Sport News

By Martin · Published 18th May 2026 · Last updated 18th May 2026

Quick answer: xG (expected goals) measures shot quality by assigning each shot a probability of scoring between 0 and 1. xT (expected threat) measures the goal-scoring danger created by moving the ball into different pitch zones, whether or not a shot follows. npxG (non-penalty expected goals) is xG with penalties removed, exposing true open-play chance creation. Used together, the 3 metrics predict outcomes at 3 different depths: xT catches build-up quality, xG catches finishing quality, and npxG catches sustainable finishing quality. AMpredict runs all 3 inside its 250+ data point model because each catches signals the others miss, contributing to 89% tracked accuracy on High Confidence picks.

Most people who use football analytics know xG.

Far fewer know xT.

Almost nobody outside professional analysis knows npxG well enough to use it correctly.

Yet the gap between using just xG and using all 3 metrics together is one of the largest single accuracy improvements available to a football predictor. The 3 metrics capture different layers of the same underlying question: how well is this team actually creating and finishing chances? Each one exposes a signal the others miss.

I run AMpredict, a UK-registered football prediction service that runs xG, xT, and npxG as part of the 12 highest-weighted data points inside our three-layer prediction methodology. This guide is the complete plain-English breakdown of all 3 metrics, when each one matters most, and how to use them together for prediction accuracy above what any single metric can deliver.

What are xG, xT, and npxG in football analytics?

xG, xT, and npxG are 3 complementary football analytics metrics that measure chance creation and finishing quality at different layers. xG measures the goal probability of each shot taken. xT measures the goal-scoring danger created by ball movement into different pitch zones, before any shot occurs. npxG measures xG with all penalty-kick contributions removed. Together they expose the full picture of how a team creates and converts scoring opportunities.

The 3 metrics work together as a chain.

xT captures what happens before a shot is taken. A team building attacks that reach dangerous zones has high xT, whether or not they finish those attacks with shots.

xG captures what happens when a shot is taken. A team creating high-quality shots has high xG, even if their overall build-up (xT) isn't dominant.

npxG captures the sustainable version of xG by stripping out penalties, which are largely non-repeatable one-off events not driven by open-play skill.

Each metric adds a layer. Using only xG is like judging a chef by their finished dish while ignoring how good the ingredients were. Adding xT gives you the ingredients. Adding npxG separates the chef's actual open-play skill from lucky penalty draws.

What is xG (expected goals)?

Expected goals (xG) is a football statistic that assigns each shot a probability between 0 and 1 of resulting in a goal, based on distance, angle, body part, defensive pressure, and shot type. Summed across a match, a team's total xG estimates how many goals they "should have" scored against average opposition finishing.

We covered xG in depth in our dedicated piece on what expected goals means in football, but the short version:

A penalty carries roughly 0.76 xG (historical conversion rate around 76%). A one-on-one with the goalkeeper sits at 0.30-0.40 xG. A header from the edge of the box sits at 0.02-0.05 xG. A 30-yard speculative shot sits below 0.02 xG.

Sum those up across a match and you get the team's total xG. Across a season, xG correlates 0.65-0.75 with future goals, making it the strongest single-metric predictor in football.

Read the full technical foundation on expected goals (xG) for the maths, but the practical point is that xG measures shot quality specifically, not build-up quality or sustainable open-play skill.

What is xT (expected threat)?

Expected threat (xT) is a football statistic that measures the goal-scoring danger created when a team moves the ball into different zones of the pitch, weighted by the probability that possession in each zone leads to a goal. Unlike xG, xT is calculated on all ball movements (passes, dribbles, carries), not just shots.

The core insight of xT is that dangerous positions matter even without shots. A team that moves the ball into the penalty area 25 times per match without shooting is doing something genuinely threatening; xT captures that. xG would miss it entirely because no shot was taken.

xT works by dividing the pitch into a grid (typically 12x8 zones) and assigning each zone a "goal probability" value based on historical data. A pass or carry that moves the ball from a low-value zone to a high-value zone generates positive xT. A pass that moves the ball backward or sideways into a lower-value zone generates negative xT.

Sum a team's xT across a match and you have a measure of build-up quality that's independent of finishing. Teams with high xT but low xG are creating dangerous build-up without finishing it (usually a striker problem or a shot selection problem). Teams with low xT but high xG are converting minimal build-up efficiently (usually set-piece dependent or counter-attack driven).

xT is a newer metric than xG (introduced in a well-known 2019 paper by Karun Singh) and is less widely available on free public data sources. Comprehensive advanced football data and detailed match analytics provide some xT-adjacent metrics like progressive passes and possession value; full xT data is more common in paid analytics environments.

What is npxG (non-penalty expected goals)?

Non-penalty expected goals (npxG) is a football statistic identical to xG except that all penalty-kick contributions are removed. Because penalties carry roughly 0.76 xG each and are largely non-repeatable events (a team can't draw penalties on demand), removing them exposes a more sustainable measure of open-play creation.

The reason npxG matters is regression.

A team that drew 6 penalties over 10 matches has 4.56 xG added to their totals purely from penalty situations. That output isn't sustainable at the same rate: penalty frequency varies significantly across seasons and depends heavily on referee tendencies, opposition fouling patterns, and VAR-era judgement calls.

npxG strips this out. It shows you what the team is creating and finishing in open play, which is the layer most predictive of future performance.

Concrete example. A team with 1.8 xG per match and 0.5 xG per match from penalties has 1.3 npxG per match. Compared to a team with 1.8 xG per match and 0.1 xG from penalties (1.7 npxG), the second team is a stronger open-play creator despite identical raw xG. The npxG reveals what the raw xG hides.

For prediction, npxG is often more useful than raw xG when projecting future performance, particularly for teams that have benefited from unusually high or low penalty award rates.

When does xG matter most vs xT vs npxG?

Each of the 3 metrics matters most in different prediction contexts. xG dominates when the shots taken tell the whole story. xT dominates when a team's build-up is disconnected from their finishing. npxG dominates when penalties have skewed recent output above or below sustainable levels.

Practical decision framework:

Use xG as your default. For most predictions, xG is the strongest single-metric signal available and should form the base of your goal-market analysis.

Add xT when the team under-shoots. A team creating xT but not converting it into shots (usually because of poor decision-making in the final third) is often under-rated by raw xG. Their build-up is real; their finishing chain has a gap.

Add npxG when penalties spike or drop. A team that has drawn an unusually high number of penalties in a recent stretch is likely to regress toward its baseline. Their raw xG overstates their sustainable output. Check npxG to see the underlying reality.

Use all 3 together for high-confidence picks. The layered signal (build-up + finishing + sustainable finishing) is stronger than any individual metric. This is why serious prediction systems including AMpredict integrate all 3.

How do xG, xT, and npxG interact?

xG, xT, and npxG interact by describing 3 stages of the same scoring pipeline: build-up (xT), finishing (xG), and sustainable finishing (npxG). Teams strong on all 3 are elite creators. Teams strong on some but weak on others reveal specific tactical or personnel weaknesses.

Consider the possible profiles a team can show across all 3 metrics.

Elite across all 3 (high xT, high xG, high npxG). The team is creating dangerous build-up, converting it into quality shots, and sustaining this in open play. Very strong prediction signal for future scoring. Examples across recent seasons include peak Manchester City and peak Liverpool.

High xT, low xG (build-up without shots). The team reaches dangerous zones but doesn't finish sequences with shots. Usually a striker problem, a shot-selection problem, or a lack of penalty-box presence. Prediction signal: expect regression upward if the shot conversion improves.

Low xT, high xG (efficient minimal build-up). The team creates limited build-up but converts it into quality chances. Usually a counter-attacking or set-piece-dependent side. Prediction signal: sustainable only if the build-up style is repeatable against varied opposition.

High xG, low npxG (penalty-dependent output). The team's raw xG looks elite but strips down dramatically when penalties are removed. Prediction signal: expect regression when penalty rate normalises.

Low across all 3 (weak overall). The team isn't creating build-up, isn't taking quality shots, and isn't finishing sustainably. Prediction signal: expect continued low output unless personnel or tactics change.

Recognising which profile a team fits is often more useful than knowing any single metric in isolation.

How do AMpredict's models use all 3 metrics?

AMpredict uses xG, xT, and npxG together across all 3 layers of its methodology: statistical modelling weights each metric based on the specific market being predicted, AI pattern recognition scans historical scenarios where similar metric profiles produced known outcomes, and human expert review verifies whether current metric readings reflect stable performance or recent disruption.

Layer 1: Statistical model weighting. For over/under markets, xG carries the heaviest weight because shot quality directly predicts goal totals. For match winner markets in tight fixtures, npxG carries more weight than raw xG because penalty rates skew short-term output. For total shots and shot-based markets, xT can outweigh both because it captures build-up dominance.

Layer 2: AI pattern recognition. The AI layer scans the 12,000+ match training base for scenarios matching the current team's metric profile. It identifies patterns like "when team A has high xT but low xG for 5+ matches against similar opposition, they regress upward within 8 fixtures 82% of the time" and applies confidence adjustments to the mathematical output.

Layer 3: Human expert review. Analysts verify whether current metric readings reflect stable underlying performance or short-term disruption. A team's npxG dropping suddenly might mean a striker injury (temporary) or a tactical system change (permanent). Human review distinguishes these cases before the prediction reaches the VIP prediction portal.

The layered approach captures signals no single metric can. Pure xG-based analysis misses build-up quality. Pure xT-based analysis misses finishing quality. Pure npxG-based analysis loses signal when penalty rates are stable. Using all 3 together with proper weighting captures the full picture.

What are the limitations of xG, xT, and npxG?

All 3 metrics share 3 core limitations: they work best across larger sample sizes, they don't fully capture individual finishing skill differences, and they don't account for game-state effects. Each metric also has metric-specific weaknesses that predictors need to understand.

Shared limitation 1: Small sample noise. Over 1-2 matches, all 3 metrics can mislead. Over 5-10 matches, they become reliable. Over 15+ matches, they become highly reliable. Always use rolling averages, not single-match spikes.

Shared limitation 2: Average-finishing assumption. All 3 metrics assume average finishing across shooters. In reality, elite strikers consistently outperform xG, and poor finishers underperform. Individual finishing skill introduces noise the metrics don't capture.

Shared limitation 3: Game-state independence. A team trailing 3-0 in the 80th minute might rack up xG and xT against a defending opposition. That output looks strong in the aggregate but reflects an artificial scenario. Serious analysis strips out or weights down late-match trailing scenarios.

xG-specific limitation: Doesn't capture defensive pressure at time of shot. Advanced xG models incorporate defender proximity, but public xG data often doesn't.

xT-specific limitation: Rewards possession-heavy play that doesn't always produce goals. A team can generate high xT through sideways passing in advanced areas without ever creating actual shots.

npxG-specific limitation: Discards genuinely repeatable penalty-drawing skill. Some teams consistently draw penalties year after year through specific tactical patterns; stripping penalties entirely may over-correct.

Serious prediction systems use all 3 metrics alongside 4-6 other signals rather than as stand-alone oracles. The 250+ data point model at AMpredict weights each metric appropriately for each market, avoiding the single-metric traps.

How can you use xG, xT, and npxG in your own predictions?

You can use xG, xT, and npxG in your own predictions in 4 steps, each doable in under 20 minutes per match. The improvement over pure form-based prediction is typically 15-25 percentage points on suitable goal markets.

Step 1: Pull xG and npxG for both teams over the last 10 matches. Use FBref or Understat. Compare xG to actual goals scored, and npxG to open-play goals scored. Flag any team diverging by 1.5+ goals as a regression candidate.

Step 2: Check the penalty differential. If one team has drawn significantly more or fewer penalties recently than their season average, factor in likely normalisation. Their raw xG may overstate or understate sustainable output.

Step 3: Look for xT-xG gaps if you have access. Where public xT data is available, check whether teams are creating xT above what their xG suggests. This signals under-conversion of good build-up, often a regression opportunity.

Step 4: Match team profiles to markets. Elite-across-all-3 teams are strong for over/under 2.5 markets. Penalty-dependent teams are risky picks for future match winners against strong defensive opposition. Under-shooting build-up teams may be undervalued in outright match winner markets in the short term but reliable in medium-term projections.

Do this for 20-30 predictions. Compare accuracy before and after. The improvement on goal markets specifically is usually the largest, because xG-family metrics are strongest at predicting goal-related outcomes.

If you'd rather skip the manual work and tap into a system where all 3 metrics run alongside 247 other data points through the three-layer methodology, AMpredict was built for exactly that.

The Bottom Line

xG, xT, and npxG are the 3 most powerful predictive metrics in modern football analytics, each capturing a different layer of the scoring pipeline: build-up quality (xT), finishing quality (xG), and sustainable finishing quality (npxG). Using them together delivers accuracy no single metric can match, and using none of them means predicting football with tools 15 years out of date.

Casual analysis stops at form and possession. Intermediate analysis reaches xG. Professional analysis uses all 3, weighted appropriately by market and by team profile. The gap between the 3 tiers is measurable, repeatable, and entirely visible in tracked accuracy figures.

At AMpredict, all 3 metrics feed the mathematical layer, get pattern-verified by the AI layer, and get human-reviewed for context that the data hasn't absorbed. That's why the tracked accuracy on our High Confidence picks stays at 89% across large samples: not because the metrics are magic, but because they're used with the discipline that turns raw stats into actual predictions.

Want xG, xT, and npxG working alongside 247 other signals? Compare AMpredict membership options and get all 3 metrics layered into every prediction, before your next weekend kickoff.