xGenius, page 6
What was once a set of niche and widely mocked metrics has now become a go-to tool for foxy top-tier managers and players.
Tuchel was also intrigued by Smartodds’ assessment of Borussia Dortmund’s performances so far that season. Jürgen Klopp’s team had gone into the Bundesliga’s winter break in the relegation zone, but Smartodds’ data showed that they had been extremely unlucky. Tuchel ended up taking over from Klopp at Signal Iduna Park a few months later and kept in loose contact with some of the analysts he had met at Benham’s offices, calling on them to provide qualitative analysis and in-match data on an informal basis during his stint at the club.
Tuchel’s xG education in the Smartodds offices clearly made an impression, given his comments made several years later following his Chelsea team’s defeat to Arsenal. Tuchel’s interview came the day after Ralph Hasenhüttl had used Expected Goals to defend Southampton’s performance in their defeat at Burnley. The list of managers who have used xG to justify their own team’s results over the last few years continues to grow: Arsène Wenger, Frank Lampard, Mikel Arteta, Gareth Southgate, Dean Smith, Thomas Frank, and Graham Potter, among countless others. The last three on that list are unsurprising inclusions, having all been schooled in the Brentford and Brighton schools of xG analysis.
Players have also started publicly mentioning Expected Goals. James Maddison has told interviewers of his desire to increase his xG figures. After scoring for Leicester City against Chelsea, Maddison spoke of how his manager and a member of the analytics department had spoken about where goals come from based on xG. ‘Me, the gaffer, and Jack the analyst – Jack will be buzzing that I’ve name-dropped him – sat down and looked where I could get more goals,’ he told the assembled media. Meanwhile, Jack Grealish once noted that he had the most Expected Assists in the league. The England star famously told an interviewer he didn’t know what an ‘encyclopaedia’ was, but has no trouble rattling off xA figures. What was once a set of niche and widely mocked metrics has now become a go-to tool for foxy top-tier managers and players.
The science of winning football matches is what pundits and the media are fundamentally interested in. Who are the best teams and players? Why are they the best? How can others get as good as them? The media are getting better at answering these questions as the tone becomes increasingly analytical. The fundamental difference between a fox and a hedgehog, and perhaps the ultimate difference between a good analyst and a bad analyst, is that the former sets out to prove themselves wrong. Foxes are always stress-testing their systems and changing their opinions; meanwhile hedgehogs are usually one-sided in their argument. The increasing number of foxes appearing in the media won’t necessarily mean the end for the Roy Keanes of this world. Quite the opposite. As the tone gets more analytical, the importance of light-heartedness will also grow. The utopian ideal is to blend analysis and insight with passion and debate. A world where foxes and hedgehogs live alongside each other in harmony.
Chapter Summary
Prediction-makers can be categorised as either hedgehogs or foxes depending on their open-mindedness and ability to incorporate new information into their models.
The nature of football punditry means pundits often display hedgehog tendencies, although sports media is becoming foxier.
Managers and players have started talking about xG more openly, which has helped analytics reach a wider audience.
Adopting a fox-like mentality is crucial to unravelling the science of winning football matches.
A blend of serious analysis aligned with warm and lively debate is the best way for the popularity of xG to continue to grow.
Note
6 ‘The Goal Probability Philosophy’ doesn’t have as nice a ring to it.
4
The Data Consultancy
The Brains Behind the Analytical Revolution
‘War is 90 per cent information’
Napoleon Bonaparte
You arrive at the Smartodds office at 11:30 am, as agreed. Upon entrance you’re greeted by a small reception area. To the left you spy a glass-walled meeting room housing an oval table and a whiteboard, on which several maths equations have been crudely scribbled with a marker pen. Next to the meeting room stretches an unextraordinary corridor. You don’t know it yet, but at the end of this hallway is a larger room, an office space in its own right, where Brentford’s analytics department carry out perhaps the most profound and insightful football analysis anywhere in the world.
You’re greeted by an employee of the company. ‘Very nice to meet you,’ he says. ‘Please follow me.’
He leads you down the stairs on the other side of the reception area and on to a vast trading floor. You walk through a corridor of desks, each one housing at least four large computer monitors. The men sitting here are fully dialled into their screens. A woolly mammoth could walk in through the front door and go unnoticed. These are Smartodds’ clients, professional gamblers who pay to access the company’s broad range of tools and data services. The setup is what you imagine the trading floor of an investment bank might look like, albeit with khaki shorts and t-shirts replacing the Armani suits. This collection of desks precedes a separate group of desks at the end of the room, with less impressive tech and only two monitors per station. These worktops are where the data collectors sit, logging stats which feed into the company’s advanced algorithms. Framed football jerseys and other items of memorabilia line the bare brick walls. It’s approaching midday, but hardly anyone is positioned at these worktops. It’s lunchtime on a weekday, so not many football matches are on and in need of analysing.
The man stops at one of the desks. ‘Please have a seat.’
You applied for a job at Smartodds on a whim after reading about the vacancy online. The job specification offered a vague description of the role: you’d be contracted as a data collector for a company that specialises in providing in-depth quantitative and qualitative research and analysis in football. Whatever that means.
The process to get you this interview was fairly simple. The company sent you a 31-page document entitled the ‘Smartodds Watcher Handbook’. The front cover showed a picture of an anonymous footballer lying down on his back looking up at the sky in dismay, presumably having just missed a big goalscoring chance. Underneath this image was written in red writing, ‘Confidential – Property of Smartodds.’ The document introduced itself by saying it provided ‘a comprehensive guide to the watching process and outlines what Smartodds expects for any person who provides watching services for the company.’ It was, in essence, a manual for how to assess the danger of goalscoring opportunities in football matches.
The first page of the document gave a brief history of the Watcher Operations team. The team was formed in 2006 to give real-time match analysis to its clients. They offered a detailed reflection of how games were developing, primarily through providing live data such as shots on target, shots off target, corners, and possession percentages. ‘However, such statistics are in many respects restrictive and ambiguous,’ the manual explained. ‘For example, shots off the post or glaring misses from three yards are classified as shots off target – the same category as a harmless stray shot from 30 yards. Consequently, a new process of rating the danger of attacks was developed with the Watcher Operations Team.’
The method of rating attacks was defined by five core classifications:
Delivery: Any meaningful attack should be logged as a delivery. Examples include any time the ball is dribbled into the opposition box but no shot takes place, shots from outside the box which fly harmlessly over the bar, or efforts at goal which are easily blocked by defenders. Attacking set pieces also count as deliveries, including all corners, long throw-ins, and free kicks inside the opponents’ half, regardless of whether or not these situations were under- or over-hit.
Half Chance: One step up from Delivery. There was an opportunity to score but nothing particularly dangerous. Examples include a shot from the edge of the box which would have required a keeper error to go in, a header from 12 yards which an attacker should be able to direct goalwards but there are other factors going against him, or any free kick from around the edge of the box.
Chance: One step up from Half Chance. There was a decent opportunity to score. Examples include if the goalkeeper makes a good save (either at full stretch or fingertips), a shot from 18 yards which went narrowly wide but would have gone in if the other side of the post, the attacker was going to shoot from a dangerous area but the defender makes a last-ditch tackle, or a big goalmouth scramble where the ball could have gone anywhere but no one has real control of the ball.
Oooh: One step up from a chance. An extremely near miss, where there seemed to be at least a 50 per cent chance of a goal being scored. Examples include penalty misses, one-on-ones, an awkward last-gasp clearance off the line, a cross flashed across goal which narrowly misses the boot of a striker from close range, or any attempt which hits the post or crossbar. This rating gets its name from the sound that can be heard inside a stadium when a team misses a glorious chance.
Goal: Self-explanatory.
The guide also instructed you to award ratings based on controversial refereeing decisions. If an attacker is tackled clumsily in the box but no penalty is given, that merits the awarding of a Half Chance to the attacking team. If a team score a goal but VAR rules it out for a marginal offside, a Chance should be awarded. If the referee fails to play advantage when the team are about to be through one-on-one, an Oooh should be logged. In this sense, the way Smartodds assess chance creation is superior to standard xG models, which would log these situations as 0.00(xG) as no shot has actually taken place.
The Watcher Handbook is keen to stress that only one rating can be provided from an individual attack. Suppose two shots happen in quick succession that each merit the awarding of a Chance. Rather than logging two separate chances, you should upgrade the whole sequence to one Oooh. The Expected Goals method deals with sequences which feature more than one shot in a similar way. Consider a Manchester City penalty which took place against Watford late in the 2019/20 season. Raheem Sterling’s spot-kick was saved by Ben Foster, but the ball rebounded to Sterling who tapped in from close range. Penalties are always worth 0.77(xG) and the rebounded shot was worth 0.90(xG). As such, an analyst might be tempted to award Manchester City 1.67(xG) from this sequence, but this would be wrong because it’s impossible to score more than one goal from a single attack. So how does xG deal with this problem?
The key is working out the probability that the attack doesn’t result in a goal. In other words, how likely is it that Manchester City don’t score either of these chances? The maths behind this particular situation is as follows:
(1 − 0.77) × (1 − 0.90) = n
0.23 × 0.10 = 0.02
So there’s a 2 per cent probability that Manchester City don’t score either chance. Therefore:
1 – 0.02 = 0.98(xG)
So Manchester City should be awarded 0.98(xG) for this attack.
Smartodds’ data collection system essentially measures the danger of each attack which takes place, and can be considered a variant of Expected Goals. In many ways, it’s more powerful than standard xG, which only measures shots that actually take place. A situation where a striker rounds the goalkeeper but runs out of room and ends up dribbling the ball out of play won’t be registered in xG. Neither will a ball flashed dangerously across the face of goal that narrowly eludes an on-rushing forward. Smartodds’ watchers are trained to log such situations as Chances or Ooohs, meaning these dangerous attacks are accounted for in the dataset.
Along with the handbook, the company also provided 100 video clips of goalscoring opportunities and tasked you with assigning a rating to each one based on the danger scale outlined above. You’d worked your way through the clips over the course of the last week or so. The short videos came in sets of five and after each set, Smartodds’ bespoke software would provide iterative feedback on your performance. If you had correctly assessed how dangerous the scoring opportunity was, you’d get a green tick next to that clip. If you’d made a mistake, the computer would tell you what you’d done wrong:
Sorry, you incorrectly marked this ‘Chance’ as an ‘Oooh.’ Notice how the player’s heavy touch narrows the angle and reduces the danger of the attack.
That’s correct, this attack was a ‘Delivery’ that carried little threat.
Unlucky, that opportunity is actually a ‘Half Chance.’ Look at the number of defenders between the striker and the goal.
Clearly your assessment of these 100 chances had been accurate enough to advance you to the final stage in the process. You take a seat at the desk as the ‘interviewer’ turns on both of your screens. On the left-hand monitor he loads up a full replay of a match between Heidenheim and Kaiserslautern in the second tier of German football. On the right-hand monitor, he loads up the company’s internal data-collection system. You’re tasked with watching the entire game and rating every attack that happens according to the danger scale outlined in the handbook. Afterwards, your performance will be assessed and, if you’re accurate enough in separating your ‘Chances’ from your ‘Ooohs’, you’ll be offered a contract with the company. This is certainly different from any job interview you’ve ever done before.
You complete your task, clicking the ‘Delivery’ button every time a weak attack takes place, the ‘Half Chance’ button every time a long-shot ends up comfortably wide, and so on. The match finishes 3-1 to Heidenheim, but the system at the end of the game shows that Kaiserslautern had actually created more Half Chances, Chances and Ooohs than their opposition. A few weeks later you receive an email. You are now officially a Watcher at Smartodds.
World-Class Prediction-Making
Over the next year or so, you watch hundreds of matches for the company. You’re paid £20 per game, with an additional £20 if the match goes to extra time. You watch up to four matches per day, taking small breaks in between games to re-energise yourself with a Mars bar from the vending machine before embarking on another 90-minute stint. Some of the games are recorded matches which took place over the course of the previous week. The stats for these matches presumably go into the large bank of data that the company stores. Other matches are live. These games are particularly interesting to work on because the data feed is sent in real-time to the company’s clients, football traders sitting just across the room from you. They are also watching the game live and following the Deliveries, Ooohs and so on which you’re logging. This data helps inform their opinions and predictions. Alongside your logging of the company’s bespoke statistics, you are tasked with providing a brief update on the flow of the match every 15 minutes. You’re asked to note which team is on top, how cagey the game has been, any tactical insights you can give, and other information that might provide context to the raw stats.
Most of the games you watch are from obscure leagues. The company houses satellites that pick up the feeds from a number of different countries worldwide. One minute you’re watching Slovan Liberec take on Slavia Prague in the Czech first division, the next you’re logging data as Farense play Atlético CP in the Portuguese second tier. You watch a fair amount of lower-league football from England as well and occasionally even getting assigned to a low-level Premier League clash.
The unique proprietary data which Smartodds collects is the foundation of their success. The business has built up a bank of chance-quality data which tells them the danger rating of every attack which occurs in professional football. From this, they can build a picture of how good every team is at creating and preventing goalscoring opportunities. They have an accurate gauge of the attacking and defensive ability of every team. Smartodds were one of the first organisations to uncover the xG treasure chest, and have been living off the riches ever since.
The company’s models do also incorporate factors besides the ability of teams to create dangerous opportunities. Smartodds found that a team’s chances of winning away from home were slightly raised if they had a shorter journey to the game. This means local derbies are marginally easier to win than games on the other side of the country. A network of international informants also provide ‘soft’ information which is factored into Smartodds’ predictions. Dressing-room strife or an injury to a key player could shift the odds of a team’s success.
The brightest guys in football don’t tend to originate from football. They’ve moved over from other industries – banking, technology, consulting and so on. One industry stands alone from others in terms of how many football innovators it’s produced: the gambling industry. These people aren’t your amateur gamblers, the blokes down the pub placing a bet on Saturday afternoon to receive their weekly dopamine hit. These are the professionals. The ones who are using sophisticated algorithms to guide their bets, who approach football betting as a trader would approach the stock market. Matthew Benham, the owner of Brentford and Smartodds, once said, ‘I never bet for fun or the thrill. We are all about calculating probabilities with the help of mathematical models.’
Benham studied physics and worked as a derivatives trader in the City. It was there he learned the art of exploiting inefficient markets, a skill he would later bring to sports betting. Seeking out undervalued propositions is the key to succeeding in any market, whether that be of the financial, betting or player transfer variety. Just down the road from Smartodds’ Kentish Town office resides Brighton & Hove Albion owner Tony Bloom’s Starlizard, a company with a near-identical structure and business model to Smartodds. Both of these men live and die by the guidelines outlined in this book. They fully understand the science behind winning football matches and are incredibly skilled forecasters of footballing outcomes. Their data consultancies are the brains behind the decisions being made at their clubs, and are the primary factors why both Brentford and Brighton have been able to haul their way up the Football League ladder. The one rule in the Smartodds offices is that you’re not allowed to celebrate a goal. There could be someone sitting at an adjacent desk who has lost thousands because of that goal. But the thing is, none of these bettors are likely to cheer a goal anyway, nor would they be likely to commiserate one. Their approach to betting is like trading. They work on the percentages. They understand luck and probability, which arms them with a thick skin to the whims of chance and protects them from emotional misguidance. But above everything else, they know the science of what wins football matches, and thus know that the odds are in their favour. To get worked up over a single goal is as unproductive as a casino manager getting worked up over a single hand of black jack.
