Passes are the invisible architecture of a match — they open space, create chances, and sometimes end entire attacks. Measuring pass quality requires more than counting completions: it needs a probabilistic lens that judges not just whether a pass reached a teammate, but how valuable and how risky that action was. In this article I walk through what an xPass-like metric looks like, how to build a risk profile around passing choices, and how coaches and analysts can turn those numbers into practical decisions.
Why pass quality matters
Simple pass completion percentage flattens complexity. A ninety-percent passer who always plays safe square balls isn’t necessarily helping his team more than an eighty-percent passer who attempts line-breaking passes into the final third. Teams need a way to quantify the expected benefit of a pass, not just the success rate.
Good pass quality metrics highlight creativity, reward risk-taking when appropriate, and expose hidden value producers. For recruitment and training, they help separate players who preserve possession from those who consistently improve their team’s chances to score.
What is xPass?
In practice, xPass is a label often used for a modelled expected value of a pass — the probability-weighted benefit or the probability that a pass will lead to a positive event (shot, shot-on-target, or increased possession value). It borrows the logic of expected goals (xG) and expected threat (xT) to evaluate passes based on context, not outcome alone.
Technically, xPass can be implemented as a supervised model that predicts the likelihood of a successful pass and the downstream increase in scoring probability if the pass succeeds. Inputs typically include start and end coordinates, body part, pass type, opponent pressure, time in match, and information about the intended receiver’s options.
What is a risk profile?
A passing risk profile maps the downside of a pass — the chance and impact of losing possession and the field location where that loss would occur. Risk is not binary; it’s a combination of probability (how likely is a turnover) and cost (how dangerous is the position for the defending team if possession is lost).
Risk profiles can be tailored to player roles and match states. A central defender’s acceptable risk envelope differs from that of an attacking midfielder. Good profiles allow coaches to measure whether a player’s choices align with tactical instructions and to compare players on a common risk-reward scale.
How xPass and risk profile work together
Pairing xPass with a risk profile creates a two-dimensional view: expected reward on one axis and expected cost on the other. This lets analysts quantify net expected value for each pass as expected benefit minus expected cost, helping to identify high-upside, low-risk actions and flag reckless plays.
For example, a long vertical pass that raises the team’s chance to score by 0.03 (xPass = +0.03) but carries a 25% chance of a turnover in the defensive half (high risk) might be judged differently depending on scoreline and time. Combining both metrics makes that context explicit.
Measuring and computing xPass: data and features
At minimum you need event data: pass start/end location, time, pass outcome, and some annotation about pressure or body part. Tracking data adds teammate and opponent positions, which markedly improves model fidelity by capturing spatial options available to the receiver.
Typical features and why they matter:
| Feature | Why it matters |
|---|---|
| Pass distance and direction | Long or forward passes often change threat and have different completion probabilities. |
| Receiving position relative to defenders | Open receivers can control and progress play; tightly marked ones are less likely to convert a pass into a chance. |
| Pressure on passer | Under pressure, precision drops and turnover risk increases. |
| Game state and time | Teams behave differently when winning, losing, or late in the match; expected values shift accordingly. |
Model choices vary. Logistic regression or gradient boosting can predict pass success, while more complex architectures (random forests, neural nets) can estimate downstream scoring impact. The soccer-action and EPV frameworks provide patterns for chaining events and valuing outcomes.
Building a risk profile: components and calibration
Start by estimating turnover probability for each pass, either directly (model predicting turnover) or indirectly (1 – pass completion probability). Then attach a cost to losing possession at the pass end location — this cost is typically the expected threat (xT) or expected possession value (EPV) of the location for the defending team.
Calibration matters. Use historical match data to map locations and turnover contexts to actual conceded chances. Without calibration, a risk metric can overstate danger in innocuous areas or understate it in high-transition zones where turnovers lead quickly to shots.
Practical workflow: from data to insights
A practical pipeline looks like this: gather event and tracking feeds, extract features for each pass, build separate models for pass success and for downstream value, then combine outputs to compute expected reward and expected cost. Visualize results at player, lineup, and match levels.
In my work with club analysts, we iterated quickly — a baseline xPass model ran in a few hours and revealed clear signals about certain midfielders who created value despite modest completion rates. We used per-90 and per-pass normalizations to compare players fairly across minutes played and team styles.
Use cases: coaching, scouting, and in-game decisions
Coaches can use xPass and risk profiles to nudge player behavior: instruct a fullback to avoid risky passes in defensive third or encourage an attacking midfielder to accept higher turnover probability because their passes increase scoring chances. Those directives become measurable.
Scouts benefit from a standardized lens when comparing players across leagues. A player with consistently positive net expected pass value in a mid-table league may be a better buy than a flashy passer whose risks often lead to dangerous turnovers.
Pitfalls and limitations
No model captures everything. Contextual factors such as player intent, set-up phase from earlier plays, and opponent tactical adjustments are difficult to encode fully. Models trained on one league may not transfer cleanly to another due to differences in tempo and style.
Data quality is another constraint. Event data has annotation noise; tracking data is expensive and not always available. Be transparent about uncertainty, and validate models against out-of-sample matches and known tactical episodes before making high-stakes decisions.
Getting started: tools, datasets, and further reading
For practitioners, useful tools include the socceraction library for EPV-style frameworks and the common machine learning stacks (scikit-learn, XGBoost). Commercial providers such as StatsBomb, Opta/StatsPerform, and Second Spectrum offer the event and tracking data needed to build robust models.
Begin with a simple pass-success model, then add a downstream value module. Iterate and visualize: heatmaps of net expected value, per-player scatter plots of reward vs. risk, and match timelines showing cumulative net value are all effective communication tools for coaching staff.
Measuring pass quality with an xPass approach and a complementary risk profile gives teams a principled way to balance creativity and caution. It turns vague judgments into numbers you can test, explain, and act on. Start small, validate often, and use the metrics as a conversation starter between analysts and coaches — not as a final arbiter.
Sources and further reading:
