Exact scores are the market that makes bettors dream — and drains their bankrolls. A +700 line on 2-1, a +1100 on 3-1 — you can already picture the multiplied stake. Except the true probability of any specific score rarely exceeds 12%, and sportsbooks take 12 to 18% margin on this market — versus 4 to 6% on the 1X2. In plain terms: it is mathematically a losing market in the long run.
Today's most likely scorelines
Live from our Dixon-Coles goals model — probabilities included, because honesty beats hype: our top scoreline hits ~15% of the time.
{# Depuis le 21/08/2026 les scores probables sont réservés aux membres : ce bloc ne montre plus qu'UN match en démonstration. CTA vers l'essai. #}That does not mean analyzing exact scores is pointless. On the contrary: it is the core of our AI's statistical engine. Every Over/Under 2.5 or 1X2 pick ProbWin publishes comes from a 7×7 matrix where each cell holds the probability of a specific score. This article explains the Poisson method — as refined by Dixon-Coles — that generates this matrix, walks through a concrete example computed on a real Ligue 1 match, and lists the 5 most probable scores for J1 League matchday 18 on June 6, 2026.
You will leave with a clear grasp of the calculation, a mental generator you can apply to any match, and the reason why we have never published an official pick on the exact score market.
Why exact scores are soccer's hardest bet
Three structural reasons that no "system" ever escapes.
1. The distribution is flat. Across 1,000 Premier League or Ligue 1 matches, here is how observed scores break down:
| Score | Observed frequency | Fair odds (no margin) |
|---|---|---|
| 1-1 | 11.2% | 8.93 |
| 1-0 | 9.8% | 10.20 |
| 0-0 | 8.4% | 11.90 |
| 2-1 | 8.1% | 12.35 |
| 0-1 | 7.9% | 12.66 |
| 2-0 | 6.8% | 14.71 |
| 1-2 | 6.2% | 16.13 |
| 3-1 | 4.5% | 22.22 |
| Other scores | 37.1% | — |
The most common score — 1-1 — lands only once in nine matches. The top 8 combined account for barely 63% of games: 37% of results are scattered across rare scores (3-2, 4-1, 0-3, 2-3...). No model, however sophisticated, can "predict" an event at 11% with certainty. It can only estimate the probability.
2. Bookmaker margins explode. Compare the implied margin across three markets on the same match:
| Market | Typical margin | Why |
|---|---|---|
| 1X2 | 4–6% | Liquid, ultra-competitive |
| Over/Under 2.5 | 4–7% | Very high volume, attentive traders |
| Exact score | 12–18% | Low volume, emotional demand |
On exact scores, sportsbooks know bettors are chasing the thrill, not the edge. They exploit it. A fair line of 10.20 on 1-0 becomes a posted line of 8.50: a 17% gap. To beat that margin, your model would need to outperform the book by 18 points — something no public model achieves on this market.
3. End-of-match randomness is massive. In the final 15 minutes, a 1-0 scoreline has a 28% chance of ending 1-1, 12% of ending 2-0, and 6% of ending 2-1. Substitutes come on, the leading team slows down, the trailing team presses higher. The pure variance of that closing phase trips up most deterministic models.
These are the three reasons we publish our picks on 1X2 and Over/Under 2.5, never on exact scores. The model precision exists; the profitability after margin does not.
How our soccer AI calculates every score: Dixon-Coles + Poisson
The statistical engine underlying soccer analysis is the Poisson distribution applied separately to each team. The idea: if you estimate the average number of goals the home team is expected to score (λ_home) and the away team (λ_away), you can compute the probability of any final score.
P(home scores x goals) = exp(-λ_home) × λ_home^x / x!
P(away scores y goals) = exp(-λ_away) × λ_away^y / y!
P(score x-y) = P(x) × P(y)
That is the "naive" version. It works reasonably well, but it has three flaws that any serious soccer analyst knows.
Flaw 1: goals are not independent. If team A leads 2-0 at the 70th minute, they play defensively, and the probability of a 3-0 drops. Naive Poisson ignores this.
Flaw 2: low scores are underestimated. Across 50 years of data, scores of 0-0, 1-0, 0-1, and 1-1 occur more often than Poisson predicts. The model systematically pushes scores upward.
Flaw 3: timing matters. A match from last season carries less information than one from last week. Averaging goals over 5 years dilutes recent signals.
Dixon-Coles (1997) fixes flaws 1 and 2 with a correction function τ(x, y, λ_home, λ_away, ρ) applied to the four corners of the matrix (0-0, 0-1, 1-0, 1-1). The ρ parameter is fitted so that the empirical frequency of low scores matches observed data. Here is the formula we use:
import math
def poisson(k, lam):
return math.exp(-lam) * lam**k / math.factorial(k)
def tau(x, y, lam_h, lam_a, rho):
if x == 0 and y == 0:
return 1 - lam_h * lam_a * rho
elif x == 0 and y == 1:
return 1 + lam_h * rho
elif x == 1 and y == 0:
return 1 + lam_a * rho
elif x == 1 and y == 1:
return 1 - rho
return 1.0
def prob_score(x, y, lam_h, lam_a, rho):
base = poisson(x, lam_h) * poisson(y, lam_a)
return base * tau(x, y, lam_h, lam_a, rho)
The ρ parameter is negative in practice (~-0.013 in our calibrations): it boosts 0-0 and 1-1 draws while dampening overly clean 1-0 / 0-1 wins — matching what we actually observe.
For flaw 3 (timing), we apply exponential time decay on training data: a match from season N-1 gets a weight of 0.5, season N-2 gets 0.25, and so on. This is how the model progressively forgets older performances and stays responsive to current form.
Finally, λ_home and λ_away do not come out of thin air. We compute them from three fused sources:
- the attack and defensive strength of each team, time-adjusted;
- xG (expected goals) over the last 10 matches, measuring the true quality of chances created;
- home advantage, calibrated by league (1.12 in Ligue 1, 1.18 in J1 League, 1.08 in Premier League).
The full λ calculation is detailed in our xG guide for sports betting.
Score probability table: a step-by-step example
Let us take a real match computed by our engine: Saint-Étienne vs OGC Nice, Ligue 1 matchday 36 (May 26, 2026). The values come directly from our soccer_model_probs table:
- λ_home (Saint-Étienne expected) = 1.40 goals
- λ_away (Nice expected) = 1.10 goals
- ρ = -0.0131
Here is the 7×7 probability matrix calculated with the Dixon-Coles formula above:
| Home \ Away | 0 goals | 1 goal | 2 goals | 3 goals | 4 goals | 5 goals |
|---|---|---|---|---|---|---|
| 0 goals | 8.37% | 8.86% | 4.97% | 1.82% | 0.50% | 0.11% |
| 1 goal | 11.33% | 12.81% | 6.95% | 2.55% | 0.70% | 0.15% |
| 2 goals | 8.04% | 8.85% | 4.87% | 1.78% | 0.49% | 0.11% |
| 3 goals | 3.75% | 4.13% | 2.27% | 0.83% | 0.23% | 0.05% |
| 4 goals | 1.31% | 1.45% | 0.79% | 0.29% | 0.08% | 0.02% |
| 5 goals | 0.37% | 0.40% | 0.22% | 0.08% | 0.02% | 0.00% |
The 36 cells sum to 99.58% — the remaining 0.42% is the probability mass for scores beyond 5-5, negligible. Read this way, the matrix tells a clear story:
- 1-1 is the most probable score at 12.81%, yet it is still wrong more than 87% of the time;
- the 7 most likely scores (1-1, 1-0, 0-1, 2-1, 0-0, 2-0, 1-2) total 62.2%;
- the probability that home scores 3+ goals (3-x, 4-x, 5-x) is ~16%;
- the probability of Over 4.5 combined goals is only ~8%.
By summing cells in the matrix, we directly derive the markets we publish:
| Market | Model probability | Fair odds |
|---|---|---|
| Saint-Étienne wins | 43.6% | 2.29 |
| Draw | 27.0% | 3.70 |
| Nice wins | 29.4% | 3.40 |
| Over 2.5 goals | 45.2% | 2.21 |
| Under 2.5 goals | 54.4% | 1.84 |
These probabilities are the real product. The exact score is just an intermediate output — useful to the analyst, barely exploitable at the sportsbook given the margin.
Join ProbWin for free
Access our AI predictions for NBA, NHL, Soccer & Tennis. Sign up in 30 seconds.
From exact score to 1X2 and Over/Under: why ProbWin chose these markets
A common question is: "if your AI computes the full matrix, why do you not publish exact score picks?" The answer comes down to three words: margin, variance, sample size.
Margin. Back to the Saint-Étienne vs Nice example. Our model says 12.81% on 1-1 → fair line 7.81. At the best available sportsbook, that line is 7.00. For a 1-1 pick to be profitable long-term, our estimate would need to be at least 12% higher than the book's. On the same match's 1X2, the best available line is 2.25 vs our estimate of 2.29: edge of 1.8% — far more attainable, and far more likely to be real and persistent.
Variance. With 100 exact score picks at 10% probability, your 95% confidence interval on your hit rate runs from 4% to 16%. You need 500+ picks to distinguish a genuine edge from luck. With Over/Under 2.5 (probabilities around 50%), 100 picks are enough to sense the signal. The data converges ten times faster.
Sample size. Our Learner needs several hundred settled bets to validate that a pattern is robust (see our complete AI soccer predictions guide for the mechanics). On exact scores, reaching 500 picks per league takes two seasons — during which the model receives no feedback signal at all. On 1X2, the improvement loop runs every week.
That is why our primary soccer market is 1X2 + Over/Under 2.5. Not a technical limitation — a deliberate optimization: we bet where the margin leaves room for an edge, not where it crushes the signal.
Asian Handicap markets offer margins closer to 1X2 but with more complex mechanics; we discuss them in our odds, margin and probabilities guide but do not publish public picks there as of today.
The 5 most probable scores for J1 League on June 6, 2026
J1 League matchday 18 falls on Saturday, June 6, 2026. Our pipeline has not publicly covered this league at the time of publication, but here are illustrative calculations run with the same Dixon-Coles method on three featured fixtures. The λ values draw on publicly available season statistics (average xG, goals scored/conceded at home and away, time decay over the last 20 matches).
Kawasaki Frontale vs Sanfrecce Hiroshima — λ 1.55 / 1.10 (Kawasaki fifth-best attack in the league, Sanfrecce solid defensively but without a standout star, home advantage 1.18).
| Score | Probability |
|---|---|
| 1-1 | 12.20% |
| 1-0 | 10.79% |
| 2-1 | 9.34% |
| 2-0 | 8.49% |
| 0-1 | 7.62% |
Market summary: 1X2 = 47.0 / 25.6 / 26.7 — Over 2.5 = 48.8%. Tight market; no official pick published until a book line shows 5%+ edge.
Kashima Antlers vs Vissel Kobe — λ 1.30 / 1.30 (top-of-table clash, two equivalent attacks, solid defenses on both sides).
| Score | Probability |
|---|---|
| 1-1 | 12.72% |
| 1-0 | 9.49% |
| 0-1 | 9.49% |
| 2-1 | 8.16% |
| 1-2 | 8.16% |
Market summary: 1X2 = 36.4 / 26.7 / 36.4 — Over 2.5 = 47.7%. A textbook exact score trap: the top 5 barely cumulate 48%, meaning more than one match in two ends on a less probable score.
Urawa Reds vs Okayama — λ 1.65 / 0.95 (Urawa strong favorites at home against a promoted side).
| Score | Probability |
|---|---|
| 1-0 | 12.10% |
| 1-1 | 11.79% |
| 2-0 | 10.11% |
| 2-1 | 9.60% |
| 0-0 | 7.58% |
Market summary: 1X2 strongly favors Urawa, Over 2.5 ~46%. The Urawa 1-0 is the modal score — typical sportsbook line 7.50 to 8.00. Calculated probability 12.10%, fair line ~8.26. Unless you find a line above 9.00, the bet is a long-term loser despite an apparently correct read on the match.
Key insight: these three examples show how the modal score of a match rarely exceeds 12%. You can have the game dynamics exactly right and lose 8 times out of 10 on the exact score. That is why our AI systematically pivots to 1X2 or Over/Under 2.5 whenever a real edge appears.
How to use a free AI exact score generator for your bets
You can generate the 7×7 matrix for any match in five minutes, by hand, with no subscription. Here is the step-by-step method you can apply tonight.
Step 1 — Estimate the λ values. For each team, take goals scored over their last 10 matches and divide by 10. Weight it by goals scored at home / away only, based on their role in the match. Multiply λ_home by 1.1 (home advantage for most European leagues) or 1.18 (Japan, MLS). Check that your λ values are in the typical range: between 0.8 and 2.2 in 95% of cases.
Step 2 — Build the matrix. Use the Python code above (15 lines), or a spreadsheet. For Excel/Google Sheets:
P(score H-A) = POISSON.DIST(H; lambda_H; FALSE)
* POISSON.DIST(A; lambda_A; FALSE)
* tau(H, A, lambda_H, lambda_A, rho)
For a quick first pass, skip τ and set ρ = 0 — you will lose 2–3 points of accuracy on low scores, but that is acceptable for a first read.
Step 3 — Read the matrix. Identify the 5–7 most probable scores. Sum the cells where H > A (home win), H = A (draw), A > H (away win): you have your model's 1X2 distribution. Sum the cells where H + A ≥ 3: you have Over 2.5.
Step 4 — Compare to book lines. Convert your probability to a fair line (1 / probability). Compare to the best available line. If the book line is at least 5% above your fair line, you may have a genuine edge. If it is below, walk away (our value bet guide explains the full method).
Step 5 — NEVER bet the exact score itself. That is the trap to avoid at all costs. Use the matrix to understand the match structure, then put your picks on 1X2 or Over/Under 2.5 — that is where the book margin leaves you a fighting chance.
Classic mistakes we see repeatedly in the ProbWin community:
| Mistake | Why it is fatal |
|---|---|
| Estimating λ from the last 3 matches | Massive variance, overreacts to temporary injuries |
| Ignoring context (Cup, fatigue, nothing at stake) | A 10-match average does not capture a dead-rubber end-of-season game |
| Stacking exact score bets | Variance compounds: 5 exact score bets = near-zero chance of winning all |
| Treating a +2400 line as "value" without a calculation | Probability must be > 4% AND book must show > 5% edge. Very rare. |
Frequently asked questions: AI soccer exact score
Can an AI really predict the exact score of a soccer match? No — no AI "predicts" the exact score. It calculates the probability of each score. The most likely score in a given match rarely exceeds 12%. A great model estimates probabilities better than a human; it still cannot tell you which score will actually land.
What are the most common soccer scores? Across major European leagues over the last 10 seasons: 1-1 (11%), 1-0 (10%), 0-0 (8.5%), 2-1 (8%), 0-1 (8%), 2-0 (7%). The top 6 account for roughly 52% of matches. The rest is spread across 40+ different scores.
Why do sportsbooks offer attractive lines on exact scores? Because this market is very profitable for them. The implied margin often exceeds 15%, versus 4–6% on 1X2. Recreational bettors chase the big payout — the book monetizes that emotional demand.
What is the difference between Poisson and Dixon-Coles? Naive Poisson assumes the two teams' goal totals are independent. Dixon-Coles corrects this assumption for low scores (0-0, 0-1, 1-0, 1-1) with an empirically fitted ρ parameter. Across 10 seasons of data, Dixon-Coles fits observed frequencies 3 to 5 points better on these critical scores.
Does xG (expected goals) replace Poisson? No — xG is an input to the Poisson model. The xG over the last 10 matches is used to estimate the λ values. Poisson/Dixon-Coles then turns those λ values into a full score distribution. They are two complementary steps: xG to calibrate, Poisson to distribute. See our xG guide.
Why does ProbWin not publish exact score picks? Three reasons: the book margin (12–18%) crushes any achievable edge; variance requires 500+ picks to validate a signal; our Learner converges faster on 1X2 and Over/Under 2.5. Our AI computes the full matrix, but it pivots to the profitable markets.
What is the best free exact score generator? The most accurate generator is the one you build yourself with Poisson/Dixon-Coles and 15 lines of Python. It is free, auditable, and you understand exactly what it does. Online tools are often black boxes running naive Poisson with no recent calibration — quality varies enormously across sites.
How do I know if an exact score bet is profitable? Calculate your model probability p. The fair line is 1/p. Compare it to the best available book line. If the book line is ≥ 1.07 × your fair line, you may have a genuine edge (5%+ margin against the book). If it is lower, you lose long-term. Our value bet guide walks through the full method.
Next step: let our AI do the calculation for you
The Dixon-Coles method is elegant and educational — it is exactly what runs inside our engine. But applying it to every Ligue 1, Premier League, Serie A, La Liga, Bundesliga, J1 League, and MLS fixture, plus the Cups, plus the 2026 World Cup starting in eight days, takes hours per day, iron data-ingestion discipline, and a Learner that loops after every matchday.
That is what our AI does. You read the analysis and decide. Our soccer channel publishes daily:
- 1X2 probabilities across the 5 major European leagues
- Over/Under 2.5 picks when edge exceeds 5%
- match-by-match analysis explaining the reasoning in plain language
- the model confidence displayed as a percentage on every pick
We do not promise a guaranteed track record, because short-term variance is unpredictable. We promise the complete, shared, auditable method — that is our product.
To explore today's soccer picks, or try the channel free for 14 days:
And if you want to dig deeper into the general methodology of our soccer engine, our AI soccer predictions guide explains how Scout, Analyst, Executor, and Writer chain together to produce each analysis — from odds download to match card publication.
