Struggle with AFL footy tips? Even computer algorithms cannot beat randomness

The home and away rounds of the AFL season have ended, and clever pundits are celebrating their wins in the tens of thousands of office tipping comps run around the country.

How did you go? How well do you know your footy?

Because even if you think you are clued in to Australian rules football, tipping winners is hard, and the data proves it.

ABC Sport is live blogging every round of the AFL and NRL seasons in 2025.

The winning tippers in two of the most prominent online AFL tipping competitions, with a combined 1.3 million participants, correctly picked 165 games out of 207 matches.

That's an average of 79.71 per cent correct tips, or to put it another way, the best of the best tippers, among a collective group of more than 1.3 million people, got slightly more than one out of five games wrong.

The average for the largest of these competitions, with around 900,000 participants, is 132.6 correct tips, which is an average of 64 per cent or less than two out of three games right.

But being able to tip as well as those at the top of leaderboards is extremely difficult, and there’s a strong argument that even those tippers are outliers who got lucky.

The best computer algorithms can't do quite as well, but they do perform consistently. So why is it so hard to tip correctly?

In a nine-match round, the odds of randomly picking every match correctly are one in 512, not accounting for draws.

The chances of randomly tipping every game correctly in a 207-game home-and-away season would far exceed the number of stars in the universe.

There are simply so many variables in a game of football, and deciding which ones affect the outcome of the match is fiendishly difficult.

Consider too, that this season there have been 19 games decided by five points or less, such that one kick could have changed the result.

Even if we take into account this season has been a relatively easier because tipping against the bottom three teams has been close to a sure thing, we can't get away from the fact that every week there are enough evenly matched games, which makes them toss ups.

Max Barry is a novelist who, in his spare time, loves to dabble with data and runs an AFL stats website called Squiggle.

Squiggle is home to Barry's own computer model for tipping AFL games and another 27 models designed by data analysts.

Barry's model correctly tipped 152 games.

The best two models on the Squiggle site correctly tipped 158 games, which is a strike rate of 76.3 per cent, or slightly better than three out of four.

Those two models would have placed in the top 0.24 per cent biggest tipping websites — they did extremely well, without any human involvement.

One of them, called Matter of Stats, was programmed by data analyst Tony Corke, who said getting 76 of the tips right was "exceptional".

Corke said the difference between the best human tippers and his computer model was the ability to get close games right.

"At the end of the day, the person that wins the tipping competition is usually the person who happened to do best in the coin-toss games," he said.

"If you ask 1,000 people to tip eight coin tosses in a row, one of them will tip all eight," Corke said.

So, how are Corke and Barry doing it?

Barry says it was a matter of compiling the variables in any game of football and creating an algorithm that gives those variables a weighting.

The most obvious one is home ground advantage.

Over the 150-odd years of the V/AFL, the home team has won about 57.4 per cent of all games.

This season it was 57.5 per cent or 115 home wins out of 200 (excluding the seven games played at neutral venues during the Gather Round in South Australia).

But that number was skewed by West Coast, which only won one game at home, North Melbourne, which won three, and Essendon and Richmond, which both won four.

In other words, tipping the home team and always tipping against the bottom four teams would have given you a relatively healthy strike rate of 64 per cent.

Barry said one of the unusual factors in Australian rules football was how much information you can get from the raw scores.

"So, just from the fact that teams tend to win 88-62, and scores like that, means that you get a lot more information than, say, a soccer match where someone might win 3-1," Barry said.

And breaking that down, the percentage and the number of scoring shots also contain a lot of important data.

"Percentage contains more information about whether teams are squeaking out wins or if they're getting wins by big margins," Barry said.

"And the second thing would be, look at scoring shots rather than scores."

If a team kicks seven goals and 19 behinds for a score of 61, they would have had 15 more scoring shots than a team that racked up the same score by kicking 10 goals, one behind.

"The point is that a team has to do a lot right to get a goal, and when they do everything right — except the very last step of converting that chance into a goal rather than a behind — it is a really strong signal that they're actually a pretty decent team," Barry said.

When Barry started his computer tipping model, the only two variables were home ground advantage and raw scores.

He then started adding more variables to test theories and now has over a hundred models.

"Someone would come up with a theory about why a team won this game and then I would say, 'OK, well, let's make a model that assumes that's true, and see if it actually exists or not,'" Barry said.

"There's a lot of myth about what creates winning teams and what creates wins, and none of it can be quantified.

"So, I would run these models, and a lot of them just wouldn't do any better than my existing simple models, so I would just ignore it.

"But occasionally a model would actually do something useful."

Corke's model also takes into account where games are played, but he also said he allows for what he calls away-ground disadvantage.

"Some teams play a lot worse at Docklands when they're away, but another team not so much," he said.

Corke and Barry also take into account what players have been selected using Champion Data's player ratings.

But the difference missing a star player might make — even someone like Collingwood's Brownlow medal favourite, Nick Daicos — is actually not that great.

"I think it gives you an edge, but it's not as much as you think," Barry said.

"That's not because the players themselves don't matter, but because it's quite difficult to quantify how [much] a player matters, even very good players, if you remove them from their side.

"So, it can be the case that missing Daicos is absolutely terrible for Collingwood, but in the general case, missing a superstar player doesn't affect the results all that much," Barry said.

Corke said the difference in terms of points lost by not having a star available was less than a goal at best.

"If they're an absolute gun, maybe three of four points, maybe six at a stretch, but it's nothing more than that," he said.

The key question is what weighting do you give to each variable? How do you weigh up home-ground advantage against Nick Daicos not playing, against a rainy day in front of a smaller-than-usual crowd?

"The weighting is really important and the best way to do it is to come up with a plausible range for what you think that weighting might be," Barry said.

"Then you have access to a massive set of historical data, and you run all the different variations of that weighting against the historical data, and you see which ones seem to do the best.

"The tricky thing is that at some point in the past, the historical data is probably not so relevant to modern football.

"So, you have to make a decision as to what trade-off you choose between having a lot of data, which is going to make it more reliable and having irrelevant data from football from an era that is very different to today."

The Squiggle site also contains Mean Average Error (MAE), which is the average difference between the predicted margin and the actual margin of each game.

It turns out that even the best model on the Squiggle site is out by 25.59 points.

"It means that no matter how good you are, you still can't get within four goals of the average margin, which is a fair way to be out," Barry said.

This explains why, if you get a big enough group or people tipping, some of the outliers will do better than the computers, which history suggests always happens.

"And that is why these models, which are very sophisticated and built by very smart people — much smarter than me — regularly get beaten by someone's mum tipping their teams with the mascot she likes," Barry said.

"It's that old thing about a million monkeys in a million years will tip the perfect season of AFL football," Barry said.

"So, you get a big enough group of people, and even if they're all tipping randomly, there's gonna be some people who are gonna do extremely well and they're going to beat the model.

"The models will generally murder any small group of tipsters; it's only when you get enough monkeys in a group that you start to get the outliers who start getting better."

The reason that it is only the outliers who beat the computers may be explained by the fact that humans have an inordinate amount of difficulty coming to terms with probability and randomness, compared to statisticians who program computers to wade through numbers to find relevant patterns and ignore irrelevant ones.

"People will get this idea that Geelong's due," Corke said.

"They've lost the last three. They must be due to win the next one.

"There's no particular reason why they should, but nonetheless, we humans have that feeling about these kind of trends can't persist.

We see what we want to see and make stories to fit patterns that may or may not be there.

But according to Barry, that is the beauty of football, and it's where his dual lives as a novelist and the data-driven world of statistics find synchronicity.

"I don't think that combination of numbers and narrative really exists anywhere else quite like it does in sport," he said.

"The numbers can reveal the nature of the story for me.

"So, it means you're not just taking as gospel what you're being told and why something happened, but you get a bit of a deeper insight into the real forces that are driving what you're seeing on the field.

What then of us mere mortals, who weekly take a stab in the dark, resisting the urge to tip against our own team and letting the instincts of our gut tell us who's going to win a game of football?

"I think that's part of the magic of tipping, that we all go in with very different theories and the way we tip to a degree reflects the kind of outlook we have on life and what we believe about how predictable the world is," Barry said.

"We don't want to all just leave it to the computers to predict who's going to win."