the numbers, the season was exceptional in several ways:
Leicester won with 81 points, enough to win only one other of the last 20 Premier League seasons. It would have given them an average league position of 2.7 over that period.
They scored the joint lowest number of goals of any premier league winner in the last 20 years, and only four of winners conceded more.
They rose from 41 points the previous season to 81, then fell back to 44 the next year. For context, this is the largest season-on-season positive points swing in premier league history. The next highest was Newcastle in the 21/22 season, which was only 22 points, and followed a £60m year-on-year wage bill increase.
They relied on an unusually small, unusually healthy squad; they used the smallest number of players of any team that season, and ten of their first-team players appeared in at least 32 league matches.
Their wage bill was £30m below the league average, and £90m less than the three largest clubs. Wage bill and past performance are the two best indicators of future league position. The fact that only two Leicester players managed to achieve meaningfully sustained success after the league win, suggests that the wage bill was not artificially deflated due to a shrewd buying strategy.
All of this indicates that betting companies were on good grounds, even with hindsight bias, to judge Leicester as rank outsiders for the title. There was no indication in the lead-up to the 2015/16 season that they would be title challengers, and their performance in later years seems to confirm that this was a freak season. Collectively and individually, the players had the season of their lives. They also needed other teams to underperform at exactly the same time that they overperformed.
It is very hard to put an accurate probability to extreme events, so we can't “validate” the 5,000-1 odds. A model built afterwards estimated Leicester’s pre-season probability of winning at about 20,000–1[viii], suggesting that bookmakers under- rather than over-estimate the odds. If betting companies did make a mistake it may have been to underestimate the true randomness of football: since 2016 the proportion of teams in the “longest odds” bucket doesn’t seem to have reduced, but the longest odds on offer have shifted from 5,000-to-1 to 1,000-to-1.
What this means for organisations
Applied to performance management
Performance analysis is on strongest ground where data is plentiful, the organisation agrees on the theoretical link between good process and outcomes, and there are verified measures for good process. Where these three pillars are not in place, performance analysis can start to look a lot like football punditry: narrative-driven explanations that lack predictive value and assume that outlier events must have significance.
In the absence of “scientific” data to assess good process, organisations can be reluctant to seek out new information sources of unproven value. Data design and collection is rarely high status within organisations because data analysis resource tends to focus on high quality data, or data that is accepted by key stakeholders as being relevant. New, unaccepted data sources are vulnerable to ridicule and low investment.
Applied to risk management
As with performance management, risk analysis is strongest where data is plentiful and the organisation agrees which processes (and associated metrics) meaningfully reduce uncertainty. Good risk management quantifies uncertainty, tests assumptions, and examines whether controls change the likelihood or impact of downside risk.
Where evidence is weak or contested, risk teams face the same traps. We have a bad habit of observing an outcome, building a plausible story, and only occasionally testing its predictive value. A good result can make a weak process look effective, and a bad result can make a sound decision look foolish. Without proper analysis, risk teams can end up helping the business to reach the wrong conclusions.
Outcome measures such as breaches, errors, defaults, incidents, complaints, staff turnover and downtime can be useful, but they are lagging risk signals. Each may reflect control quality, exposure, reporting practices, external conditions or chance. On its own, the outcome does not tell us which explanation is correct.
Worse, we often focus on the outcome of processes that have not been validated: after identifying a potential harm, the business looks to what is already lying around that could concievably mitigate that harm. When this happens, risk teams can end up performing an ineffective management oversight role for tasks of unproven value.
An excessive focus on outcomes can also distort behaviour. Project teams may build contingency into plans or narrow their scope to improve the chance of meeting targets. The reported outcome improves, while the underlying “true” risk may be unchanged, or even increase.
A more useful set of questions
Risk teams should separate the outcome from the process that produced it. For any material risk, we need to build a clear picture of:
What is the uncertainty? This requires an understanding of the processes and decisions that create the uncertainty and, wherever possible, a quantification of that uncertainty.
What processes reduce that uncertainty? We need to create robust, falsifiable theories, with clearly stated assumptions.
What does process success look like? Too often in risk frameworks the definition of success for a control or decision is left ambiguous. We confuse level of effort and intent with quality of process.
How do we measure the performance of those processes, and how do we gain evidence that those processes meaningfully reduce uncertainty? Where measures don’t exist we need to work with the business to develop them. Where there is a cost to collating data this needs to be justified in the context of the estimated uncertainty reduction and the cost of running the process.
Without these questions, risk teams can drift from assessing uncertainty into performance management or activity tracking. A list of successfully completed business tasks may show that people are busy; it does not show that downside risk has reduced.
We also need to embed this question set in our work so that we can help businesses to drive innovation – in both methods of assessment and methods of control. As with Brentford City, we can only drive positive change when we have the freedom and license to collate new types of data, and conduct new types of analysis, even though their business benefit is not yet proven.
In Conclusion
The lesson from football is that good organisational analysis recognises the difference between good performance and good outcomes, it creates and tests theories on what improves performance, and it seeks out new data where existing information is insufficient.
Hopefully that has sparked some ideas. I am hoping to start a discussion with this blog. If you find it interesting, please share! If you would like to contribute or share feedback, please comment below or message me on LinkedIn!
[i] Das Reboot: How German Football Reinvented Itself and Conquered the World,2016, Vintage Publishing, ISBN: 9780224100144
[ii] Wunderlich, F., Seck, A., & Memmert, D. (2021). The influence of randomness on goals in football decreases over time. An empirical analysis of randomness involved in goal scoring in the English Premier League. Journal of Sports Sciences, 39(20), 2322–2337. https://doi.org/10.1080/02640414.2021.1930685
[iii] In 2006, Steven Gerrard scored arguably the greatest goal in FA Cup final history to take the game into extra time against West Ham. He later admitted that because he was suffering from cramp he decided to shoot immediately from 30 yards in injury time rather than bringing the ball down and building an attack
[iv] Anderson, C. J. & Sally, D. (2014). The numbers game: why everything you know about football is wrong. Penguin Books.
[vi] https://www.theguardian.com/football/blog/2023/may/13/leicesters-relegation-battle-is-a-harsh-maths-lesson-for-all-mid-table-clubs
[vii] Alex Duff (2024), Smart Money: The Fall and Rise of Brentford FC, ISBN-13: 978-1408719404
Comments