What football can learn from Ferrari’s costly strategic mistake in 2010
Formula 1 is arguably the most data-driven sport out there. According to a case study from Amazon Web Services, during each race, 120 sensors on each car generate 3 GB of data and about 1.500 data points per second, which results in billions of data points every Grand Prix weekend.
However, as we all know, being data-driven is no guarantee for success and doesn’t protect you from failure. One of the most prominent failures happened in the final race of the 2010 campaign. Fernando Alonso was leading the Championship ranking, followed by Mark Webber in second and Sebastian Vettel in third position. Alonso needed a top 3 finish if Mark Webber would have won, top 4 if Sebastian Vettel would have won.
Although all odds were in favor of Alonso and despite him having the best car, the Spaniard ended up seventh while Vettel won the race and, consequently, the Championship. Mark Webber finished 8th.
So, how could that happen? As former team principal Stefano Domenicali explained, Ferrari focused too much on Mark Webber, failing to see the bigger picture, and overestimated Alonso’s ability to overtake his competitors on a race-track (Abu Dhabi) they did not know. Consequently, after an early safety car and Webber’s decision in an early pit stop, Ferrari reacted in the same way, which turned out to be the wrong decision as Alonso was not able to overtake his competitors as they expected him to. The man responsible for the decision? Chief strategist Chris Dyer. However, and important to note, Dyer did not make the decision solely by himself but supported by a modern decision support system (DSS), which relied on big data.
The intersection of the human mind and machines
Chris Dyer was offered two options for Alonso’s strategy by the DSS. Based on Ferrari’s decision-making processes, Dyer was not allowed to consider another option and deviate from the DSS but stick to the alternatives offered. Nevertheless, as everyone agreed on in hindsight, the DSS did not offer the best solution, which would have probably secured the title. The reason: The DSS did not incorporate the conditions of the rack-track and thus, the difficulties of overtaking.
Without putting this into a football context yet, Ferrari’s failure is a prime example of the difficulties in the interplay between machines and the human decision-maker. Even statistical algorithms are vulnerable to certain errors and biases as they involve the risk of omission errors, whereby viable alternatives are left out. Besides, it emphasizes the importance of understanding how the models work and which data eventually enters the DSS as there is oftentimes information that the DSS cannot incorporate but the human mind.
In a football context, this can be important when evaluating a player's performances and why he did perform extraordinarily well or, on the contrary, disappointed. In that way, for example, Eintracht Frankfurt was able to acquire Luka Jovic after a difficult period at Benfica or Brentford with Neal Maupay.
Understanding where the problem is and stems from allows you to assess whether it’s something you can solve or not. Or as Daryl Morey puts it: “You have to figure out what the model is good and bad at, and what humans are good and bad at.” The importance of learning how to deal with data limitations while avoiding cognitive biases is also underlined by Liverpool’s Lead Data Scientist William Spearman: “If you have the best scouts and the best analysts, they are going to be better than even the best models because their mental model takes into account many factors that the data doesn’t control for.”
The importance of a coherent organizational structure
On the one hand, Ferrari had a coherent organizational structure to the extent that the data-driven approach was accepted and implemented throughout the entire organization. This puts them already one step ahead of most football clubs today, where the data guys oftentimes work in silos, not being incorporated in the decision-making process. On the other hand, however, Chris Dyer was ultimately fired even though he followed the procedure, sticking to the alternatives of the DSS and even chose the best option available.
This not only indicates Ferrari’s need for a scapegoat but also shortcomings when it comes to the evaluation of a decision as they apparently focused on the quality of the outcome. Annie Duke refers to this process in her book Thinking in Bets as Resulting, which basically says that you draw conclusions from the quality of the outcome to the quality of the decision. Her favorite example is Pete Caroll’s controversial call in Super Bowl XLIX.
A clear and objective evaluation process is essential in Formula 1 as it is in football or every other business in the world. When judging transfer decisions, for example, one should not only be looking at the quality of the outcome but the quality of the decision as explained by Tiago Estêvão. Naby Keita’s time at Liverpool has not been a success story so far, especially due to numerous injuries that never allowed him to fully settle into Klopp’s starting 11. That said, considering his performances at Salzburg and Leipzig, this was arguably a reasonable transfer when the decision was made.
In his book Superforecasting, Philip Tetlock emphasized the importance of keeping track of predictions and decisions. He summarized this process as forecast, measure, revise, repeat. After every decision, the decision-maker has to evaluate it and draw conclusions from it to improve the decision-making process, which can eventually help to enhance both luck and skill. Assuming the aforementioned transfer of Naby Keita is assessed as wrong in hindsight, the club can then figure out the reasons and problems in their decision-making process that led to the assessment.
Know what the critical information is and put them into context
One of Dyer’s most crucial mistakes was to focus on Mark Webber, the closest competitor in terms of points before the race, instead of focusing on the overall goal: winning the Championship. Regardless of whether he would have acted differently, this prevented him from questioning the alternatives of the DSS.
Formula 1 is a prime example of decision-making with DSS under time pressure. When Webber decided to pit early, Dyer only had a few minutes to look at the data, consult with other engineers, and ultimately make a decision. Numerous studies point out the difficulties of decision making in high-velocity contexts, which can be detrimental to human decision-making outcomes as heuristics are used to misjudge or ignore important information (speed-accuracy trade-off).
In a football context, the time-pressure can be most often found during the 90 minutes of a match when coaches have to make numerous decisions in a short period of time. This requires understanding what a coach wants to see and knowledge about which KPIs are actually related to winning the match, which is the overall goal. A field study by Daniel Memmert and Robert Rein, for example, found evidence that the best teams control the most space, particularly in critical zones of the playing field (30m in front of the goal and penalty area). With a pitch control model, one can see where to create space on the pitch and where you might be vulnerable. Or in a transfer context: figuring out the metrics that really help to predict the development of a player and which price is appropriate.
Conclusion
Ferrari’s case does not offer any new revolutionary insights but highlights possible sources of errors and how to not do it. Formula 1 has evolved a lot since 2010 and the decision support systems are far more advanced than they used to be, driving the key decision of the teams. Football has made huge steps to become more data-driven as well although the models will arguably never be as accurate and predictable as they are in Formula 1. That said, they can and should still guide decision-making combined with critical thinking. What it ultimately all comes down to is understanding the strengths and weaknesses of statistical models (and the human mind) and getting the processes right including objective tracking/evaluation.