Every week, millions of adults pretend that they’re football managers. I am one of them. We live out our dreams in the Fantasy Premier League (FPL), an online game that’s become a global phenomenon.
With a budget of 100 million (digital) pounds, we build virtual squads of footballers. If their real-life counterparts play well, we win points. By the end of the season, the team with the most points wins. But what exactly do they win? Well, that depends who you ask.
Here’s the official answer: over 10 million players compete for the grand prize of — drum roll, please — two tickets to a football match. Yet most of us fight for a far greater honour: bragging rights.
Win your mini league of friends and bask in the glory of your exalted social status. Finish last and face eternal shame for your pathetic football knowledge.
The stakes are high. So high, in fact, that some players now employ controversial advisors: data analytics.
These means I can no longer trust my rivals. Are they even making their decisions? Who’s the real winner when the teams are guided by AI? What happens to a game that’s managed by machines?
Regrettably, I now must ask these questions of myself.
A new era for fantasy football?
After several seasons of mixed results in FPL, I began searching for analytical support.
Thankfully, I soon found my saviours. There was Mohamed, aka Ragabolly, an Egyptian based in the US, whose performance simulators are the stuff of legend. There was Ben Crellin, a Brit who builds spreadsheets to navigate the season. There were even entire companies dedicated to FPL analytics.
The biggest of them all is Fantasy Football Hub, a startup based in the UK. After launching with a blog focus in 2019, Fantasy Football Hub began experimenting with point projections.
“It quickly became quite apparent that this was hugely popular,” says Will Thomas, the company’s CEO and founder. “We doubled down on it.”
Using a mind for stats developed during psychology research, Thomas built a longitudinal multilevel regression model for the platform.
The model scours data from football analytics kingpin Opta. It then predicts every possible permutation of points: goals, assists, clean sheets, minutes played, you name it. All the information is pooled into predicted points and transfer recommendations.
The tips have proven popular. About 40,000 paying subscribers and 200,000 registered users are now members of the Hub. Annual recurring revenues have surpassed £2.5mn. Micah Richards, a TV pundit who won a real Premier League title with Manchester City, has joined as a partner.
But that’s not what impresses me. The Hub earns my respect by making a noble offer: win your FPL mini-league or get your money back.
A deal like that is too hard to ignore.
Following the model
I signed up to the service and explore the analytics. I copied the model’s top team to start the season. Discussions with my friends were no longer needed.
But then the plan began give me pause. Something just didn’t feel right about AI’s attempts to beat my friends. Suddenly, it hit me: the tool has endorsed players from the loathsome Tottenham Hotspur, the arch enemies of my beloved Arsenal FC.
Then I recall a tip from Thomas: don’t just blindly follow the model. Instead, use the analysis as “a sense check.”
With my good sense reactivated, I replace the wretched beasts from Spurs. AI will no longer be my boss, but merely my employee.
Still, the model lacks a certain human touch. I miss speaking to my pals about FPL — until I find a superior alternative: an AI that acts like a friend.
A little more conversation
As the FPL summer break drew to a close, a new model caught my eye. Built by Fantasy Football Fix, the tool follows the conversational trend set by ChatGPT.
Fix bills the app as “the world’s first AI chat assistant dedicated to FPL.” The company has named the bot — can you guess? — ChatFPL.
Ask a question and ChatFPL will answer with substantiating evidence. Request information and the bot will reply with digestible data.
The depth of detail is impressive. I compare multiple players across their expected goals, minutes, and future prices. My transfers are optimised for budgets and opponents on specific gameweeks.
Unlike my pals, the bot provides exactly what I want. It’s also a fairly fluid conversationalist. With each response, ChatFPL invites follow-up questions. Our dialogue then moves in new directions.
Tom Brown, Fix’s head of marketing, compares the app to an “informed friend” with whom you “bounce ideas.”
Like his competitors at Hub, Brown defies claims that the AI replaces human decision-making. “We give recommendations — it shouldn’t be just an answer,” he says.
The interaction does enrich my experience. It gives ChatFPL a semblance of humanity.
Under the hood, however, the system is fully computational.
Working as a team
ChatFPL’s artificial brain is an evolving roster of LLMs. The Pro version currently runs on Anthrophic’s Claude 3.5 Sonnet, while the Lite model uses OpenAI’s GPT-4o-mini. When new LLMs emerge, the team QA tests them and then considers integration.
“We’re hooked into all of the main models,” says Adam Moss, Fix’s technical director. “But the key part is that the model acts as an interpreter to the user’s question. It then determines what additional information needs to be injected into the prompt.”
The system also continuously updates the data through various APIs. In my experience, however, ChatFPL still sometimes needs a push into the present day.
At one point, the bot suggests buying Dominic Solanke from Bournemouth. A solid choice, but there’s a big problem: Bournemouth had just sold the striker to the odious Tottenham Hotspur.
After an indignant reminder, the bot swiftly delivers updated suggestions. Sensibly, they arrive with a profuse apology. I forgive the transgression to focus on what really matters: winning my mini league.
Building an FPL champion
My strategy is decided. I shall blend the analytics with my own ideas.
I scour the AI-generated teams projected to maximise my points. I ask ChatFPL’s advice on my own amendments. I feed the updated squad to the Hub’s AI transfer recommender. I pick my final squad.
I have become an FPL cyborg: a nerdy football fan merged with analytics. I wait with baited breath for the results.
The fusion decision soon pays off. After the first round of fixtures, I’m top of my mini-league. But the decisions are becoming harder.
As the new season progresses, the challenges grow. No longer do I merely have to buy 11 players for my squad. I now also have to contend with injuries, form fluctuations, and a tight budget.
For Brown, that’s when the app truly proves its worth. “You can go on a group chat with your mates and ask them but they’re in different places, so they can’t really give you advice,” he says. “ChatFPL will be an informed source to talk to about your situation.”
It’s a fair point: my friends are now my rivals. They can no longer be trusted. Even if they could, their advice would be questionable at best.
I try to instead build a relationship with my tools. I seek their advice, ask them questions, listen to their views. The final decisions, however, are mine alone. My bots are closer to assistant coaches than dictatorial managers.
Fix endorses my approach. “AI does seem to be sucking the fun out of FPL for certain people,” says Brown. “Before, it was open for interpretation — there wasn’t a right and wrong way to play. We want to go back to that.”
At times, the models can even expand the constraints of our consciences. Paradoxically, the AI will often think outside the box.
The human element
FPL players increasingly follow the same “templates” shared online by dubious experts. They drain our individuality and creativity.
AI provides an escape from their restraints. Groupthink, confirmation biases, contrast effects, and all our other prejudices are cast aside in the quest for points.
There’s a parallel here to a broader divide in AI. On one side are the algorithmic biases that misidentify people, reinforce stereotypes, and amplify discrimination. On the other side are the systems that expand our horizons.
A shining example of the latter is DeepMind’s AlphaGo. The model developed radical strategies for a fiendishly complex board game, stunning world champions with unorthodox moves. AI can do the same for FPL.
“It brings a different element to it, which takes our inherent psychological biases out of the equation,” says Thomas of Fantasy Football Hub.
One memorable example for Thomas was vouching for Jean-Philippe Mateta.
At the time, the Crystal Palace striker was an unfashionable choice. Neither he nor his team were in form. But his underlying data was promising. He would also soon benefit from two top creative players rejoining his team. Most importantly, he cost a fraction of the league’s established strikers.
The model astutely identified a bargain. Mateta promptly embarked on a lengthy scoring streak that propelled Palace up the table.
AI vs the eye test
The Mateta pick was reminiscent of “Moneyball,” an analytics approach that’s legendary in real sports. Struggling sides have deployed the techniques to unearth diamonds in the rough. Human experts will often overlook these players. Yet data shows they’re undervalued, abnormally skilled, or just need a helping hand.
Such oversights are common in fantasy football. “A lot of FPL players use the eye test,” Thomas says. “But the AI doesn’t care about that.”
Over time, the models will also add new personal touches. Fix plans to customise advice around our strategic styles, risk appetites, and football preferences. Before long, AI could tailor the tips for winning my mini-league. I could ban the mere mention of Spurs.
I could also build a richer relationship with my AI partner. “We’re exploring being able to talk to it verbally,” says Brown. A comfort, perhaps, for the lonely. But one that won’t dispel the fears about AI replacing human roles in FPL. Not just strategically, but also socially.
Still, that’s of little concern to me. I’m leading my mini-league and no-one knows about my AI assistant. Please don’t tell them my secret — I have a reputation to automate.
Leave A Comment
You must be logged in to post a comment.