How many ways are there to play football?

Introducing team tendencies and team styles

There’s a sort of unofficial checklist for a manager’s first press conference. They’ll praise the supporters, thank the board, crack a tepid joke or two and get a little misty reflecting on the proud traditions of such a massive club (not an easy line to sell when you’re taking the helm of the Dodge City Prairie Dogs, but such is the price of leadership).

For a grand finale, the new gaffer will take a swig of Evian, lean into the microphone and announce a bold new tactical vision: from now on, the team will play attractive, attacking football.

Always attractive, always attacking. The phrase recurs with such alarming regularity that you start to wonder whether there are any other adjectives in the managerial repertoire.

“Coach, how’s your wife?”

“Attractive, thanks for asking.”

“And your children, are they—”

“Quite attacking, yes.”

Luckily for all of us, there’s more than one way to play the game. Some styles of play are downright ugly. Some are even rumored to involve defending.

This is a fun area of football analytics, because there are almost as many ways to try to measure styles of play with data as there are tactical approaches. At The Athletic, I once mapped out Five Kingdoms of Football from season-level stats, but for futi we wanted to get more granular than that — not just analyzing style down to the match level, but zooming all the way in on how teams behave in specific tactical situations.

The result is two new futi models: team tendencies, which break down stylistic traits by phase of play, and team styles, which sort a team’s tendencies in each match into a general category.

Modeling something as nebulous as style is neither easy nor entirely objective. It requires a lot of judgment calls based on what statisticians call “domain expertise” — in other words, ball knowledge. Everything you’re about to read about was run several different ways, poked, prodded, argued over and run again until it told us something that more or less made sense, but that doesn’t mean it’s not still open to debate.

Futi will improve future versions of these models based on your feedback, so today we’re releasing lots of new data in our interactive tables and explaining how they work.

Team tendencies

The first step in the journey toward measuring styles of play was to come up with a list of individual stats that seemed like they might tell us something useful about a team’s tactical approach, such as their pass completion percentage.

A problem with numbers like that is they can blur important distinctions. Some teams like to play short passes in the buildup to bait the opposing team’s press and then break forward quickly with direct, low-percentage passes. That’s a pretty distinctive style of play. Average it all together, though, and the team that swings between stylistic extremes in different parts of the game might come out with the same pass completion percentage as a middle-of-the-road team with no real style at all.

So step two was to add tactical context to our stats by separating them by phase of play. That meant that instead of averaging a team’s pass completion rate over the whole match, we could let the models evaluate how the team in our example plays deep in the buildup (slow, safe) independently from their approach to midfield progression (YOLO, rocket emoji).

Our pockets bulging with phase-level data, next we turned to a technique called principal component analysis (PCA) to help sort it all out. Maybe you’ve read about this in How to Win the Premier League, Ian Graham’s book about how Liverpool’s data department analyzes the game. It’s an algorithm that can take in a lot of stats, figure out which ones are related to each other, then group them together and weight them to spit out “components” that summarize the biggest differences in those statistics across teams or players.

For each tactical tendency, we ran a PCA that reduced all of the relevant stats into a couple of components. Instead of answering “Does this team press high or not?” with one basic ratio or a confusing tangle of competing numbers, we let the algorithm boil each batch of weighted stats down into a sophisticated but easy-to-grasp spectrum.

Here’s futi’s current list of team tendencies:

Patient Buildup - Direct Buildup: Intricate possession or faster, more aggressive play in the deepest phase of possession?

Wide Buildup - Central Buildup: Build up the flanks or through the middle of the pitch in the first phase of possession?

Patient Progression - Direct Progression: Intricate possession or faster, more aggressive play in the middle phase of possession?

Central Attack - Wide Attack: Attack through the middle of the pitch or from the wings?

Secure Transitions - Direct Transitions: Play it safe after winning the ball back or push the tempo in transition?

Counterpress - Consolidate: Press aggressively to win back lost balls or fall back to recover defensive structure?

Defensive Block - High Press: Stay compact to deny space or press to win the ball during the opponent’s buildup?

Control - Chaos: More organized possession or transitions and contested phases?

You can explore this data for the 2025 MLS season in the Team Tendencies tab of futi’s new interactive tables.

A snapshot of the Team Tendencies tab at futi-data.streamlit.app

If you view the team tendencies tab with the “Values / Percentiles” toggle set to “Percentiles,” the numbers in the table are pretty self-explanatory: the closer a team is to 100, the more its play leans toward the end of the spectrum that’s named in the column header. So a percentile of 90 in the column labeled “Patient Buildup” means the team builds slower than most of the league, while a percentile of 10 means they play more direct in that phase. These percentiles are scaled against other teams in the same season.

That’s not true on the “Values” side of the toggle, where the numbers are scaled at a match level across multiple seasons. Even though the numerical limits here run from -50 to 50, teams don’t go all the way to the ends of the scale because the season averages shown here aren’t as extreme as the match-level values that set the scale.

Because team tendency values are scaled across multiple seasons, league playstyle trends can shift the baseline over the years. For example, only one MLS team in 2025 was above the historical average for “High Press” because the entire league’s playing style has changed significantly in recent years. To help you get a feel for historical changes — and to keep MLS sickos occupied for days — we put three seasons of tendency data in the interactive tables.

Team Styles

Tendencies let you dig into the details of a team’s tactics, but futi also needed an at-a-glance label for a team’s playing style in each match. We wanted a way to describe football that was short, simple, attractive and — dare we say — attacking.

That’s where team styles come in.

Sort of like how the tendency model takes a bunch of stats and turns them to a few summary metrics, team styles start with match-level tendency data and reduce it further to a few distinctive tactical labels. This time we used a dimension reduction algorithm called UMAP that plots teams with similar tendencies near each other, followed by a clustering algorithm to group nearby teams together.

We found that by far the biggest distinction, at least in MLS, was between teams that are patient in possession and those that play direct (unsurprisingly, the “Patient Buildup” tendency is usually correlated with “Patient Progression”). That gave us the first aspect of team styles. Because we wanted team styles to tell you something about defending, too, we derived the second aspect from the “Counterpress” and “High Press” tendencies, which also tend to be correlated.

That gave us a nice clean grid of four styles in two dimensions. One axis measures whether teams press aggressively or fall back in defense, while the other measures whether they build in a patient, controlled way or prefer to cause chaos with long balls and direct attacks.

Here are futi’s four team styles:

Press and Possess is the classic Guardiola ideal, bringing the ball forward with controlled possession and keeping it in the opponent’s half with a tight pressing net.

Control and Regroup sides are patient in possession but less aggressive off the ball, preferring to fall back into a compact defensive shape after losing possession.

Launch and Squish teams play for field position, thumping the ball forward quickly and charging after it to suffocate the opponent with aggressive pressing.

Bunker and Counter is pure reactive football, the stuff that’s kept underdogs scrappy since time immemorial. These teams fall back instead of pressing and try to hit the space behind the opponent with quick attacks.

Futi measures these styles at the match level, since the same team may play Press and Possess at home one week, then switch to Bunker and Counter on the road against a stronger opponent. In the Team Styles tab of the interactive tables, you’ll see each team’s most common style across the whole 2025 season as well as the percentage of regular season games in which the team played each style.

A snapshot of the Team Styles tab at futi-data.streamlit.app

One thing you’ll probably notice in the data is that Control and Regroup was by far the most common style in MLS last season — in fact, it was the predominant style for over half the league. This is mostly because of the historical trends you can trace in the Team Tendencies tab, since MLS teams are more patient with the ball but less aggressive against it than they used to be. We’re still weighing the pros and cons of scaling team styles within each league and season for a more diverse mix, but in this first version of the model we’ve opted not to so that users can compare team styles across seasons and competitions.

As you explore the interactive tables, we hope you’ll hop in the futi Discord to share your thoughts. Let us know what makes sense to you and what doesn’t feel right — we want user feedback to help guide future versions of futi’s style models.

futi in the wild

Shout out to Jake Burgess, who used last week’s phase of play data release to make the very first fan analysis post from futi’s advanced stats:

And to Charles Wilson, who sniffed out the team styles data before we’d even announced it and produced this comparison of MLS’s Bunker and Counter teams’ expected goal difference:

If you use futi data in your work, let us know so we can share it with people. And if you’d like to talk to futi’s creators for an article or podcast, give us a shout. We’ve got a new podcast episode dropping early next week with American Soccer Analysis: