Sport News

By Martin · Published 3rd November 2026 · Last updated 3rd November 2026

Quick answer: The Poisson distribution is a mathematical formula that predicts how likely a team is to score exactly 0, 1, 2, 3 or more goals in a match, based on the average number of goals it is expected to score. It works for football because goals are rare, independent events spread across 90 minutes, which is exactly the pattern Poisson describes. Feed in a team’s expected goals, and the formula outputs the probability of each goal count; multiply two teams’ goal probabilities together and you get the chance of every scoreline. It is the single most widely used model in football prediction, and this guide shows the full scoreline grid with worked numbers. AMpredict uses an advanced version of Poisson as one component of its wider model.

If you have ever wondered how a prediction model turns “this team is good” into “there is a 27% chance of a 2-1 win,” the answer is almost always the Poisson distribution.

It is the workhorse of football forecasting. Bookmakers use it, prediction models use it, and anyone who has built a football spreadsheet has probably used it without knowing the name. Despite sounding intimidating, the idea behind it is simple, and the maths is something you can follow with a calculator.

This guide explains the Poisson distribution in plain English, shows exactly how it predicts goal counts and scorelines, builds a full scoreline grid with worked numbers, and then checks the model’s predicted scoreline frequencies against how often those scorelines actually happen in real football. By the end you will understand the tool that sits under most maths-based predictions, including the step-by-step process in our guide to how mathematical football predictions are calculated.

What is the Poisson distribution?

The Poisson distribution is a formula that calculates the probability of a given number of events happening in a fixed period, when those events are rare and occur independently at a known average rate. In football, the events are goals, the fixed period is the match, and the average rate is the team’s expected goals, which makes Poisson a natural fit for predicting how many goals a team will score.

The distribution was developed in the nineteenth century to model rare events like the number of letters arriving in a day, and it turns out football goals behave in much the same way.

The key features that make something suitable for Poisson are that the events are countable whole numbers (you can score 0, 1, 2 goals, never 1.5), they are relatively rare within the time period, and each one does not depend heavily on the last. Goals fit all three reasonably well. A team does not score continuously; it scores occasional, discrete goals across 90 minutes, which is precisely the shape Poisson was built to describe.

The one input the distribution needs is the average rate, written as the Greek letter lambda (λ). In football, λ is simply a team’s expected goals for the match. Everything else follows from that single number.

Why does the Poisson distribution work for football?

The Poisson distribution works for football because goals are low-frequency, roughly independent events, which matches the distribution’s core assumptions closely. Most teams score between zero and three goals per game, goals are spread unpredictably through the match, and one goal does not strongly dictate the next, so the real-world pattern of football scoring lines up well with what Poisson predicts.

This fit is not perfect, but it is good enough to be genuinely useful, which is why the model has endured for decades.

Consider how goals actually arrive. Across a season, if you count how often teams score exactly zero, one, two, three or more goals, the real distribution closely tracks the Poisson curve for the relevant average. Low-scoring counts are common, very high counts are rare, and the shape tapers off in the way the formula predicts. That alignment between theory and reality is what gives the model its credibility.

The main place football deviates from pure Poisson is independence. In reality, goals are not perfectly independent, a team that goes 1-0 up may sit back, a team chasing a game throws players forward, which slightly distorts the pattern. Advanced models correct for this, but the basic Poisson model remains remarkably accurate for a tool this simple, and it reliably outperforms the gut-feeling guessing that ignores the underlying numbers entirely, the same lesson behind why form tables lie.

How does the Poisson formula calculate goal probabilities?

The Poisson formula calculates the probability of a team scoring exactly k goals as (λ to the power of k, times e to the power of minus λ, divided by k factorial), where λ is the team’s expected goals and e is the constant 2.718. In plain terms, you put in the expected goals and the goal count you are interested in, and it returns the probability of exactly that many goals.

Let me make this concrete with a team expected to score 1.5 goals, so λ = 1.5.

The formula for scoring exactly 2 goals is (1.5² × e^−1.5) / 2!, which is (2.25 × 0.223) / 2 = 0.251, or about 25%. The same formula run for each goal count gives the full picture of how that team’s scoring is likely to break down:

Read down the column and the pattern is clear. A team expected to score 1.5 goals is most likely to score exactly one, fairly likely to score none or two, and increasingly unlikely to score three or more. The probabilities add up to 100% across all possible goal counts. The only input required was λ = 1.5, and that single number comes from the expected-goals work covered in what expected goals means and how it predicts outcomes.

How do you build a scoreline probability grid?

You build a scoreline probability grid by calculating each team’s goal probabilities separately with Poisson, then multiplying them together for every combination. The probability of a specific scoreline is the home team’s chance of scoring its goals multiplied by the away team’s chance of scoring its goals, because the two are treated as independent events.

Let me take a sample fixture with a home team expected to score 1.6 goals and an away team expected to score 1.0 goal.

First, run Poisson for each. The home side (λ = 1.6) has roughly these chances: 0 goals 20%, 1 goal 32%, 2 goals 26%, 3 goals 14%. The away side (λ = 1.0) has: 0 goals 37%, 1 goal 37%, 2 goals 18%, 3 goals 6%.

Now multiply each pairing. The grid below shows the probability of each scoreline, home goals down the side, away goals across the top:

Each cell is one team’s probability times the other’s. For example, a 1-0 home win is the home side scoring 1 (32%) times the away side scoring 0 (37%) = 11.8%. The single most likely scoreline in this fixture is a tie between 1-0 and 1-1, both at 11.8%. Extend the grid to higher scores and you have the probability of every realistic result in one table.

How do you turn the grid into match and goal market probabilities?

You turn the scoreline grid into match-result and goal-market probabilities by adding up the cells that belong to each outcome. All the cells where the home team outscores the away team sum to the home-win probability, the equal-score cells sum to the draw, and the rest sum to the away win. Goal markets like over 2.5 are built the same way, by summing every scoreline that qualifies.

This is the step that converts a wall of scoreline percentages into the numbers people actually use.

From the grid above, extended to its full size, the outcomes work out roughly as follows:

The match-result figures add to 100%, because one of the three must happen. These clean probabilities are what you then compare against a bookmaker’s odds, which are inflated by a margin, to judge whether there is value, the comparison explained in how bookmaker odds are calculated. A single Poisson grid, in other words, produces every market at once.

How accurate are Poisson scoreline predictions versus reality?

Poisson scoreline predictions are accurate enough to be useful but not perfect, matching real scoreline frequencies closely for common results while slightly underestimating very low-scoring and very high-scoring games. Across large samples, the model’s predicted frequency of each scoreline lands close to how often that scoreline actually occurs, which is why it remains the standard tool despite its known limitations.

Here is roughly how Poisson’s predicted scoreline frequencies compare to the real-world frequency of those scorelines across a large sample of matches:

The common scorelines match almost exactly, which is where the model earns its keep. The small gaps appear at the extremes: real football produces slightly more 0-0 draws than pure Poisson expects, because of the independence issue, teams that reach 0-0 late often actively protect the point, which the formula does not know. Advanced models add a correction for this, but the basic version is still close enough to be genuinely predictive. No model is ever exact, which is part of why sure wins are mathematically impossible: even a well-calibrated Poisson grid only ever gives probabilities, never certainties.

How does AMpredict use the Poisson distribution?

AMpredict uses an advanced, correlation-adjusted version of the Poisson distribution as one component within its wider model, rather than relying on it alone. The basic formula provides the scoreline backbone, but it is corrected for the dependence between teams and combined with 250+ other data points, AI pattern recognition, and human review before a prediction is published.

Poisson is a foundation, not the finished building.

On its own, the simple model ignores context the basic formula cannot capture, late team news, fixture congestion, the slight dependence between the two teams’ scoring. AMpredict’s version keeps the Poisson scoreline logic but feeds it richer expected-goals inputs and corrects its known weaknesses, then layers the result into the full three-layer system described in the complete guide to mathematical prediction. That combination is what lifts accuracy on High Confidence picks to a tracked 89%.

You can see the finished scoreline and market probabilities on the AMpredict homepage, with the full set across every confidence tier inside the VIP packages. The free data behind the expected-goals inputs, if you want to build your own Poisson grid, is published by FBref and Understat, with the methodology behind expected goals set out by Opta’s analysts.

The Bottom Line

The Poisson distribution is the mathematical engine behind most football scoreline predictions. It takes one input, a team’s expected goals, and returns the probability of each goal count, and by multiplying two teams’ goal probabilities together you build a full scoreline grid from which every market, match result, over/under, both teams to score, falls out. The common scorelines it predicts match real-world frequencies closely, which is why it has been the standard tool for decades.

Its limits are real but manageable. It slightly misreads the extremes because it assumes the two teams score independently, when in reality game state changes behaviour. Advanced models correct for this, but even the basic version turns raw expected goals into genuinely predictive probabilities, far better than guesswork.

At AMpredict, a corrected version of Poisson supplies the scoreline logic inside a much larger model, combined with hundreds of data points, AI, and human review into calibrated forecasts tracked at 89% on High Confidence picks.

Want the finished scoreline probabilities without building the grid yourself? See today’s predictions on the AMpredict homepage, or compare the membership plans to get the full model on every fixture.