Sport News

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

Quick answer: A mathematical football prediction is calculated in 6 steps: gather each team’s scoring and conceding rates, adjust them for home and away effect, work out how many goals each team is expected to score in the specific fixture, use the Poisson distribution to turn those expected goals into the probability of every scoreline, add up the scorelines to get match-result and over/under probabilities, then adjust the final numbers for context like injuries and rest. Worked through a sample fixture, this turns raw data into a figure like “62% home win.” The maths is simple arithmetic, and this guide shows every number from start to finish. AMpredict runs a far more detailed version of this process across 250+ data points per match.

Most people think mathematical football prediction is a black box.

Data goes in, a percentage comes out, and the bit in the middle is assumed to be impossibly complex. It is not. The core of it is arithmetic you could do with a calculator, and the best way to understand it is to watch one fixture travel from raw numbers to a final probability, with every step shown.

That is exactly what this guide does. I will take a sample match, feed in realistic team data, and walk through the full calculation: attacking and defensive strengths, expected goals for the fixture, the Poisson step that produces scoreline probabilities, and the final adjustments. By the end you will be able to follow, and roughly reproduce, how a maths-based prediction is built. The broader theory behind it sits in the complete guide to mathematical prediction; this is the hands-on version.

What does a mathematical football prediction actually calculate?

A mathematical football prediction calculates the probability of each match outcome by first estimating how many goals each team is likely to score, then converting those goal estimates into probabilities for every scoreline and every market. The end product is a set of percentages, such as the chance of a home win, a draw, over 2.5 goals, or both teams to score.

The whole process rests on one idea: if you can estimate how many goals each team will score, you can estimate everything else.

Goals are the atomic unit of football. Once a model has a reliable estimate of each team’s expected goals in a specific fixture, the probability of any scoreline, and therefore any market, follows from there using the Poisson distribution. So the calculation really breaks into two halves: work out the expected goals for each team, then turn those into probabilities.

The example that follows uses a sample fixture with realistic numbers so every figure is transparent and reproducible. I will call them Team A (at home) and Team B (away), and give them the kind of scoring profiles you would see from an upper-mid-table and a mid-table Premier League side.

Step 1: How do you measure a team’s attacking and defensive strength?

You measure a team’s strength by comparing its scoring and conceding rates to the league average, which produces an attacking strength and a defensive strength multiplier. A team that scores more than average has an attacking strength above 1; a team that concedes more than average has a defensive weakness above 1. These ratios are the raw material for every later step.

Start with the league baseline. In a typical Premier League season, teams average roughly 1.45 goals scored per game at home and 1.15 away, but to keep the worked example clean I will use a simplified league average of 1.40 goals per team per game.

Now the sample data for our two teams across the season so far:

From these we calculate four ratios, each one dividing the team’s rate by the league average.

Team A attacking strength: 1.80 / 1.40 = 1.29. Team A defensive strength: 1.05 / 1.40 = 0.75. Team B attacking strength: 1.30 / 1.40 = 0.93. Team B defensive strength: 1.50 / 1.40 = 1.07.

So Team A scores about 29% more than average and concedes about 25% less. Team B scores about 7% below average and concedes about 7% more. In reality these raw rates would be replaced or supported by expected goals, which measures chance quality rather than finishing luck and predicts better, as covered in what expected goals means and how it predicts outcomes. For the worked example, goals keep the arithmetic visible.

Step 2: How do you adjust for home advantage?

You adjust for home advantage by applying a home and away factor to each team’s expected goals, because teams reliably score a little more at home and a little less away. In the current era that effect is modest, worth roughly a tenth to a quarter of a goal, so the adjustment is small but real.

Home advantage has been shrinking for years, as set out in home advantage in 2026, so I will use a restrained home factor of 1.10 and an away factor of 0.95. These say the home side’s expected goals get lifted 10% and the away side’s trimmed 5%.

I will fold these factors into the next step rather than applying them separately, so the expected-goals figures come out in one calculation. The key point is that the home effect is now a gentle nudge, not the heavy thumb on the scale it was a decade ago, which is why over-weighting home advantage is one of the most common errors in amateur prediction.

Step 3: How do you calculate expected goals for the fixture?

You calculate each team’s expected goals for the specific fixture by multiplying their attacking strength by the opponent’s defensive weakness, the league average, and the relevant home or away factor. This produces the single most important number in the whole process: how many goals each team is likely to score in this exact match.

The formula for each team is: attacking strength × opponent’s defensive strength × league average × home/away factor.

Team A expected goals: 1.29 (A attack) × 1.07 (B defence) × 1.40 (league avg) × 1.10 (home) = 2.13.

Team B expected goals: 0.93 (B attack) × 0.75 (A defence) × 1.40 (league avg) × 0.95 (away) = 0.93.

So the model expects roughly 2.13 goals from Team A and 0.93 from Team B in this fixture. Those two numbers, 2.13 and 0.93, are the engine of the prediction. Everything from here is turning them into probabilities. This is also the stage where a full model layers in the other high-value inputs, fixture congestion, lineup stability, and the rest of the 12 data points that move predictions most, which nudge these expected-goals figures up or down before the probability step.

Step 4: How does the Poisson distribution turn expected goals into probabilities?

The Poisson distribution turns expected goals into probabilities by calculating how likely a team is to score exactly zero, one, two, three or more goals, given its expected-goals figure. Because goals are rare, independent events, they follow this distribution closely, which makes it the standard tool for converting an expected-goals number into scoreline probabilities.

The Poisson formula gives the probability of a team scoring exactly k goals as (λ^k × e^−λ) / k!, where λ is the team’s expected goals and e is about 2.718.

Applying it to Team A (λ = 2.13) and Team B (λ = 0.93) gives each team’s chance of scoring 0, 1, 2, 3 goals:

Read the table and the match already has a shape. Team A is most likely to score one or two; Team B most likely to score zero or one. The single most probable individual scoreline here is Team A 2-0 or 2-1. Now we combine the two columns to get full probabilities.

Step 5: How do you combine the numbers into match-result probabilities?

You combine the numbers by multiplying each team’s goal probabilities together for every possible scoreline, then grouping those scorelines into outcomes. The probability of a specific scoreline is Team A’s chance of scoring its goals multiplied by Team B’s chance of scoring its goals, and summing the relevant scorelines gives the home win, draw, and away win probabilities.

For example, the probability of exactly 2-0 is Team A scoring 2 (27%) × Team B scoring 0 (39%) = 10.5%. The probability of 1-1 is Team A scoring 1 (25%) × Team B scoring 1 (37%) = 9.3%. Do this for every scoreline up to a sensible maximum and group them.

There is the headline number: a 62% home win probability for Team A. The three figures add to 100%, because exactly one must happen, which is the clean property that lets you compare them against a bookmaker’s margin-loaded odds, as explained in how bookmaker odds are calculated.

The same scoreline grid gives the other markets. Summing every scoreline with three or more total goals gives an over 2.5 goals probability of about 52%, and summing every scoreline where both teams score at least one gives a both-teams-to-score probability of about 48%.

Step 6: How do you adjust the final numbers for context?

You adjust the final numbers for context the raw model cannot see, such as injuries, suspensions, rest days, and team news, because these change a team’s true expected goals after the historical data was collected. This step moves a prediction from a pure data output to a realistic forecast, and it is where pure maths needs help.

Suppose that after the calculation above, Team A’s first-choice striker is ruled out and the side is playing its third match in seven days. Both facts lower Team A’s real expected goals below the 2.13 the model produced from season-long averages.

A reviewer might trim Team A’s expected goals from 2.13 to around 1.85 to reflect the weaker attack and fatigue, which pulls the home-win probability down from 62% to something closer to 56%, lifts the draw, and nudges the away side up. The exact adjustment is a judgement call, and it is precisely the context that fixture congestion and injury timing introduce, covered in fixture congestion and how it reshapes a prediction model. This is also why pure mathematical models plateau and why human review still beats pure AI: the arithmetic cannot read the morning’s team news.

How does this simple model compare to what AMpredict actually runs?

The six-step model above uses a handful of inputs; AMpredict runs the same underlying logic across 250+ data points per match, then adds AI pattern recognition and human review. The worked example is a faithful miniature of the real process, but a production system uses far richer inputs and more sophisticated adjustments, which is what lifts accuracy on High Confidence picks to a tracked 89%.

The differences are in depth, not in principle.

The point of the simple version is to show that nothing in mathematical prediction is mysterious. It is strengths, expected goals, Poisson, and context, scaled up. A real model just does each step with more and better data, and expresses the result as a calibrated probability with a confidence tier rather than a single raw number, which is how prediction confidence is actually calculated. You can see the finished output on the AMpredict homepage, with the full range inside the VIP packages.

The Bottom Line

A mathematical football prediction is built in six clear steps: measure each team’s attacking and defensive strength against the league average, adjust for home and away, calculate expected goals for the specific fixture, use the Poisson distribution to turn those into scoreline probabilities, sum the scorelines into match and goal markets, and adjust for the context the data cannot see. Worked through a sample fixture, Team A’s 1.80 scoring rate and Team B’s leaky defence became a 2.13 expected-goals figure, which became a 62% home-win probability, then a context-adjusted 56%.

None of it is magic. It is arithmetic plus judgement, and the only thing separating this calculator-level version from a professional model is the depth of the data and the quality of the adjustments. Understanding the steps means you can read any maths-based prediction knowing exactly what produced the number.

At AMpredict, this same logic runs across 250+ data points per match, sharpened by AI and human review into calibrated, confidence-tiered forecasts tracked at 89% on High Confidence picks.

Want to see the finished numbers rather than do the maths yourself? View today’s predictions on the AMpredict homepage, or explore the membership plans to get the full model on every fixture.