About OriginalXG

What We Do

OriginalXG generates football match predictions using Expected Goals (xG) data and statistical probability modelling. We cover the top 5 European leagues: Premier League, La Liga, Bundesliga, Serie A, and Ligue 1.

Unlike tipster sites that rely on gut feeling, recent form, or head-to-head records, our predictions are built entirely on shot quality data. xG measures the quality of chances a team creates and concedes — not just the scoreline — which makes it a far more reliable predictor of future performance.

How the Model Works

1. Data collection: We track per-shot xG data for every match across all 5 leagues, calculating rolling home and away averages for each team.
2. Expected goals calculation: For each upcoming match, we combine the home team's attacking xG with the away team's defensive xG (and vice versa) to estimate expected goals for each side.
3. Probability modelling: These expected goals feed into a statistical model that calculates the probability of every possible scoreline — from 0-0 to 5-5.
4. Outcome probabilities: We sum scoreline probabilities to get overall home win, draw, and away win percentages.

Our Data Sources

• Match and xG data: Understat — the expected goals each team created and conceded in every match across the five leagues we cover. This is what the model runs on.
• Live odds: The Odds API — match odds from UK bookmakers, pulled fresh before each set of predictions. We use these to score ourselves against the market after the fact, never as an input to the prediction itself.
• Historical odds: football-data.co.uk — real closing odds going back to 2014/15. These define which historical matches our eleven-season backtest covers.

That's everything. We don't use team news, injuries or confirmed lineups — the model has no way to read them. It works purely from the quality of chances each team creates and concedes.

Prediction Schedule

Each gameweek, our predictions are re-run as fresh data comes in:

Version	When	What's included
Early Predictions	Tuesday	The week's first predictions for all upcoming matches across the 5 leagues, with top picks highlighted. We publish on Tuesday rather than Monday so that Monday-night results are already in the numbers.
Updated Predictions	Friday	A refresh before the weekend. The predictions only change if new matches have been played since Tuesday — most weeks they won't, and we'd rather say so than pretend otherwise.
Final Predictions	Saturday morning	The last run before kickoff. This is the version we score ourselves against.
Whenever results land	Automatic	If a match finishes outside the usual weekend — a Monday night game, or a midweek round — the predictions for the next fixtures are rebuilt straight away rather than waiting for the next scheduled run.
Weekly Email	Friday	One email a week with the weekend's predictions and our running accuracy. Friday because by then every midweek result is in, so you get the most complete picture we have.

What makes a Top Pick

A Top Pick is a predicted home win at 65–75% confidence. Both halves of that are deliberate, and both come from checking where the model gets things wrong rather than from picking numbers that looked good.

Home only. At the same confidence level, our home predictions have come in 73.6% of the time and our away predictions 59.5%. Home advantage is partly made of things expected goals cannot see — crowd, travel, the rub of refereeing decisions — so when the model makes an away side a heavy favourite on shot quality alone, it is missing something.

Capped at 75%. This is the part most sites would quietly leave out. Above about 75% our model becomes over-confident: those predictions have come in 67.8% of the time, not the 78% claimed. So the model's most certain calls do not get our top badge, however certain they sound. The badge reflects how often that kind of prediction has actually landed, not how sure the model says it is.

Everything else at 55% confidence or above is flagged Worth a look — those have come in 62.2% of the time. Below 55% we say nothing, because there is nothing worth saying.

Accuracy & Transparency

Every prediction is logged and compared against actual results. Our accuracy dashboard updates automatically after every gameweek — no cherry-picking, no hiding bad weeks. We also compare our accuracy against Chris Sutton's BBC Sport predictions and the bookmaker favourite.

Backtested across 11 seasons (2014/15–2024/25) with real bookmaker odds. Full breakdown →

What We're Not

• We are not a tipping service. We don't tell you what to bet on.
• We are not affiliated with any bookmaker. We don't earn commission from gambling.
• We show you what the data says and let you draw your own conclusions.

FAQ

Why xG instead of just looking at results?

Results include a lot of noise — lucky deflections, missed penalties, goalkeeper heroics. xG strips out the randomness and measures the underlying quality of chances. A team that creates 2.5 xG per game but only scores 1.5 goals is due for a correction upward. xG helps us see that before the results catch up.

How accurate is the model?

No model predicts football perfectly — the sport is beautifully unpredictable. Our target is >45% accuracy on match outcomes (random guessing would get you ~33%) and >12% on exact scores (random is ~3%). Check our live accuracy stats to see how we're doing.

Do you cover cup matches?

Currently we focus on league matches only, where we have the richest xG data for consistent modelling. Cup matches with teams from different divisions are harder to model reliably.