Expected goals (xG) tells you how good a shot was—but what about the pass that created it, or the carry that broke the line? Possession value models (PVMs) assign a number to every action on the ball, so analysts can compare build-up, progression, and risk across players and teams. Here’s how they work and how they fit with xG in 2026.
What Possession Value Models Are (and Why They Matter)
Possession value models are frameworks that estimate how much each event on the pitch changes the probability of scoring (or conceding). [cite: StatsBomb / Hudl possession value models explained] Instead of only rating shots, they rate passes, dribbles, tackles, and interceptions so that players who don’t score or assist still get credit for moving the ball into dangerous areas or stopping threats. That makes PVMs central to modern football analytics: they turn “who did what” into “how much was it worth.”
Expected Threat (xT): The Best-Known Possession Value Model
Expected threat (xT) is the most widely cited possession value metric. Introduced by Karun Singh in 2018, it splits the pitch into zones and assigns each zone a value based on the probability that a goal will be scored from that location. [cite: StatsBomb; Hudl blog] Players earn xT when they move the ball from a lower-value zone to a higher-value one—via a pass or a carry. A progressive pass into the box is worth more than a sideways pass in midfield; a carry that beats a defender and enters the penalty area is valued accordingly.
xT has limits: it typically values only ball-moving actions (passes and carries), not defensive actions or shots, and many implementations use basic event data rather than pressure or defensive line height. Even so, it’s the go-to for explaining “value in build-up” to coaches and fans, and it’s a standard output of platforms like StatsBomb.
Beyond xT: VAEP, OBV, and Other Possession Value Models
Researchers and providers have built models that go further. VAEP (Valuing Actions by Estimating Probabilities, 2020) assigns value to all actions—including defensive ones—and considers both scoring and conceding risk, though it can be sensitive to team strength. Possession value (PV) style models use short-horizon probabilities (e.g. probability of scoring in the next 10 seconds). On-ball value (OBV), used by StatsBomb, is trained on xG and evaluates every action with offensive and defensive components while avoiding some of the possession-history bias of earlier PVMs. [cite: StatsBomb; Hudl; academic work on VAEP and OBV]
In practice, xG answers “how good was the shot?” and possession value models answer “how good was everything that led to it—and everything that didn’t end in a shot?” Together they give a much fuller picture of performance than goals and assists alone.
How PVMs Fit Into 2026 Analytics
In 2026, the best practice is to combine shot-based metrics (xG, post-shot xG) with possession value (xT, OBV, or VAEP-style models) so that both finishing and build-up are measured. PVMs are also used in probabilistic forecasting and in AI tools that recommend line-ups, set pieces, or transfers. For casual fans, the takeaway is simple: every touch can be valued, and the numbers behind “best passer” or “most progressive carrier” increasingly come from these models. For more on xG, xT, and AI in football, stay tuned to ai-football.news.
Sources
- StatsBomb: Possession value models explained (statsbomb.com; Hudl blog).
- Karun Singh, expected threat (xT).
- Academic and industry work on VAEP, OBV, and possession value (StatsBomb, KU Leuven, Soccermatics).