VAEP Football Analytics: The Complete Guide to Valuing Every Action

Picture a midfielder who barely shows up on the scoresheet. No goals, a couple of assists all season. But watch closely and he's the one making the right call under pressure, again and again: the interception that snuffs out a counter-attack, the pass that cuts through a back four, the tackle on the edge of the box that kills a chance before it starts.

The traditional numbers ignore him, and the betting market often does too. Yet the work he does is exactly the sort that wins and loses matches, and it's worth real money to bettors who know where to look.

That's where VAEP (Valuing Actions by Estimating Probabilities) comes in. Instead of counting goals, assists or even expected goals, it puts a number on every on-ball action, based on how much that action shifts the chance of scoring or conceding. If you care about football analytics, it's one of the neatest ways to see what the scoreboard hides.

Below we explain how it works, how it stacks up against xG and xT, and where we think it can sharpen your betting.

VAEP was developed by researchers at KU Leuven in Belgium, working with SciSports, and set out in the 2019 paper Actions Speak Louder Than Goals. The idea is wonderfully simple: how much did this one action change the game?

A machine-learning model looks at every pass, tackle, dribble, clearance and shot and gives each a value built from two probabilities: the chance of the team scoring within the next 10 actions, and the chance of it conceding within that same window. The change in each, before and after the action, is what you're left with.

That dual focus is what sets VAEP apart from simpler football analytics metrics. It credits attacking contribution (raising your scoring chances) and defensive contribution (cutting your conceding chances) in a single number. For bettors, that means credit for the players whose best work never makes the match report.

Goals and assists have a blind spot the size of the pitch. A defender can dominate a match without going near the scoresheet, and a midfielder can run the game without a single headline stat to his name.

Even expected goals, the best-known advanced metric in the sport, only values shots. The centre-back with five crucial interceptions, the playmaker whose defence-splitting passes just miss their target, the winger whose pressing forces error after error: none of it registers in xG.

VAEP values everything that modern event data tracks: passes and crosses, dribbles and carries, tackles and interceptions, clearances and recoveries, fouls and lost balls, plus the shots and assists. The original research argued that this picks up information conventional player stats miss, which is exactly why it's handy for spotting blind spots in the market. It's still one model with its own assumptions, though, so if you're working on a sports betting model, treat VAEP as one input among several.

The maths involves some serious machine learning, but the idea is easy to hold in your head.

Two gradient-boosted classifiers, trained on historical match data, do the work. One predicts the chance of scoring in the next 10 actions; the other predicts the chance of conceding in the same window. The model looks at the three most recent actions to understand the context, and draws on dozens of variables: the type, result and body part of the action, where it started and finished, distance and angle to goal, the score and time remaining, home or away, and (where the data provider supplies them) extras such as pitch control and defender pressure.

The sum itself is simple:

V(action) = change in scoring probability - change in conceding probability

A fall in the conceding probability is a negative change, so it adds value.

Here's an illustrative example, with rounded figures for teaching rather than numbers from a published model. A midfielder plays a progressive pass from the centre circle to the edge of the opponent's box. The chance of scoring soon was roughly 0.6% while the ball sat in a safe area, and is roughly 5% now it's somewhere dangerous, with little change in the conceding chance. That's a VAEP of roughly +0.04 for a single pass. An action that lifts the scoring chance by 0.05 and trims the conceding chance by 0.01 scores 0.05 - (-0.01) = +0.06.

Being able to put a figure on contributions that used to be invisible is exactly what makes VAEP useful for spotting market inefficiencies.

VAEP isn't the only way to put a value on possession. Several frameworks do a similar job, and knowing how they differ helps you pick the right tool for the question you're asking.

These frameworks usually agree on the broad picture and differ on individual actions and on how much context they account for. That's why comparing them side by side beats picking one and ignoring the rest. VAEP's calling card is that it rewards the defensive side of the game, not just progress up the pitch.

For a closer look at the simplest of the group, read our guide to Expected Threat (xT) in football analytics. For the full tour, see our guide to possession value models in football analytics.

The practical value of VAEP for sports betting analytics comes down to one thing: it shows you what the standard stats don't.

Goals and assists under-rate players whose work is in build-up play, defensive actions and pressing, and VAEP tries to measure exactly that. A player with a strong VAEP per 90 but modest goal and assist numbers is a candidate for a closer look, especially if the market is pricing him on reputation or a recent scoring run. Get a full season's minutes behind the number first, though. One hot month can fool you.

What does the evidence say? The original research and later academic work report that the framework identifies players whose contribution traditional statistics undervalue, and that possession-value models can add some predictive information beyond xG. How big that gain is depends on the data and the test, so check any claimed edge on your own dataset before staking money on it.

The pre-assist

A through ball splits the defence and sets up a team-mate, who squares it for a goal. The traditional stats credit only the final pass. VAEP gives the through ball its share too, because it moved the ball into a dangerous area and lifted the team's scoring chances. (That's an illustration of the logic, not a specific match.) For anyone looking at player props, it's a way to find the creators behind the creators.

The risky back pass

A defender plays a short back pass near his own goal. VAEP can score it negative: the ball has moved away from the opposition goal, and possession in a dangerous area may raise the chance of conceding. Spot a team that keeps doing this and you've found one worth watching in goals and clean-sheet markets.

The crucial tackle

A sliding tackle on the edge of the box scores high, because it cuts the opponent's scoring chances and may win the ball back. It's how you find defenders whose value goes beyond what clean sheets and tackles-won counts can tell you.

The failed but useful action

A cross misses its target, but the clearance falls to a team-mate in a better position. Possession-value models can give that sequence a small positive value, because the ball ended up somewhere better within a short window. It's a handy lens on teams that create plenty without finishing.

Hunt for undervalued teams. Look for sides with a high aggregate VAEP but poor goal conversion. They may be creating danger more often than their goals suggest and be due some positive regression, which matters most in longer-term markets such as points totals and relegation battles.

Add it to a match model. Feed team-level VAEP in as an extra input alongside xG, then test whether it improves out-of-sample predictions before you rely on it. Gains are usually modest and depend heavily on the model and the data.

Treat live signals with care. A team stringing together high-value possession without scoring may be due a goal, but in-play signals are noisy and data feeds aren't always reliable. Japan led Belgium 2-0 at the 2018 World Cup and lost 3-2, a fair reminder of how quickly a game can turn. Our complete guide to live betting covers the in-play side in more depth.

Look at player props. High-VAEP players tend to get involved in dangerous situations more often, which can translate to shots on target, touches in the opposition box and passes into the final third. For more on those markets, see our guide to player shot markets.

Track transfers. A club signing an under-rated, high-VAEP player may outperform expectations, while one that loses its VAEP leader can slip even if the departing man was never a prolific scorer.

Profile risk and style. Teams full of players attempting high-value but risky actions tend to be more volatile, which is useful for over/under and handicap markets. Sides that create most of their value in wide areas may struggle against packed defences, while those with strong defensive VAEP can make better underdogs than the traditional stats suggest.

No metric is perfect, and knowing where VAEP's edges are makes it more useful, not less.

It only sees what happens on the ball. Off-ball movement, positioning, pressing without contact and defensive shape are invisible, so the striker whose run drags two defenders out of position gets no credit for the space he creates. It also leans on event data: different providers can produce different values for the same match, and not every league has detailed coverage.

Individual values are hard to explain in plain language, so look at the aggregate patterns rather than single actions. Match to match, VAEP is only moderately stable. Shots get high values but are rare and noisy, which is why some analysts leave them out of aggregate ratings. And players at stronger teams tend to pile up positive VAEP simply because their sides spend more time in favourable game states, so compare like with like.

When you compare VAEP vs xG, the key distinction is scope. Expected goals measures shot quality, the probability that a shot becomes a goal. VAEP measures action value: how any action, not just a shot, affects the probability of scoring or conceding.

Our view is that you don't have to choose. xG tells you about chance quality and finishing, VAEP tells you about overall contribution, and the two together give you a much fuller picture than either alone.

• Use per-90 VAEP to find under-rated players. Check the sample size first: a full season's minutes, not a good month.
• Read team VAEP alongside xG, not instead of it. When the two disagree with the results, you've found a team worth a second look.
• Look for high-VAEP, low-scoring teams in long-term markets. Points totals and relegation battles are where positive regression has time to show up.
• Test any VAEP-based model out of sample before staking real money on it.
• Keep live stakes small. In-play signals are noisy, and data feeds can lag.
• Know the limits: on-ball actions only, dependent on the data provider, and not very stable from game to game.

Used carefully, VAEP helps you ask better questions about players and teams, which is where the football betting strategies that work tend to start. Whatever you back, decide your stake before kick-off and stick to it.

• Possession Value Models: Football Analytics Guide for 2026
• Expected Threat (xT) in Football: Complete Analytics Guide
• How to Build a Sports Betting Model: Complete Guide
• Live Betting Explained: Complete In-Play Betting Guide 2026

18+. Please gamble responsibly. For free, confidential support visit BeGambleAware.org.