6 minute read
What I've learnt about business through playing fantasy premier league
/ Read original article here /
During the lockdown of 2020 I started managing a Fantasy Premier League (FPL) team. I needed something to distract me, and to reconnect with friends during the isolation. At the same time, I was also managing Predictive Insights a machine-learning and economics start-up - as we tried to survive the first few years of our business’s life. This is a bit of a ‘fun’ take on the commonalities between the two, and on what I’ve learnt during the process.
FPL is currently played by over 9 million ‘managers’ - for context 26.8 million people in the UK watched live Premier League coverage in 2020/21. Each manager tries to select a combination of English Premier League (football) players who will do well over the coming games and earn points for specific outcomes - like scoring goals or keeping clean sheets.
There is a limited budget, better players cost more, and players change value depending on demand by managers. There are other constraints too:
one transfer of a player is free each week, and each subsequent transfer costs points;
each team needs a balance of players in each position; and
there is a limit to the number of specific players that can be selected from each Premiership team.
The full set of rules is here
In summary, this is a ‘game’ of luck, skill, and analytics - the same things that seem to matter in the business world.
So what did I learn about business through playing FPL?
1. You need to predict the future
In FPL you have to pick your team based on what you think will happen in the next, or next few, games. Aspects, like form, upcoming fixtures, and inherent characteristics of a player are all used by managers. Many people use their own knowledge about football, or the ‘eye-test’ of watching specific players.
Increasingly, managers are using data, models and advanced analytics to calculate ‘expected value’ (EV) or the number of points a player is expected to score in the next gameweek. Managers following this approach then try to maximise the EV of their team each week. Given my limited football knowledge, this is the analytics-based approach I try to follow.
A number of these EV predictors are available on the internet (like Fantasy Overlord or FPL review) but others (like me) try to build our own. FPL has an API with a lot of detailed data which can be used to build these models and is accessed with tools like fplscrapR.
At Predictive Insights we do a similar thing in the business space - make forecasts of what is going to happen based on large and diverse sources of data, as well as robust statistical analysis. In our experience we’ve found that this data-driven approach is more accurate than traditional approaches and can free up people to do things that they are good at, like managing people, interacting with customers, or thinking strategically about business direction. Playing FPL has helped me think about business problems in a similar way to the FPL problem - what are the important things to predict which will actually have an impact on the business.
2. It’s about optimising where you spend your resources
In FPL you have £100 million at the beginning of the season to spend on buying 15 players. Once you’ve figured out which players you expect to score the most points in the upcoming gameweek (or gameweeks) you need to get the best combination to maximise future points with the budget constraint you have. This can be a tricky problem to solve.
Innovators like Sertalp B. Çay have brought analytical optimisation techniques to FPL over the past two seasons and this approach is paying off - Sertalp is in the top 5,000 of players at the moment (in other words, the top 0.05% of players). Introductions to these approaches and other useful tools are available on his website
I’m a big fan of his live gameweek tracker which shows, in real time, how I’m expected to perform relative to different groups.
At Predictive Insights we’ve used some of these optimisation approaches to help clients better schedule staff, to decide on what to stock and where to place it, and how to prioritise orders and enquiries. Like in FPL, these analytical approaches can bring considerable monetary savings, but also, importantly, improvements in customer and staff satisfaction.
Optimising where and how I spend my time has been another aspect of running a small business with parallels to FPL. I have a limited amount of time and energy and have to think hard about how to use this where it is of the greatest benefit to the business and my colleagues.
3. Strategy matters
There are a number of aspects to FPL that make it more than ‘just’ an optimisation problem. There are special ‘chips’ which can be used at different times of the season to change your team (either permanently or temporarily), to back certain players, and to use players which are on your bench. This requires identifying the best times to use these chips and having strategies for each of them.
To complicate things further, the fixture schedule is not fixed but can be changed. In the past two seasons this has been due to Covid postponements but more generally it happens because teams have commitments as they advance in the Cup and European competitions. The best resource to keep track of future fixtures and likely blanks and double gameweeks, which will require specific strategies is Ben Crellin.
As you can see in the figure above, I’ve not figured out the optimal strategy for many of these unusual gameweeks - I’m the orange, a sample of ‘elite’ managers are grey (including chess legend Magnus Carlson), and some of the players who rely heavily on analytics are in green. Most weeks I'm lagging against these better players.
Business also requires specific strategies which are often contextspecific. This is one area which machine learning projects often underestimate. In my experience, working closely with people who know the context is key for successful implementation of machine learning and data-driven solutions. These types of projects often also require organsiational change, which needs to be done in a sensitive and empathetic way, which also requires an understanding of the organisational culture.
If you have read this far, thank you!
I’m trying to put together a mini-league of players who are interested in improving at FPL through the better use of data and analytics. If you are interested please join here. I'll then try and pull your data for this season to create a benchmark for performance next season.
Neil Rankin is a founder of Predictive Insights, a machine learning and forecasting company which combines economic data, behavioural science and analytical techniques to produce accurate and useful business forecasts.
This is his second season playing FPL, and he almost made it into the top 100k in one lucky gameweek but has since slipped in rank.