Predicting a correct score isn't about guessing a result. It's about estimating how many goals each team should score, then turning those estimates into scoreline probabilities with the Poisson distribution. The most likely correct score falls out of that calculation. The method has five steps: gather the right data, measure each side's attacking and defensive strength, work out the expected goals for this match, adjust for context, then read the scoreline grid.
One honest caveat before we start: even when it's calculated properly, the most likely score usually has only a 10 to 15% chance of landing. We explain why in sure correct score predictions: do they exist?. The method won't make a correct score certain; it stops you picking one at random.
To keep things concrete, we'll work through a real fixture: Manchester United v Tottenham, Old Trafford, Saturday 10 October 2026 (Matchweek 6 of the Premier League). All the figures come from the 2025-26 Premier League season (380 matches), so you can check every step yourself.
Step 1: gather the right data
What genuinely helps:
- Goals scored and conceded, home and away separately, over a meaningful sample (ideally 10 matches or more).
- Expected goals (xG) created and conceded. They measure the quality of chances, not just how many went in. A team scoring well above its xG has often enjoyed some luck, and that doesn't always last.
- Competition averages: how many goals a home side and an away side score on average in that league.
- Absentees (injuries, suspensions) and probable line-ups.
- Context: what's at stake, fixture congestion, fatigue, weather, neutral venue.
What helps far less than people think:
- Head-to-head records (H2H): often few in number, sometimes years old, with different squads. Use them as a hint, not a foundation.
- The last result on its own: one spectacular 4-0 barely moves the needle over a season.
Step 2: measure attacking and defensive strength
The principle: compare each team with the average of its competition.
In the 2025-26 Premier League, 380 matches produced 580 home goals and 465 away goals:
- a home side scored on average 1.53 goals per match (580 ÷ 380);
- an away side scored on average 1.22 goals per match (465 ÷ 380).
Manchester United at home (19 matches in 2025-26):
- 39 goals scored, i.e. 2.05 per match → attack strength = 2.05 ÷ 1.53 ≈ 1.35 (35% above the league average);
- 24 goals conceded, i.e. 1.26 per match → defence strength = 1.26 ÷ 1.22 ≈ 1.03 (roughly average; home sides are compared with what away sides usually score).
Tottenham away (19 matches in 2025-26):
- 26 goals scored, i.e. 1.37 per match → attack strength = 1.37 ÷ 1.22 ≈ 1.12;
- 26 goals conceded, i.e. 1.37 per match → defence strength = 1.37 ÷ 1.53 ≈ 0.90.
A defence strength below 1 means a better-than-average defence; above 1, a leakier one.
Step 3: calculate the match's expected goals
Cross each team's attack with the opponent's defence, then multiply by the competition average:
- United's expected goals = United attack × Spurs defence × home average = 1.35 × 0.90 × 1.53 ≈ 1.84
- Tottenham's expected goals = Spurs attack × United defence × away average = 1.12 × 1.03 × 1.22 ≈ 1.41
These two numbers, usually written λ (lambda), sum up the whole match. If you have xG data, you can run the same calculation with xG created and conceded instead of goals: it's often more stable, because xG is less sensitive to luck.
Step 4: adjust for context
Averages don't know what's happening this week. This is where your own analysis matters:
- A first-choice striker missing: lowers his team's expected goals. We'll test that effect in step 5.
- A back-up goalkeeper: can push up the opponent's expected goals.
- A neutral or relocated venue: shrinks home advantage.
- A congested calendar: likely rotation, lower intensity.
- What's at stake: a team happy with a draw often shuts the game down, which lowers both λ.
- The sample itself: last season's numbers describe last season's squads. Transfers, a new manager or promotion can change a team a lot, which is why serious models blend older data with the current season's.
Stay measured: a 5 to 15% adjustment is already significant. Go beyond that and you risk swapping the data for your gut feeling.
Step 5: turn expected goals into scorelines
This is where the Poisson distribution comes in. It gives the probability that a team scores exactly k goals when it's expected to score λ on average:
P(k goals) = λᵏ × e^(−λ) / k!
With λ = 1.84 for United and λ = 1.41 for Tottenham:
| Goals | 0 | 1 | 2 | 3 | 4 |
|---|---|---|---|---|---|
| Man United (λ = 1.84) | 15.9% | 29.2% | 26.9% | 16.5% | 7.6% |
| Tottenham (λ = 1.41) | 24.4% | 34.4% | 24.3% | 11.4% | 4.0% |
A scoreline's probability is the product of the two (assuming the two attacks are independent). For example, 1-1 = 29.2% × 34.4% ≈ 10.1%. Here's the full grid (United goals down the side, Spurs goals across the top):
| Utd \ Spurs | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| 0 | 3.9% | 5.5% | 3.9% | 1.8% |
| 1 | 7.1% | 10.1% | 7.1% | 3.3% |
| 2 | 6.5% | 9.2% | 6.5% | 3.1% |
| 3 | 4.0% | 5.7% | 4.0% | 1.9% |
What the grid tells you
- Most likely score: 1-1, at 10.1%. Next come 2-1 (9.2%), then 1-0 and 1-2 (7.1% each).
- Add up the cells and you get the 1X2: 47.7% United win, 22.7% draw, 29.7% Tottenham win (figures rounded).
- Over 2.5 goals: about 63%. Both teams to score: about 64%.
The classic trap: United are favourites, yet the most likely score is a draw. That's normal. United's wins are spread across lots of scores (1-0, 2-0, 2-1, 3-1…), whereas the draw is concentrated on 0-0, 1-1 and 2-2. If you want a score consistent with the most likely outcome, take the most likely score among United wins: here 2-1 (9.2%), just behind.
The effect of one adjustment
Now suppose United's first-choice striker is ruled out and you trim their expected goals by 10% (1.84 → 1.66). The grid shifts: 1-1 rises to 10.9%, 1-0 to 7.7%, and the 1X2 becomes roughly 43% / 24% / 33%. A single absentee moves several percentage points around: that's why team news matters so much.
Remember what these numbers are: a teaching example built only from last season's goals. They are not Elofoot's prediction for this match, which uses more inputs (xG, current form, league position, team news).
The method's limits (worth knowing)
- The simple Poisson model slightly underestimates draws, especially 0-0 and 1-1, because the two teams influence each other (a side in front slows down, a side behind pushes). Corrected versions exist, such as the Dixon-Coles model.
- The data matter more than the formula: badly estimated expected goals produce a wrong grid, however perfectly you calculate it.
- Small samples mislead: early in the season, for promoted sides or for national teams who play rarely, averages are fragile.
- The unexpected stays unexpected: a red card, a penalty, a goalkeeping error.
Do the maths by hand… or not
You can reproduce this method in a spreadsheet: a few averages, two multiplications and the POISSON.DIST function in Excel or Google Sheets (with the last argument set to FALSE). It's a great exercise for understanding a match.
If you'd rather save time, this is exactly the logic behind Elofoot's predictions: a statistical model estimates each team's expected goals from the match data, calculates the 1X2 probabilities and the most likely score, then the AI explains the scenario in plain English. The numbers come from the model, never from the AI's imagination. It's all set out in our methodology and on the AI football match analysis page, and the concept is summarised in the glossary: correct score.
- Today's matches: football predictions today.
- Every Premier League fixture: Premier League predictions, and all competitions on AI football predictions.
- For the current Premier League picture after five matchweeks, read Premier League predictions: how AI calculates the correct score.
- Your first full prediction is free, no card needed.
Responsible gambling. This method is for understanding a match, not for guaranteeing a win. Betting is for adults only (18+ in the UK and most countries, older in some) and carries real risks: debt, dependency, isolation. Never stake money you need and never chase losses. If gambling stops being fun, talk to someone you trust or a local support service. In Great Britain, the National Gambling Helpline (run by GamCare) is free and open 24/7 on 0808 8020 133.
Frequently asked questions
What's the best method to predict a correct score?
Estimate each team's expected goals (one side's attack against the other's defence, ideally using xG), adjust them for context, then apply the Poisson distribution to get the probability of every scoreline. The most likely correct score is the one with the highest probability in the grid.
What are the chances of getting the correct score right?
In a typical match, the most likely score has roughly a 10 to 15% chance of happening. In this article's example, 1-1 comes out at 10.1%.
Should I rely on head-to-head records for a correct score?
With caution. H2H records are often few and old. They can confirm a trend, but recent form, xG and team news are more reliable.
Why is the most likely score often a draw when one team is favourite?
Because the favourite's win is spread over many different scorelines, while the draw is concentrated on a few (0-0, 1-1, 2-2). Each individual winning score can therefore be less likely than 1-1.
Is there an app that gives the correct score before the match?
Some tools, Elofoot included, calculate the most likely score and its probability before kick-off. None can guarantee it: be wary of any app or channel promising a "sure" or "fixed" correct score.
