Frenkie de Jong, Ajax, and Artificial Transitions: Reverse Engineering Insights on Turnovers
A few months ago I wrote a piece on how high risk passing and an organized press can be combined to turn a side’s own turnovers into scoring opportunities. Therein, I argued that Liverpool’s crossing and diagonal heavy system is one that creates chances not just through successful passes, but also through unsuccessful ones. Essentially, these high difficulty passes into danger areas were attempted not just because of what they could create if they fell to one of Liverpool’s men, but also for what they could create if they fell to an opponent. Successful passing outcomes are by no means irrelevant by this logic, though. If Liverpool could, they’d much rather just complete every pass. And this is another facet of Liverpool’s style that seems statistically optimized – the side’s most valuable passes by expected pass value happen to be the ones we typically associate with their style of play:
From LFC home games last season:— Omar (@topimpacat) January 4, 2021
- highest attempted value
- highest expected completion
- lowest expected completion
- highest expected value pic.twitter.com/BsVWRTfe63
The graphic above displays a metric called xPV, or expected pass value. Expected pass value is the combination of two other metrics: xPass, which essentially measures the likelihood of a pass being completed from one point on the pitch to another, and Possession Value (PV), which measures the likelihood of a goal being scored from a given point on the pitch (in the case of this model, in a gamestate-agnostic manner – it does not know the position of the players on the pitch or the speed of the ball, only the ball’s location). Using these together, as Omar has done here, gives us an idea of which passes will improve our chances of scoring most drastically.
What you’ll notice about these passes from Liverpool is that their best value passes are also passes that follow their archetypal buildup pattern: a midfielder circulates the ball wide, then one of their fullbacks crosses towards the penalty spot and six yard box. Forget negative outcomes for a moment – their positive outcomes appear equally optimized.
I noticed this trend just after finishing the aforementioned piece on turnovers – which was incidentally the same time the analytics side of football twitter was inundated with Possession Value graphics. At first I thought, ‘Well, that’s par for the course, Liverpool do lots of smart stuff.’ However, as more of these Possession Value graphics stared me in the face, I began to realize something – the nature of Possession Value is always going to be in favor of these aggressive, low danger area to high danger area passes. This is natural of course, and what you would expect. The largest delta values will be generated by passes from very low danger areas – the fullback spaces, for instance – to very high danger areas – the wings and the penalty box. This has another natural consequence: PV will almost always prefer high risk passing to high risk dribbling. Consider that a 60 yard run is almost impossible, whereas a 60 yard pass is difficult but executable. As a result, the average dribbling event will be of lower value than the average passing event. Some extremely high difficulty dribbles, like those from wide areas into shooting positions, are attributed high value, but almost any take on outside the opposition penalty box has minimal value in the eyes of PV. Add on that dribbling success rates are lower than pass completion percentages and you have a perfect cocktail for the Expected Possession Value model to devalue these events.
This is all without considering a variable of massive importance: remember when I said these Possession Value models were ‘gamestate agnostic?” If that means nothing to you, essentially it signifies that the probability of a goal being scored is not influenced by how many players a pass or dribble beats, nor does it take into account pass velocity, where defenders are, or which direction they’re facing. This is much like the earliest expected goal models: shot location alone was used to assess the probability of a goal being scored. These models were massively improved when things such as whether a shot was taken during a counter attack or in settled play were taken into account.
These contextual variables are important in many different ways, for passes and dribbles alike. But their impact is outsized in the context of the model when it comes to dribbles. Why? Because dribbling, especially from deep, is fundamentally about altering phase of play more than it is about gaining yards. If a player receives under pressure in their own defensive third, proceeds to beat two men, then passes to a teammate on the wing, the crowd doesn’t gasp because the ball has moved ten yards up the pitch – they gasp because the attacking team has created disorder in the opposition defence. Where this comes full circle is how it relates to turnovers and transitions. In the previously discussed piece on turnovers, I explored how deliberately conceding possession provides opportunities to win the ball back and score in transition yourself. This is contingent on an important asset, though – a replicable means for winning the ball back after you concede it. For most, this is the counterpress. Counterpressing is lovely, but its also physically and tactically demanding. Many teams don’t have the squad build or assets to institute these principles effectively, and, without it, the value in conceding possession deliberately suddenly becomes a liability. Giving up the ball if you can’t get it back is purely negative.
So where does that leave the rest of us? If you can’t press like demons, is it back to relying on stagnant possession and moments of individual brilliance to create your goal scoring opportunities? Not exactly.
Consider what makes a turnover valuable in the way of creating goalscoring chances: for starters, when a team wins the ball, their opponent is typically in its attacking structure. This means they’re not compact, they have greater numbers forward, and they are in a general state of disorder trying to regain their defensive shape and slow the ball simultaneously. Defenders often have to turn their backs to the opponents goal to run back, and it is in these moments that an attacking side can exploit their adversaries decreased organization and awareness.
Now, while we take a short detour, keep these features of defensive transition in the back of your mind.
Recall Ajax’s fairytale Champions’ League run of 2018/19. That team played Bayern Munich level in the group stage twice before beating Real Madrid and Juventus on their way to the semifinals, ultimately falling to a slightly fortunate Tottenham Hotspur in Amsterdam. Unfortunately for them, things didn’t go as well the ensuing year, when they fell into the Europa League from a tough group featuring Chelsea, Lille, and Valencia. This may not seem a significant drop off in performances given the players they lost – defensive colossus Matthijs de Ligt, key cog midfielder Lasse Schöne, and Frenkie de Jong (whom we’ll discuss at greater length in a moment) all departed for greener pastures. However, Ajax’s performance drop off year to year was likely undersold: where Ajax had produced 1.83 xG per 90 (2.0 xG per 90 in the group stage) and conceded 1.31 xG per 90 (1.17 xG per 90 in the group stage) in 2018/19, they managed only 1.12 xG per 90 and conceded 1.87 xG per 90 in the 2019/20 campaign. Essentially, they went from one of the best sides in the tournament by xG difference per 90 to one of the worst in the tournament in just a few months. The main thing keeping them competitive last season was goalkeeper Andre Onana: the Cameroonian saved 0.75 post shot expected goals per 90 – essentially single handedly stopping 4-5 goals through 6 games.
So what changed? And why do we care?
Well, a few things changed – recall the major departures we discussed. Lasse Schöne’s sale left the Amsterdam side bereft of their primary set piece taker, while Matthijs de Ligt had been the side’s aerial and physical presence at the back. Not only had the Dutchman been vital at set pieces, he’d also been incredibly important in preventing opponents from exiting pressure long. This was likely a major factor in the side’s defensive regression.
Going forward, however, a different change was afoot. Frenkie de Jong was the darling of the footballing world during the 18/19 campaign. He was the daring dribbling maestro at the heart of this Ajax side, while also playing perhaps the most important role in a revitalized Dutch national team that danced out of a difficult group featuring the two most recent World Cup winners en route to the Nation’s League final. That summer he moved to Barcelona, continuing a long tradition of Ajacied exodus to the Catalan capital. With him went Ajax’s greatest weapon in settled possession – ball carrying from deep.
Recall our brief discussion of the features of defensive transitions: defenses are not compact, they have greater numbers forward, and they are in a greater state of disorder attempting to both regain their defensive structure and slow the ball.
Now, take a look at some of Frenkie de Jong’s ball carries for Ajax…
What things look like before the carry:
And what things look like after the carry:
What things look like before the carry:
What things look like after the carry:
What things look like before the carry:
What things look like after the carry:
Notice how the game state is one of settled possession when De Jong receives the ball in each of these scenarios. Then, notice the game state when he has finished carrying the ball:
1. Are the defences compact?
2. Do they have greater numbers in front of the ball than behind it?
3. Are they in disorder – specifically due to a simultaneous effort to regain structure and slow the ball?
So what was Frenkie doing for this Ajax side that they were missing in the 2019/20 season? In a concise buzzword – artificial transitions. The Dutchman was turning settled possession into an environment that closely resembled the threat of opportunities in transition, all without Ajax having to lose the ball themselves in the first place.
Now, you might be thinking: ‘Kees, what does ball carrying from deep do that passing from deep doesn’t? Does a line breaking pass not turn defenders and simulate transition?’
The answer to that is yes, but also no. Though an effective line breaking pass or switch of play can generate some of the conditions of a transition, it is less desirable for a few reasons.
First, while a line breaking pass can compromise how compact a defence is, it is far more reliant on both a) a preexisting inefficiency in the opposition defensive shape (i.e. an attacking player has to find space within the opposition defensive structure to receive the ball, which by its very nature means the defence was imperfect to begin with) and b) a teammate, the receiver, to be in position to collect the pass. In this way, a line breaking pass is far less portable and universally applicable, as there are more variables outside of the control of the passer.
Secondly, a line breaking pass fails to meet one of our conditions of a defensive transition – after having been split by a line breaking pass, the defence is not nearly as concerned with stopping the ball (the receiver almost always has their back to goal to begin with, slowing them naturally) as it is with regaining its shape.
Finally, playing line breaking passes is far more difficult under pressure (an ever proliferating facet of the modern game), whereas dribbling from deep is arguably facilitated by opposition pressure – once a player has beaten the man immediately pressing them, they can drive into far more space than they would be able to if the defence had been sitting in a static shape.
This last point is particularly important. Ajax’s performances didn’t fall off nearly as much year to year in the Dutch Eredivisie as they did in the Champions’ League from 2018-19 to 2019-20. This is likely down to nature of their opponents in the Champions’ League – overwhelmingly, continental opponents like Juventus, Real Madrid, and Chelsea were inclined to press a seemingly weaker opponent in Ajax. This played into their hands with De Jong – he would collect the ball in the defensive third, break pressure on the dribble, and transition the side into fluid positional attacks in space. Without De Jong, transition from defence to attack was far more labored under pressure, and when Ajax did finally break into the final third it was more frequently against a set defensive block. Take, for example, some of Ajax’s goals in the Champions League in 2019/20:
These, admittedly beautiful, but extremely high difficulty goals were a departure from the football Ajax had played the year before – when they transitioned from defence to attack far faster. Take a look at the numbers:
Ajax’s shot quality dropped off year to year, likely due to an increase in speculative shots from tight angles and long distance against low blocks. Set Piece xG dropped off massively as well, probably as a result of the loss of Schöne and De Ligt. However, the most interesting cliff that the Amsterdammers fell off was with regard to their counter attacking shots generated. Counterattacking shots are defined by StatsBomb as shots generated within 15 seconds of a possession that originates from a team's own half. These are exactly the sort of possessions De Jong would generate – carrying the ball out defence, breaking the opposition pressure, and starting lightning quick attacks. Without him Ajax created these chances with half the frequency they had before.
None of this is groundbreaking on its own – you might be saying to yourself, “Doesn’t everyone know beating a man disturbs defensive shape?’ You would probably be right. What makes this insight valuable is where it ties into turnovers, and how that knowledge allows us to reverse engineer our approach to the game.
Ball carrying in the early phases of buildup can turn settled possession into an environment much resembling that of a transition, affording many of the benefits of an aggressive ball recycling and counterpress system like that of Liverpool. There is one crucial difference, though – Ajax's method comes without the same stress on the defence of frequently conceding possession. If you can create a stretched, vertical game state without ceding possession, you can actually minimize the opposition’s opportunities in transition, and thus goalscoring opportunities, while maximizing your own. If a high turnover style on both sides of the ball increases the margin of victory in favor of the superior side at the expense of control, an attack dependent on ball carrying from deep does the same while maximizing control.
Incidentally, this gives us yet another explanation as to why Ajax’s defence declined from 2018/19 to 2019/20 – deprived of a controlled means to create chances, the side were forced into a style that involved far more high risk, low probability avenues. Anecdotally, this lines up. In the group stage the year following De Jong’s departure, Ajax’s attack overwhelmingly flowed through volume crosser and shooter Hakim Ziyech. This lead to some beautiful goals, but also may have played a part in the side’s struggles conceding counterattacks.
Frenkie de Jong is an instance of this phenomenon, but he is by no means the only one. Mousa Dembélé for Mauricio Pochettino’s Spurs of the mid 2010s is another great example of the phenomenon (and the falloff in performances for that Tottenham side can be easily traced with that of the Belgian's). Ismaël Bennacer of presently high flying AC Milan is yet another. These players transform gamestate for their possession dominant sides into a stretched, transition-like environment that allows them to create chances while maintaining control. This is vital – the main premise of conceding possession in order to gain possession again is not that losing the ball is good, but rather that you must lose the ball in order to win it back and create a certain state of play. With dribbling from deeper areas, we can cut out the middle man in a lot of ways – you don’t have to absorb the risk of the other side scoring in transition if you never concede possession at all. This has a massive defensive advantage over deliberate ball concession: we no longer have to play a hectic, back and forth game that is predicated on our defensive prowess, we can now stretch play while playing a controlled manner that minimizes our opponents opportunities. Not only that, but deep lying ball carriers perform this function on their own – unlike counterpressing, or passing between the lines, dribbling from deep is something one player can do in isolation to create a transition-like environment. These factors together make dribbling from deep one of the most valuable skills in the game of football today.
All data courtesy of StatsBomb via fbref.com