Seven Years After the Rejected xG Model: V-League Learns to Read Data

Long An lost 0-3 at home in Round 22 of the 2026 V-League season, and I sat in front of my computer screen as the xG table auto-updated after the final whistle. The final figure: 0.72 expected goals per match — the lowest among all 14 teams. The club from Tan An sat 13th with 22 points after 26 rounds. One week later, I submitted a 12-page data report to my editorial board. The response arrived within 48 hours, tidy and blunt: “Football is not mathematics.” The report was never published. When the season ended, Long An were relegated exactly as the model had predicted since Round 18. In 2026, I was 26 years old, working as a data analyst for a Vietnamese football publication. My daily job was to update basic statistics: passes, pass completion rates, tackles. I wanted more. I began building the first xG (expected goals) model for the V-League, based on shot location, foot used, type of play, and defensive pressure. Drawing on my experience of tracking all 26 rounds live, I manually coded every single shot. A total of 2,847 shots across 182 matches. The central problem of Vietnamese football that year was a shortage of people who could read data. Clubs signed players based on scouts’ gut feelings, contracts based on relationships, and form judged on goal counts. A player who scored eight goals in a season was always valued higher than one with six, regardless of chances created or shooting positions. I was once rejected in 2026 because of a model. Seven years later, I am paid to write about it. My model was not complicated. Each shot was assigned a probability of becoming a goal, based on distance to goal, angle, pressure from the nearest defender, and body part used. Shots inside 6 metres, wide angle, unmarked, had a probability of 0.45. Shots from beyond 30 metres, narrow angle, had a probability of 0.04. Summed across 26 rounds, I had an xG figure for every team. The 2026 results revealed three clearly separated tiers of data. The top group — Ha Noi, Than Quang Ninh, Sai Gon — averaged over 1.6 xG per match, matching their actual league positions. The middle group — Song Lam Nghe An, SHB Da Nang, Hoang Anh Gia Lai — ranged between 1.1 and 1.4. The bottom group — Long An, Can Tho, XSKT Can Tho — sat under 0.9. Long An were at the very bottom with 0.72. What mattered was the gap between xG and actual goals. Long An scored 1.02 goals per match — 0.30 above their xG. That gap came from a single individual: the club’s foreign striker converted 28% of his shots, three times the league average. In other words, Long An survived on the extraordinary efficiency of one player, not on a system. When I split the data by individual, the picture sharpened. That striker took 41% of the team’s total shots but was involved in only 27% of final attacking moves. Most of Long An’s chances depended on a single finisher. If he was injured or tightly marked, the team had no alternative. In Round 19, when he received a red card and missed two matches, Long An managed just 0.43 xG per match — the lowest in the league. Compare that with Ha Noi that season: 12 different players scored, with xG distributed evenly across the attacking line. That distribution came from tactical structure, not chance. Ha Noi won the title with 56 points, seven clear of the runner-up. A team’s xG distribution is a far more stable indicator than individual goal counts, because goals depend on a chain of random variables a model cannot control, while xG measures the quality of chances. I attached an xG table for all 14 teams to the report, along with a prediction: Long An would be relegated unless they signed at least one backup striker and one creative number 10. The editorial board replied that the data was “too dry” for readers. They were right about one thing: my readers, at that time, were not yet fluent in the language of xG. But I believed in one principle: a single match is a story. Fifty matches are the truth. The season ended. Long An finished bottom and were relegated with 24 points. My model, rejected in Round 22, proved correct across the final eight rounds. I archived every figure and never argued with the editorial board. Seven years later, Vietnamese football data sites use xG for every V-League match. A model once dismissed as “not football” is now a professional standard. But correlation is not causation. This is the point I do not want misread when I speak about xG models. The model cannot predict everything. xG measures the quality of chances, not the ability to convert them. In the 2026 V-League, the gap between xG and actual goals at the bottom of the table reflected many factors: poor attacking quality, the opposing goalkeeper’s form, pitch conditions. A data map is not the territory itself. The second point matters more: V-League clubs today use xG but still fail to price players correctly. I once advised a club during the COVID season, proposing a 20% wage cut based on a distance-run analysis of 11 key players from the 2026 season. The head coach objected because those players “had brand value.” When football returned, those key players averaged just 8.5 km per match — 1.2 km below their pre-pandemic numbers. The data was available, but it was still misread. The bottom group of the 2026 V-League did not lack talent. They lacked a system to convert talent into probability. Long An had an excellent striker but no structure to support him. When I delivered the wage-cut proposal, they looked at me as if I were heartless. I was only delivering data, not emotion. That, in essence, has been the entire problem of Vietnamese football over the past decade. Seven years after the model was rejected, the V-League still needs one more step: reading data both to evaluate and to decide. The question for the coming season is not which club has the best xG, but which club dares to sign a contract based on xG rather than reputation.