Every supporter is a forecaster, and most of us are hopeless at it. A new football site has handed the job to a model with no badge, no derby and no memory of last season’s heartbreak, and the way it watches the game is fascinating.
Every football fan makes predictions. We make them on the bus to the ground, in the group chat on a Friday night, in the pub after a bad defeat, when we decide with great confidence that the manager has lost the dressing room. We are sure our striker is due a goal. We are sure the big club will turn it around. We are sure this is finally our year.
Most of the time we are wrong, and it doesn’t matter, because that was never really the point. Supporting a club means looking at football through a very particular lens. We remember the late winner from twelve years ago and forget the twenty dull draws around it. We rate our own players a little higher than everyone else does. We believe in the new manager bounce right up until it stops bouncing.
So I was curious, during November’s international break, to spend a day in the company of a football fan who has never supported anyone. It lives on BananaGoal, a football prediction site launched this year, and it isn’t a person at all. It is a statistical model that studies matches all over the world, makes a call on every one of them before kick-off and then, once the final whistle has gone, lets anyone see whether it was right.
A fan without a badge
The first thing you notice is how evenly it pays attention. The model doesn’t know which club has the bigger stadium, the grander history or the louder away end, and it doesn’t care. A Champions League night and a wet Thursday in a second division get exactly the same routine. That sounds obvious until you remember how football coverage usually works, with a handful of clubs taking up most of the oxygen and everyone else fighting for scraps.
On the day I looked, that even-handedness was on full display. With the big European leagues paused for internationals, the board still carried 63 matches with a published prediction: Nations League games in Europe and in the CONCACAF region, Brazil’s Serie B, Uruguay’s Primera División, the Egyptian second tier and women’s football in the United States. Most of us would not watch a minute of that. The model had studied all of it.
What it watches instead of the crest
A fan reads a match through feelings and memories. The model reads it through the numbers a good analyst would want in front of them: expected goals (xG) created and conceded, recent form, home and away records, probable line-ups, injuries and availability, previous meetings between the sides, and how the wider market rates the match in the hours before kick-off, including how that view shifts. According to the site, no person picks, approves or edits any of its calls, before or after they go live.
Expected goals matter more here than they might seem. A team can create the better chances, win the xG battle comfortably and still lose 1-0 to a deflected shot from 25 yards. A fan remembers the scoreline. The model remembers the chances, and the chances usually say far more about what will happen next week.
The part that really won me over is how it treats different competitions. Each signal counts according to how well it has predicted earlier matches in that same competition. Leagues have personalities, and the model respects that. A mid-table game in a league it reads well can end up with a stronger call than a glamour tie somewhere its signals have been less reliable. Fans tend to carry the logic of their own league everywhere they look. This one adjusts its thinking to every league it watches, which is exactly what a good scout does.
There is a clever twist in how it chooses its main call for each match, too. It rates several outcomes, from the result to the number of goals to whether both teams score, and the call it puts forward is not always the most likely one. It is the outcome where its own estimate differs most from the consensus going into the game. Put simply, it isn’t trying to name the favourite. It is looking for the places where everyone else might have the match wrong, and that is a far more interesting way to watch football.
Never more than 95
Here is where the fan with no club is least like the rest of us. When we fancy our team, we say it with total conviction: we’ll win this, no problem. The model never does. Every call on BananaGoal comes with a percentage, the chance that it comes in as the model sees it, and nothing on the site is ever shown above 95%, however lopsided a fixture looks on paper. There is no 99% anywhere, and certainly no 100%.
The site is just as clear about what that number means. If a large batch of calls is marked at 75%, about three in four should land and one in four should not. Plenty of people in football will happily tell you the first half of that sentence. Very few are as honest about the second. The cap at 95 is a small design choice, but it tells you the people behind it understand the sport. Some share of every match belongs to chance: the bobble on a tired pitch, the goalkeeper having an inspired night, the red card nobody saw coming.
Calls rated 90% or higher are kept as their own group and gathered on a separate daily page. They are rarer than you might expect. On that Thursday, the model had found four across the whole board. Some days there are more, some days none at all, and the page simply stays short rather than being padded out. I liked that a lot. It is the voice of a fan who only raises it when genuinely confident.
A Thursday nobody was watching
Six of that day’s calls were open to everyone, no account needed, and they read like a tour of football’s back roads. Armenia against Cyprus came with “at least two goals” at 72%. Cerro Largo against Progreso in Uruguay had “three goals or fewer” at 77%. Trinidad and Tobago were given a 73% chance of beating Antigua and Barbuda. The first of the six had already been settled overnight: DC Power against Brooklyn in the USL Super League, at least two goals, rated 78%. It came in.
What surprised me is that these open calls are not the model’s strongest. They are drawn from calls with a middling chance, spread across different competitions, and every visitor sees the same six for the whole day. The 90-plus group is never among them. It is a genuinely fair way to give people a taste, and a confident one. The site is happy to be judged on its everyday calls, not only on its best.
A fan who admits when it is wrong
Football has a long tradition of predictions that nobody revisits. Every August, pundits line up to call the title race and the relegation places, and by May the clips have quietly gone missing. Fans are no better. Nobody keeps a list of the times they were certain about a derby and lost it.
BananaGoal keeps the list. Once a match ends, its call becomes visible to everyone, marked right or wrong, and it is never changed afterwards. The record page shows daily results, monthly totals, a heatmap of good and bad days and every graded day, the bad ones included, and anyone can read it without an account. Every finished match on the daily board is open as well. The site’s about page sums up the thinking better than I could: “A site that shows only its winning picks is not offering a track record. It is offering an advert.”
For a newcomer, that is a bold way to introduce yourself. It means asking to be judged on everything you do, rather than on a carefully edited highlights reel. I can’t think of a more convincing way to earn trust in this corner of football.
From the Champions League to Egypt’s second tier
The scope is staggering. The competition directory lists 580 competitions from 122 countries, starting with the Premier League, La Liga, Serie A, the Bundesliga, Ligue 1 and the three European club competitions, then running down through domestic cups, lower divisions, women’s leagues and international qualifiers. Each competition has its own page with upcoming calls, graded results, the current table and a running profile of the season: how often the home side wins, how many games end level, how many goals a match produces and how often both teams score. Clubs and national teams have pages too, with recent form and a tally of how the model has fared on them. For anyone who loves the numbers side of the game, it is a wonderful rabbit hole.
The site runs in six languages (English, Spanish, French, Italian, Romanian and Russian) and lives in the browser, so it works on a phone with nothing to install. It is clean, quick to read and easy to find your way around, which matters when there are dozens of matches on the board.
Who is behind it
BananaGoal is run by two registered companies: NerdyTech HUB SRL in Romania, which looks after data protection, and KickOff Ventures LLC in Delaware, which handles international operations. That kind of openness about who runs a site is rarer than it should be. It describes itself as an analysis service. It doesn’t take bets or hold anyone’s money, and, according to the site, no betting company owns a share of it or pays it for traffic. A handful of calls are open every day, a paid membership unlocks the rest of the upcoming board, and finished matches stay public for everyone. The site is meant for adults, and every page carries the same sensible reminder: a prediction is an estimate, not a promise.
Heart and head
I won’t stop supporting my club with my heart, and I wouldn’t want to. That is the whole joy of football. I will still be sure we are about to turn the corner, and I will still be wrong about it more often than I would like.
But I came away from BananaGoal with real respect for the fan who never supported anyone. It watches the Uruguayan league with the same care as the Champions League, it never claims more certainty than the game allows, and it leaves its mistakes on display next to its successes. That is more than most of us manage. For a site that only launched this year, it already behaves like one that expects to be around for a very long time.
So the next time the group chat is certain about a derby, it might be worth asking the fan with no club what it thinks.
