When the Algorithm Says Your Team Should Lose

Sports fans used to have a relatively simple vocabulary before a game. One team was flagged as better, the other was flagged as worse. There was a favorite and an underdog, but those labels mostly lived in previews, conversations between fans and the betting lines printed in newspapers.

Today, probability is everywhere – television broadcasts display win percentages while games are still being played, sports apps update projections as scores change, advanced statistics attempt to quantify everything from player performance to championship chances, while betting markets constantly attach numbers to what might happen next.

The modern fan is surrounded by predictions everywhere, and yet sport retains an inconvenient habit of ignoring them. Here comes a decade of data analyzed by Sports Illustrated, which provides a striking reminder for all sports fans – SI examined teams across the NFL, NBA, MLB, college football and college basketball to identify those that won most frequently when the betting market expected them to lose.

The most successful underdog wasn’t an obscure team responsible for a handful of famous shocks. It was the Kansas City Chiefs.

Kansas City won 20 of the 30 games in which it was listed as an underdog during the period analyzed, producing a 66.67% win rate. Baltimore followed at 61.9%, while Pittsburgh won 60% of its games as an underdog.

In other words, one of the defining NFL teams of its era repeatedly entered games with the market saying the other side was more likely to win, then won anyway.

That tells us something important about the increasingly quantified way we watch sport. A probability is useful, but it is not the same thing as knowing what will happen.

Being an underdog doesn’t mean being a bad team

Part of the confusion comes from the language itself.

Calling a team an “underdog” can sound like a judgement on its overall quality. In reality, the label applies to a specific game under a specific set of circumstances.

An excellent team can be an underdog against another excellent team. Venue matters. Injuries matter. Rest matters. Recent performances matter. In some sports, even weather can materially change expectations.

Kansas City’s presence at the top of the Sports Illustrated ranking illustrates the distinction particularly well. The Chiefs weren’t spending a decade as one of the NFL’s weakest teams and somehow producing miracle after miracle. They were regularly among the league’s strongest teams.

There were simply individual games in which the balance of available information suggested their opponent had the better chance of winning.

The same pattern appears elsewhere in the study.

Boston had the highest underdog win rate among NBA teams at 43%, winning 92 of 214 games in which it was the outsider. In college basketball, Duke won exactly half of its 54 games as an underdog. Clemson led college football at 57.1%.

These are hardly sporting minnows.

Sometimes the algorithm, model or market isn’t telling us that a team is poor. It is saying that, under today’s circumstances, another team has a slightly better chance.

That distinction often disappears when a percentage appears on a screen.

Fans increasingly experience sport through probability

This matters because probabilities are no longer hidden in the background.

They have become part of sports entertainment itself.

A team goes down by ten points and a live win probability suddenly falls. A touchdown, three-pointer or home run sends it back in the opposite direction. Fans can watch the assessment of a game change almost as quickly as the score.

The growth of in-play and micro-betting has pushed this even further. Individual moments can now become markets of their own, with probabilities being recalculated throughout an event.

Big News Network recently examined this second-screen phenomenon and how micro-betting is changing the way people watch sport. Instead of following only the final result, viewers can now interact with a constant stream of short-term outcomes, statistics and changing prices.

That creates a very different relationship with uncertainty.

The fan isn’t simply watching to discover who wins. Increasingly, they are being told throughout the game who is supposed to win.

More information doesn’t eliminate uncertainty

There is an obvious temptation to assume that better data should eventually make sport predictable.

Teams now collect extraordinary amounts of information. Player movement can be tracked. Workloads can be monitored. Analysts can evaluate thousands of possessions, pitches or plays. Models can process variables far faster than a human observer.

The betting market has access to vastly more information than it did decades ago as well.

Yet the fundamental problem remains.

Sports aren’t repeated laboratory experiments.

A model can estimate the likelihood of a basketball player making a shot, but it doesn’t know with certainty whether the next attempt will fall. It can evaluate how often an NFL team converts a particular situation, but one missed tackle can invalidate the expectation immediately.

Even a 70% probability leaves a 30% probability that something else happens.

Over enough games, those percentages can be extremely informative. For the fan watching one particular game on Sunday, the unlikely outcome remains entirely possible.

This is where probabilities can become psychologically misleading.

People tend to translate “more likely” into “will happen.”

Sport doesn’t make that promise.

The favorite can be correctly priced and still lose

An upset also doesn’t necessarily prove that the original prediction was wrong.

Suppose a team has a genuine 70% chance of winning. If it loses, that doesn’t retrospectively mean the assessment was poor. Losing was always part of the remaining 30%.

If the same situation could somehow be replayed hundreds of times, the stronger team might win approximately as often as expected.

Fans only experience the one version that actually happened.

This is one reason upset results are so memorable. The probability disappears as soon as the final whistle sounds. There is no longer a 70% favorite and 30% underdog. There is simply a winner and a loser.

Sports Illustrated’s findings need to be understood in the same way. Kansas City winning two-thirds of its games as an underdog doesn’t prove that every individual pregame assessment was mistaken.

What it shows is that even the side considered less likely to win can possess a very substantial chance of doing so.

Different sports create different kinds of uncertainty

The SI data becomes even more interesting when the sports are compared.

Kansas City’s 66.67% underdog win rate was far above the NBA-leading Celtics at 43%. Houston led MLB teams at 51.3%, while Clemson topped college football at 57.1%.

Those differences shouldn’t automatically be interpreted as evidence that one sport is easier to predict than another. The number of games, frequency with which teams became underdogs and characteristics of each competition differ substantially.

But the comparison highlights how uncertainty takes different forms.

Basketball contains dozens of scoring possessions, but a sudden run of three-point shooting can transform a game. Baseball can turn on one pitching performance or one swing. Football has fewer scoring opportunities, making individual turnovers and explosive plays particularly influential.

Models can account for those characteristics when calculating probabilities.

They cannot remove them from the sport.

AI adds another layer to sports prediction

Artificial intelligence is making the relationship between fans, data and prediction even more complicated.

AI tools can process large datasets, identify patterns and summarize information that would once have required hours of manual research. For sports fans, that makes sophisticated analysis far more accessible.

But accessibility can create an illusion of certainty.

An AI-generated prediction may sound authoritative because it can explain itself convincingly. It can cite recent form, injuries, historical matchups and statistical trends. What it ultimately produces, however, is still an assessment of an uncertain future event.

There is an important difference between identifying that one outcome is more probable and knowing that it will happen.

The better sports analytics become, the easier it can be to forget that distinction.

Canada shows how quickly the prediction economy is changing

The growth of prediction-driven products is now creating questions well beyond the playing field.

Canada provides a particularly interesting example. Sports betting regulation differs by province, while regulators have also been considering how the newer generation of prediction markets should be treated.

In August 2026, Canadian securities regulators drew a clear distinction between conventional financial prediction products and contracts based on sports and entertainment outcomes. Their position was that sports event contracts should not simply enter Canada through the same regulatory framework used for securities and derivatives.

The issue demonstrates how quickly the concept of “predicting an outcome” has expanded beyond traditional sports analysis. Probabilities can now appear in broadcasts, betting markets, financial-style prediction products and AI-generated forecasts, even though those products may be governed very differently depending on where the user lives.

The contrast with the United States is particularly noticeable. Sports betting there is regulated largely at state level, meaning availability can change simply by crossing a state border. Sports Illustrated tracks that fragmented US market alongside its sports coverage, maintaining a regularly updated list of betting websites that evaluates legal operators on areas including odds, market coverage, payments and platform features.

Canada and the US may watch many of the same leagues, teams and athletes, but the systems surrounding prediction and wagering don’t necessarily cross the border with them.

Live probabilities can be even more deceptive

Pregame predictions at least have a clear starting point. Once the game begins, probabilities can change dramatically.

Consider an NFL favorite that concedes two early touchdowns.

Its pregame advantage hasn’t necessarily disappeared. There may still be three quarters remaining, and the underlying difference in team quality still matters. But the game state has changed, so any realistic model has to react.

The team that began as the favorite can quickly become the live underdog.

If it scores twice, the calculation may reverse again.

That constant movement can make probability feel like a running verdict on what is happening. In reality, it is closer to a continuously updated estimate.

This distinction becomes especially important when a probability moves toward an extreme. Seeing a team given only a small chance of winning can make the comeback feel almost impossible.

Almost isn’t the same as impossible.

Sports history is filled with games that eventually became famous precisely because something happened after most observers had stopped expecting it.

Algorithms are better at uncertainty than humans are

There is an irony in all of this.

The algorithm isn’t necessarily the problem.

A good model is comfortable saying that something has a 35% chance of happening. Humans are often the ones who convert that into “it’s not going to happen.”

Probabilities acknowledge uncertainty by definition.

A 35% underdog should win often enough that nobody ought to be shocked when it does. Play that situation 100 times and, if the estimate is accurate, the supposedly unlikely outcome should occur roughly 35 times.

But fans don’t experience 100 identical games.

They experience tonight.

That’s why the Sports Illustrated study is so revealing. Looking across a decade strips away some of the emotion attached to individual upsets and shows how regularly teams overcome unfavorable expectations.

Kansas City did it 20 times. Boston did it 92 times. Miami did it 156 times.

Each individual result may have had its own explanation. Together, they demonstrate just how dangerous it is to confuse probability with certainty.

The prediction is part of the story, not the ending

Modern sports fans aren’t going back to an era without data.

Nor should they.

Advanced statistics can reveal things that the eye misses. Models can challenge lazy assumptions. Live probabilities can help explain just how dramatic a comeback really is. Better information can make understanding a sport more interesting.

The mistake comes when the prediction begins to feel more authoritative than the event itself.

Kansas City could be the underdog and still win. Boston could be expected to lose and still walk off the court victorious. Duke could enter as the outsider and beat the favorite.

The algorithm isn’t necessarily saying those things cannot happen.

It’s saying they are less likely to happen.

Sport lives in the difference between those two statements.

And every so often, your team gets to spend the evening proving just how large that difference can be.