The Data Gap in Women's Football: When Algorithms Learn From Men's Matches

HOOK

Basel, the night of 27 July 2026. Spain held more of the ball, reached the box more often, and left St. Jakob-Park with silver after losing 1-3 to England on penalties, following a 1-1 draw across 120 minutes. England became the first team in history to win two consecutive Women's European Championships. The scoreboard told that story neatly, and for most viewers the story ended there.

I stayed behind with the match event file. Set against a men's European Championship semi-final from the same year, the number of positional and event data points I had in hand was roughly four times smaller. The women's match did not contain fewer phases of play. It was not less compelling either. The data collection network for women's football is simply thinner, sparser, and operated by parties that never treated it as their primary market.

"When xG rises up, I see the people sitting in front of the screen split into two worlds: those who can read, and those who can only look." On that night in Basel, both worlds were staring at a map missing nearly half its roads.

CONTEXT

To understand why, one has to look at the history of the sports data industry itself. The major providers began covering leading women's competitions — the English Women's Super League, the Women's Champions League, the NWSL — fairly early. But the depth of the data differs sharply. For a men's match in Europe's top five leagues, I can retrieve player positional data at hundredths-of-a-second resolution, zone-based touch data, and pressing metrics such as PPDA. For most women's matches, I stop at the level of basic events: passes, shots, duels, cards.

That gap produces a concrete technical consequence. The expected goals model — xG — is trained on hundreds of thousands of shots from men's football, including variables that only exist when positional data is available: shot angle, number of defenders in the ball's path, distance to the goalkeeper, body orientation of the shooter. When positional data is missing, the model is forced to reduce everything to coarse variables such as distance and shot type. It is not mathematically wrong. It is simply answering a different question from the one the match poses.

I have written about this subject before, after the 2026 World Cup. My argument then was that every model carries assumptions about the environment in which it was born. Women's football is not a scaled-down version of men's football. It is a different environment, with a different distribution of goalkeeper heights, different ball trajectories, different defensive block structures, and a different tempo of transition. Applying a model nourished on data from the first environment to the second without recalibrating parameters does not produce analysis. It produces extrapolation.

CORE

Based on my experience tracking matches across many seasons, four layers of evidence emerge.

The first layer is the audience. On 30 March 2026, Camp Nou recorded 91,553 spectators for Barcelona against Real Madrid in the Women's Champions League quarter-final. Less than a month later, on 22 April 2026, Barcelona themselves broke that record with 91,648 spectators for the semi-final against Wolfsburg. In England, the Women's Euro 2026 final at Wembley on 31 July 2026 drew 87,192 spectators as the hosts beat Germany 2-1 after extra time. These are figures that have been widely published and can be independently verified.

The second layer is commercial. Audience growth and broadcast rights have pulled investment into squads. But investment in data has not grown at the same rate. A women's club may spend money on a striker, yet rarely spends money on its own positional data collection system. The consequence is that most analysis of women's football still rests on basic event data, or on the manual observation of people sitting in front of a screen.

At the individual level, the signal of recognition is unmistakable. Aitana Bonmatí won the Women's Ballon d'Or in 2026, 2026 and 2026. Before her, Alexia Putellas took the award in 2026 and 2026. The first recipient was Ada Hegerberg in 2026. Individual fame is rising faster than the data infrastructure that serves those very players — a paradox I have not seen in any other sport.

The third layer is the model itself. On 24 May 2026, Arsenal beat Barcelona 1-0 in Lisbon to win the Women's Champions League. Barcelona were then an overwhelmingly possession-dominant side and clear favourites. A model built purely on possession share and shot counts would have given Arsenal a very low win probability. Arsenal won through a different structure: a low defensive block, compressed team distances, and the capacity to endure long spells without the ball. That is the kind of signal basic event data cannot capture, while positional data can — but positional data for that match was far thinner than for an equivalent men's fixture.

I have watched a model of my own collapse for a similar reason, albeit in a different setting. In 2026, when football returned behind closed doors, my five-year model began to drift: the draw rate rose by roughly 23% against the historical average, and home advantage contracted. I spent three months re-watching 212 Bundesliga matches after the restart to rebuild the adjustment coefficients. "Empty stadiums broke my faith in data in silence — because when the noise disappeared, I realised that data can tremble too." In women's football, the data has never been thick enough to tremble. It simply stays silent.

This has a direct market consequence. When a model is biased in one direction, the price line tilts with it. A women's team that defends well and converts chances efficiently tends to be undervalued, because the model cannot see their defensive structure. Conversely, a possession-dominant side may be overvalued, because possession share is the only thing the model reads reliably. In the Malaysian market where I work, leading women's competitions are listed only for a handful of major fixtures. Thin market depth, wide bookmaker margins, and few public data points combine to create an environment in which model error is amplified.

The fourth layer is the rulebook. Five substitutions became standard from the 2026-23 season across most major competitions. In men's football it lets deep squads rotate, while simultaneously turning the final twenty minutes into a war of attrition. In women's football, where squad depth is thinner, the effect may be even stronger — and I do not have enough data to quantify it. That is precisely the problem.

Nor does this gap exist only in Europe. The Vietnam women's national team appeared at a Women's World Cup for the first time in July and August 2026 in Australia and New Zealand, drawn in a group with the United States, the Netherlands and Portugal. All three group matches ended in defeat. But what interests me more than the scorelines is the data structure behind that team. Before the 2026 World Cup, Huỳnh Như became the first Vietnamese women's player to turn professional abroad when she joined Lank Vilaverdense in Portugal in 2026. One player going abroad creates one data point. One data point does not create a model.

In Vietnam's domestic women's championship, the number of teams is under ten, the number of rounds is small, and positional data barely exists. Analysis, where it exists, relies mainly on video and handwritten notes. Under those conditions, every conclusion about tactical trends in Vietnamese women's football must carry a line of warning about sample size.

CONTRARIAN

Here I have to argue against myself.

The easiest way to tell this story is to reduce the whole problem to a single cause: insufficient investment. If that were true, the solution would be simple — pour more data in. I do not believe in simple solutions.

"Every signal from data is not an answer; it is a door opening onto another corridor that still needs to be lit." When I tried to train a dedicated xG model purely on women's football data, the first problem to appear was not data quality but sample size. The number of shots in a leading women's season is many times smaller than in a men's season. A dedicated model on a small sample will memorise the training data and fail on new matches. A model sharing parameters with men's football carries the bias of another environment. Both choices have a price.

The third path — which I consider the more correct one — is conditional transfer: use the model architecture of men's football but recalibrate parameters with women's data, while publishing the error margin openly. It takes time, generates no attractive headline, and does not satisfy anyone who wants a decisive answer.

There is one more point I want to state plainly. Comparing women's football to men's football through metrics can backfire. Differences in shot distance, in the minute-by-minute distribution of goals, and in defensive structure may be characteristics of a sport developing along its own trajectory, rather than flaws to be corrected until it resembles men's football. "Viewers believe in drama; I believe in repetition; and drama repeats too, if one is patient enough to wait for it." But repetition only earns trust when there are enough cycles to observe. Women's football does not yet have enough data cycles for me to assert anything with high confidence.

I am 60 years old and have watched this industry for 44 years. "Age does not slow the observing eye; it only teaches me who genuinely wants to see — and mostly, no one does." What I have learned is not how to read more metrics, but how to recognise which metric I am missing.

TAKEAWAY

In the 2026-2026 season, the signal I am tracking is not the number of goals, but the number of women's matches recorded with full positional data. If that number rises, analytical quality will rise with it — not because data manufactures truth by itself, but because it allows better questions to be asked.

And if there is one thing I want readers to carry away from this piece, it is this: every metrics table has an edge. The reader's job is not to believe the number, but to know where that edge lies.