There's a metric that is produced and sold by all tracking data companies. It's called peak velocity, and it's expressed as a single figure in km/h. It's of course very useful and widely used in reporting and infotainment like this, but I think, used in isolation, it tells you almost nothing useful about a football player.
That's what this piece is about.
Before I get into what I found/did, I want to be clear about what I'm not saying. I'm not saying speed doesn't matter. It obviously does. I'm not saying tracking data companies are producing bad products — they're not, and the raw data behind this analysis is genuinely rich. What I'm saying is that collapsing a physical profile into a single peak number should not be the goal in scouting and analysis.
What the number actually measures
When a tracking system reports a player's peak velocity, it's reporting the maximum speed they hit at any point across all recorded matches. That's it. It makes no distinction between a 5-metre acceleration burst in open space with no opponent within 30 metres and a sustained 35-metre run at full speed with a defender within arm's length.
But when scouting, that distinction is actually really important.
What I built instead
The pipeline I've put together computes peak velocity for every player across several independent dimensions. Proximity to the nearest opponent at the moment of peak — ranging from alone (no opponent within 6 metres) to close (≤3m). Sprint distance — short bursts under 10 metres, medium 10-30 metres, sustained sprints over 30 metres. Sprint type — off-ball runs versus on-ball carrying. And two timing metrics: how long a player sustains their peak (time at peak, measured at ≥95% of peak speed) and how quickly they reach it from sprint start (time to peak).
The peak velocity figure is computed not as a single-frame maximum but as the median of the five fastest frames within a sprint bout — which filters out tracking glitches without losing genuine speed information. Sprints with the ball use a lower threshold of 20 km/h rather than the 25 km/h used for off-ball sprints, because carrying speed is a different physical question.
All of this runs on Gradient's World Cup 2022 tracking data. The methodology has real limitations I'll come to — it's a tournament, not a season, and the sample sizes for individual context cells can get thin — but the core argument holds regardless.
The Son-Gakpo problem
Heung-min Son peaked at 37.75 km/h at the 2022 World Cup. Cody Gakpo peaked at 37.62 km/h. For a tracking report, those are essentially identical — two elite wingers with comparable top-end speed.
Now look at the radar.
Son (blue) and Gakpo (coral), plotted across six dimensions of their sprint profile.
Son's peak alone is 37.05 km/h. His peak when an opponent is within 3 metres is 36.96 — he loses almost nothing. His time at peak is 1.17 seconds; his time to peak is 1.07 seconds. He's a runner who builds into full speed, holds it, and does so regardless of the defensive context around him.
Gakpo's peak alone is 34.29 km/h. His peak close is 30.69. His off-ball peak — the unrestricted maximum — is 37.62. His time to peak is 0.83 seconds, faster than Son's, but his time at peak is only 0.90 seconds.
The same overall number. But different profiles.
Son is a runner who sustains speed in defensive contact. Gakpo is an explosive accelerator who hits a high peak quickly and in open space but drops off significantly when the context becomes physical. For a team with space in behind, Gakpo's profile is a weapon. For a team that needs runners making contact-heavy channels in a compact defensive block, Son's profile is the more valuable one. A scout working from the headline number alone is not making that distinction.
The context cost
The scatter below maps every outfield player's peak alone against their peak when an opponent is within 3 metres.
Bubble size = number of sprint bouts. Color = close retention percentage (peak close ÷ peak alone).
The first thing is that almost everyone loses a bit of peak when there’s a player near them – which is actually intuitive – but not everyone loses the same amount of speed. Alphonso Davies, for instance, has a well-publicised peak velocity, but his retention when someone is close to him is not top-tier. This isn't a criticism of Davies, he's one of the best left-backs in the world, and it can make sense — as a defender, he's not going to use his full speed against slower opponents if it isn't needed, but a scout aware of this could use the contextualised number to watch him differently and more precisely.
The carrying question
The scatter is specifically interesting for wingers and attacking midfielders, where the gap between off-ball and on-ball peak is a real scouting variable.
Wingers and AMs highlighted (colored). All other positions grey. Color = ball retention percentage.
Son sits near the top with high retention when on the ball — one of the few who loses very little speed when carrying. Mbappé and Griezmann are nearby. Vinicius Junior has a lower off-ball peak than several players in this dataset but retains a high percentage of it with the ball — consistent with what you see from him in actual games, where his dribbling speed is exceptional even if his straight-line max is not elite.
Di Maria and Ferran Torres drop significantly when carrying the ball. For a club whose system requires carrying at pace that distinction matters before you make the decision, not after.
The CB problem
For centre-backs specifically, the contact dimension captures something that doesn't exist in any other standard physical metric: the ability to maintain speed in a physical duel.
A central defender's most demanding physical moments are when they're recovering ground with a striker's shoulder in their ribs, or sprinting back into a defensive position while shielding a runner. Peak velocity alone doesn't show whether a player can do that. Close retention does.
CBs ranked by close retention percentage — peak speed when opponent is within 3 metres as a percentage of peak speed alone. Minimum 28 off-ball sprints.
Kamil Glik at 103.81% is the extreme outlier — he appears to run marginally faster when pressed than when alone, which is either a measurement quirk or a sample size effect given the tournament's limited match count. Koulibaly at 100.27% and Romain Saïss at 99.43% are genuinely interesting for similar reasons. These are defenders who do not slow down under contact.
The gap to the bottom is significant. Nicolás Otamendi at 70.10%, Marquinhos at 70.89%, Ibrahima Konaté at 70.94%. Excellent footballers with extensive Champions League and international experience — but their peak speed numbers mask a 30-percentage-point difference in close-contact performance relative to the players at the top of this chart. Upamecano, Gvardiol and Timber – some of the most valuable CBs in the world – all retain over 80% of their peak speed when running near an opponent.
The distance question
There's a third dimension that the headline number collapses entirely: how far the player was actually running when they hit their peak.
A 5-metre acceleration burst and a 35-metre sustained sprint are completely different physical events. The first tells you about explosive power — the ability to reach high speed almost immediately from a standing or jogging start. The second tells you about sustained speed capacity — the ability to build through a run and hold near-maximum velocity across a meaningful distance. A player can be elite at one and average at the other, and the data from this tournament makes that split surprisingly stark.
Antoine Griezmann peaks at 37.17 km/h on short sprints. On sprints over 30 metres, that number drops to 27.52 — a gap of nearly 10 km/h. Ousmane Dembélé shows almost exactly the same pattern: 37.42 on short bursts, 28.91 on long runs. These are two of the most dangerous attackers in the world, and their speed is almost entirely a short-distance weapon. Now look at the other end. Gonçalo Ramos peaks at 29.20 km/h on short sprints — well below Griezmann and Dembélé — but hits 37.45 over 30 metres. Cristiano Ronaldo follows the same logic: 27.41 on short sprints, 35.17 on long ones. He barely registers in a burst context. He accelerates into his speed over distance rather than producing it immediately.
For a counterattacking system that needs runners threatening in behind over open ground, Ramos' profile is more relevant than Griezmann's even if their overall peak figures might look comparable on a standard report. For a team that presses high and needs players to generate short explosive actions across a compact pitch, the opposite is true.
Son is an interesting middle case. He generates 17.9% of his sprint bouts over 30 metres — one of the higher rates in this dataset — and peaks at 35.18 over long distance, only marginally below his short-sprint peak of 37.05. He's not a burst specialist or a pure endurance runner. He does both, which is part of what makes him so difficult to plan against.
What this is and isn't
The methodology has real limitations. The World Cup 2022 is a tournament, not a league season — some players appeared in three matches, some in seven, and the sample sizes for specific context cells can get thin enough that the numbers carry more noise than signal. Broadcast tracking also estimates a significant share of player positions rather than directly observing them, which introduces some imprecision in the velocity calculations. And the contact retention figure specifically depends on a player having a sufficient number of sprint bouts where an opponent was genuinely close — not everyone in this dataset qualifies.
But the core argument still has value. You cannot evaluate a player's specific use of their speed just by using a single number. The scout who treats Son and Gakpo as physically equivalent based on their headline peak is missing something, and that's true across many other profiles and contexts.
Data companies like Skillcorner and Gradient already have the raw material to produce this kind of contextualised profile, and can certainly do a much cleaner job than mine.
It would be interesting to see it.
See you soon.
Data: Gradient World Cup 2022 tracking data (publicly available). Claude was used to correct the draft of this post and code the pipeline.
