Possession Value Explained: How Football Actions Create Attacking Value

Possession value is a family of football analytics models that estimates how a pass, carry, dribble or other action changes a team’s likelihood of scoring or conceding. It values what possession becomes, rather than simply recording how long a team has the ball or how far it moves.

A possession-value model first estimates the value of the situation before an action and then recalculates it afterwards. The difference becomes the action value. Expected Threat, or xT, is one relatively simple version of this idea; VAEP, Expected Possession Value and provider-specific models can use different outcomes, data and definitions.

What Does Possession Value Mean in Football?

Traditional possession statistics measure how much of the ball a team controls. Possession value tries to measure what the team does with that control.

A sideways pass between centre-backs may retain possession without materially improving the attack. A line-breaking pass into space between the opposition midfield and defence may create significant value even when the move does not immediately produce a shot.

The simplified principle is:

Action value = value of the situation after the action − value of the situation before the action

If a model values a situation at 0.02 before a pass and 0.07 afterwards, the pass has added 0.05 under that model.

This does not mean the pass literally created five per cent of a goal. It means the model’s estimated outcome increased by 0.05 according to its target, input data and assumptions.

Possession State Value and Action Value

The terms are closely connected, but they describe different parts of the calculation.

State value

State value is the model’s estimate of how promising a football situation is at a particular moment. A basic model might define that state only by the ball’s location. A richer model could also include:

• The positions of teammates and opponents.
• The type and direction of recent actions.
• Defensive pressure on the player in possession.
• The speed of the attack.
• The score, match time and player numbers.
• The risk of conceding if possession is lost.

The state could be valued by the probability of scoring during the possession, scoring within a set number of actions, scoring within a defined period or scoring before conceding.

Action value

Action value is the change between two states. It attributes the increase or decrease to the pass, carry, dribble, tackle, interception or turnover that connected them.

This distinction matters because a player can receive the ball in a valuable state without having created that value. Equally, a player can improve a poor state without the final position becoming especially dangerous.

Possession state describes where the attack is now. Action value describes how much the latest action changed it.

Why Possession Percentage Is Not Enough

Possession percentage records control of the ball but does not distinguish harmless circulation from productive progression.

Imagine two teams each finishing with 50% possession:

• Team A circulates the ball across its defensive line but rarely breaks the opposition midfield.
• Team B attacks more directly, progresses through central spaces and repeatedly reaches the edge of the penalty area.

The possession split treats their control as equal. The attacking consequences may be very different.

Raw passing totals can produce the same problem. A player completing 70 low-risk passes may contribute less attacking value than one completing 30 passes that repeatedly eliminate defenders or find teammates between the lines.

This is why analysts need to consider which football statistics actually capture meaningful performance, rather than treating volume as quality.

How Possession Value Models Work

There is no single universal possession-value calculation. Different models describe the game, define success and assign credit differently. Most still follow three broad stages.

1. Describe the situation before the action

A model converts the available information into a representation of the current football state. Event-data models normally use recorded actions and locations. Tracking-data models can incorporate the movement and position of the ball and players.

2. Estimate the value of that state

The model estimates a chosen outcome, such as the probability of scoring before the possession ends or the combined probabilities of scoring and conceding over a short horizon.

The target is crucial. A model predicting a goal within ten seconds is answering a different question from one estimating whether the possession will eventually produce a goal.

3. Recalculate after the action

After a pass, carry or other event, the model values the new state. Subtracting the original state value produces the estimated value of the action.

A successful pass into the penalty area will normally add attacking value. Losing the ball beside one’s own penalty area is likely to produce negative value. A backwards pass may reduce immediate threat but remain the correct tactical choice if it helps the team escape pressure or avoid a dangerous turnover.

A Worked Passing and Carrying Sequence

Consider the following illustrative attack. The numbers are invented model values used to demonstrate the calculation; they are not estimates from a real match or a universal possession-value scale.

The move produces no shot and therefore no expected goals. Conventional statistics record completed passes, carries and an unsuccessful cutback. Possession value provides a more detailed account of how the danger developed and where it was lost.

The sequence also shows why an analyst should not count only positive contributions. The progression created value, but the final turnover sharply reduced it.

How Expected Threat Fits Within Possession Value

Expected Threat is one of the best-known possession-value approaches. A typical xT model divides the pitch into zones and estimates how likely possession in each zone is to lead to a goal.

A pass or carry receives positive xT when it moves the ball into a more threatening area:

xT added = destination xT − starting xT

This makes xT useful for evaluating ball progression before a shot. It can identify line-breaking passes, carries into advanced areas and other actions that increase danger without producing a goal, assist or shot.

However, a basic location-based xT model does not observe the complete situation. Two players can receive the ball in the same zone under very different conditions. One may have space, forward momentum and several passing options; the other may be surrounded by defenders and facing away from goal.

xT is therefore best understood as one implementation of possession value, rather than a synonym for every possession-value model.

Possession Value, xT and Field Tilt Compared

Field tilt measures which team controls advanced territory. Possession value asks whether particular actions and situations improved the expected attacking or defensive outcome.

A team can record strong field tilt by sustaining pressure around the final third while producing little high-value progression. Another team may have limited territory but create substantial value from a small number of fast transitions. The metrics answer related but different questions.

Other Possession-Value Frameworks

VAEP

VAEP stands for Valuing Actions by Estimating Probabilities. It values on-ball actions according to how they change the probabilities of scoring and conceding over a defined future period.

Including both sides of the calculation can reveal the cost of risky play. An ambitious forward pass may offer a large attacking reward when completed but create negative value when its failure exposes the team to a counterattack.

Expected Possession Value

Expected Possession Value, or EPV, can use tracking data to model the positions and movement of the ball and all players at a particular moment.

This richer information can help distinguish the action selected from the options available. Event data might record a completed sideways pass. A tracking model may also identify that a more valuable forward option was open.

Provider-specific models

Commercial data providers and clubs can build their own on-ball or possession-value metrics. They may differ in:

• The outcome being predicted.
• The time horizon used.
• The definition of a possession or game state.
• The actions eligible to receive value.
• The treatment of unsuccessful actions and turnovers.
• The use of event, freeze-frame or tracking data.
• The adjustment for game state, opposition and team strength.

Two providers can therefore assign different values to the same action without either number necessarily being erroneous. Outputs should not be combined or compared until their definitions and scales are understood.

What Possession Value Can Reveal

Productive ball progression

Possession value can distinguish progression that meaningfully increases danger from movement that merely gains distance. A long pass down the touchline may advance the ball without creating a strong attacking situation. A shorter pass into a central pocket may eliminate several defenders and open multiple routes to goal.

Contributions before the final action

Goals and assists concentrate credit near the end of an attack. Possession value can identify the centre-back who broke the first line, the midfielder who carried through pressure or the player who supplied the pass before the assist.

Risk and reward

Players who attempt ambitious actions may generate substantial positive value while also surrendering dangerous possession. Reviewing positive value, negative value and net contribution separately can provide a more balanced picture than ranking players only by successful progressive actions.

Team attacking patterns

Aggregated possession value can show how a team creates danger: through central combinations, wide progression, carries, direct passing or transitions. Splitting the output by zone, action type and phase can reveal patterns hidden by a single team total.

Recruitment Applications

Possession value can help recruitment teams identify players whose contribution is understated by goals, assists or pass-completion percentages.

Potential applications include:

• Finding defenders who consistently break the first line of pressure.
• Identifying midfielders who move possession into more useful states.
• Comparing a player’s positive creation with the cost of turnovers.
• Assessing how players generate value from comparable roles and situations.
• Supporting video review with a shortlist of valuable or costly actions.

The metric should not become a context-free recruitment ranking. A player in a dominant team may receive more possession, operate in better territory and benefit from stronger passing options. Role, league, team style, physical qualities and tactical fit remain important.

The intended use also matters. As explained in GoalIQAI’s comparison of recruitment models and betting models, player recruitment and match forecasting can use related evidence while targeting different decisions and time horizons.

Match-Analysis Applications

At team level, possession value can help an analyst investigate where a match was shaped rather than relying only on possession and shot totals.

An analyst might examine:

• Where each team added and lost value during build-up.
• Whether pressure forced opponents towards low-value areas.
• Which players or combinations progressed possession.
• Whether dangerous attacks came from sustained possession or transitions.
• Whether turnovers created immediate conceding risk.
• How the pattern changed after a goal, substitution or red card.

These findings can sit within a broader structured match-analysis framework alongside shot quality, tactical match-ups, team news, market expectations and uncertainty.

Possession value may help explain performance, but it is not automatically predictive. Analysts still need to test whether it adds useful information beyond established variables such as team strength, expected goals and market prices.

Defensive Actions and Off-Ball Value

Possession-value models are usually strongest when evaluating recorded on-ball actions. Defensive and off-ball contributions are harder to attribute.

A tackle or interception can receive value when it changes the probability of scoring or conceding. However, the model may struggle to credit:

• A defender whose positioning prevents a pass from being attempted.
• A pressing run that forces play backwards.
• An off-ball run that creates space for a teammate.
• Communication that preserves the defensive structure.
• A player who follows tactical instructions by avoiding an unnecessary challenge.

Event data records what happened on the ball, not every action that shaped the decision. Tracking data can improve the representation of space, pressure and available options, but it does not remove the need for modelling assumptions and video interpretation.

A low defensive possession-value total therefore does not prove that a defender contributed little. It may mean that the metric observes only a limited part of the player’s role.

Why Match Context Changes the Interpretation

The same action can have different meanings depending on the match situation.

A team protecting a one-goal lead late in the game may reasonably prefer a low-risk backwards pass. A trailing team may accept more turnover risk in exchange for rapid progression. Red cards, two-leg ties and competition formats can also change the cost of retaining or losing possession.

This connects possession value with game state in football analytics. If the model does not account for score, time, player numbers and match incentives, those factors must be added during interpretation.

Team strength and role also affect opportunity. Players in dominant sides generally receive more touches in valuable areas and operate with stronger supporting options. Per-90 figures, value per touch and possession-adjusted measures can help, but no normalisation removes every contextual difference.

What Possession Value Does Not Prove

• It does not prove that the highest-valued action was the correct decision. The model may not observe tactical instructions, fatigue, pressure or every available option.
• It does not measure every football contribution. Off-ball movement, positioning and communication may be absent or only indirectly represented.
• It does not make outputs from different providers comparable. Identical labels can conceal different targets, scales and datasets.
• It does not remove team and role effects. Opportunity and surrounding quality influence the value a player can accumulate.
• It does not establish future performance by itself. A descriptive metric must be tested before it is used in a predictive model.
• It does not mean every negative action was a mistake. Reducing immediate attacking value can be tactically sensible when control or defensive security matters more.

How to Interpret Possession Value Properly

Before using a possession-value figure, ask:

• What outcome is the model estimating?
• What time horizon does it use?
• How is the game state represented?
• Does it measure both scoring and conceding risk?
• Which actions can receive value?
• Are unsuccessful actions and turnovers included?
• Does it use event data, tracking data or both?
• Does it adjust for game state, pressure and opposition strength?
• How are team style, possession share and player role handled?
• Has the output been validated for the decision being made?

Possession value is not one universal statistic. It is a modelling family. The number becomes meaningful only when the analyst understands what is being valued, how the value changes and which parts of football remain outside the model.

Key Takeaways

• Possession value estimates how an action changes a team’s likelihood of scoring or conceding.
• State value describes how promising a situation is; action value measures the change between two states.
• Expected Threat is a possession-value approach focused mainly on progression between pitch locations.
• Field tilt measures advanced territorial control, while possession value estimates the quality created or lost through actions.
• Different models can produce different outputs because their targets, data, time horizons and action definitions vary.
• Possession value can reveal build-up contribution, progression and turnover risk before a shot occurs.
• Defensive positioning, off-ball movement, tactical instructions and team context may be only partly represented.
• The metric supports recruitment and match analysis but does not independently prove player quality, decision quality or predictive value.

Related Guides

• Expected Threat (xT) Explained
• Field Tilt Explained
• Game State in Football Analytics Explained
• How Professional Football Bettors Build a Match Analysis Framework

Stay Ahead of the Market

Explore the GoalIQAI Football Betting & Analytics Knowledge Base and subscribe for new evidence-led guides to football metrics, analytical methods and better decision-making.