AIでサッカーの試合を見極める方法 — 2分で数字を読む

How to Scout Football Matches with AI — Read the Numbers in 2 Minutes

This is Part 1 of a three-part series. Part 1 (this article) covers SCOUTING — how to find matches worth investigating. Part 2 covers EXECUTION — reading Asian Handicap lines, Over/Under, and understanding settlement. Part 3 covers REVIEW — evaluating performance over time and learning from results.

>

A video walkthrough of this article is coming soon. When it's live, it will be embedded here.

<!-- VideoObject placeholder: embed URL will replace this comment -->

---

What This Article Is (and What It Isn't)

This is not a statistics lecture. You don't need to understand Poisson distributions or maximum likelihood estimation to use OddsFlow predictions.

This is a practical walkthrough of the OddsFlow predictions page — every number, what it means, and what to do with it. By the end, you'll know how to scan 30 matches in under two minutes and narrow them down to the two or three that deserve a closer look.

We use a real match — Sao Paulo vs Atletico Paranaense (Brazilian Serie A, July 23 2026) — as the running example. The AI predicted away win, the live signals acted on it, and the match ended 1-2. You'll see exactly what the numbers said before kickoff and how the signals played out in real time.

We call this first step SCOUTING. It's the first thing you do before making any decision.

---

Step 1: The Match List — Your Starting Point

When you open the Predictions page, you see a list of upcoming matches grouped by league. Each card shows the two teams, the kickoff time, and a set of numbers.

Before diving into numbers, notice the four tabs at the top:

Start with "Upcoming." That's your scouting ground.

You'll also see a league filter on the left (desktop) or top (mobile). If you follow specific leagues, use it. If you're open to anything, leave it on "All Leagues" — sometimes the best value hides in a league you don't follow.

There's also a Favourites star. Tap it on any match to save it. This is useful when you're scouting in the morning and want to revisit matches before kickoff.

---

Step 2: The Three Numbers That Matter Most

Click on any match card and you'll see the detailed prediction view. Let's use our real example: Sao Paulo vs Atletico Paranaense (Brazilian Serie A, at the MorumBIS in Sao Paulo).

2a. Win Probability Bar

The first thing you see is a horizontal bar split into three colours:

Sao Paulo 22.7%  |  Draw 31.2%  |  Atletico Paranaense 46.1%

These three numbers always add up to 100%. They represent OddsFlow's estimate of each outcome's probability, generated by the Dixon-Coles model — a bivariate Poisson regression that calculates attack and defence ratings for every team in the league.

What to look for in this match:

• Away favourite despite playing away. The model gave Atletico Paranaense a 46.1% win chance — almost double Sao Paulo's 22.7%. That's a strong call. When a team is favoured at this level on the road, the model's attack-vs-defence ratings are saying the away side is significantly stronger.
• High draw probability (31.2%). Nearly one in three. The model sees a real possibility of a stalemate. Combined with the away lean, this tells you Sao Paulo's attack is weak (they're unlikely to win) but their defence can hold for a while (draws are common).
• A lopsided split like 65% / 20% / 15% tells you the model sees a clear favourite. That doesn't mean the favourite is worth betting on — it depends on the odds the bookmaker offers. A 65% chance at odds of 1.40 is worse value than a 35% chance at odds of 3.50.

How it works: The Dixon-Coles model estimates an attack parameter and a defence parameter for each team. These parameters are combined with a home advantage factor to produce a Poisson distribution of goals for each side. The joint probability of all possible scorelines gives you the win/draw/loss split.

2b. Expected Goals & Over 2.5

Below the win probability bar, you'll find:

Expected Goals: 1.99    Over 2.5: 32.2%    BTTS: 38.1%

Expected Goals (xG): 1.99

This is the total number of goals the model expects in this match — derived from the same Dixon-Coles score matrix.

For Sao Paulo vs Atletico Paranaense, 1.99 expected goals means the model anticipated a low-scoring affair. The individual expected goals break down as Sao Paulo ~0.73 and Atletico Paranaense ~1.25. The away team was expected to score more, but neither side at a high volume.

Over 2.5: 32.2%

Only a 32.2% chance of three or more goals. The model was saying this was most likely a one-goal or two-goal match. Under 2.5 was the favourite at nearly 68%.

BTTS (Both Teams to Score): 38.1%

Both Teams to Score is a simple yes/no market. The model adds up every scoreline where both teams score at least once (1-1, 2-1, 1-2, etc.) and gives you the probability.

At 38.1%, the model leaned towards BTTS No. Combined with the away lean and low expected goals, the picture was: Atletico Paranaense wins a tight, low-scoring match — probably 0-1 or 0-2.

2c. Why These Three Numbers Come First

Win probability tells you *who* is likely to win. Expected goals tells you *how many goals* to expect. BTTS tells you about the *shape* of the match — whether both teams will contribute.

Together, they give you a 10-second snapshot. Most matches, you can glance at these three numbers and decide whether to dig deeper or move on.

Move on when: Win probability is heavily one-sided (70%+), expected goals are low (under 2.0), and BTTS is below 40%. That's a likely 1-0 grind with no interesting angles.

Dig deeper when: The model disagrees with the market (away favourite when the public assumes home), the win probability is close, expected goals are above 2.5, or BTTS is above 50%. Any sign of a competitive, open match is worth a click.

In the Sao Paulo match, the model had a clear opinion (away win) but the expected goals were under 2.0 — suggesting the value was on the result, not on goals.

---

Step 3: The Score Spread — The Match's Fingerprint

Scroll down on any match detail page and you'll see the score distribution — a ranked list of the most probable exact scorelines. Here's what the model showed for Sao Paulo vs Atletico Paranaense:

The Dixon-Coles model doesn't just predict who wins — it computes a probability for every scoreline from 0-0 to 7-7 (an 8×8 matrix of 64 possible outcomes). The numbers you see here are the top six, ranked by probability.

What this distribution told us:

The most likely score was 0-1 (away win, clean sheet) at 16.5%. Three of the top four scores were away wins or draws. The model's conviction was clear: Atletico Paranaense was the side to back.

Notice 1-2 at 7.9% — that was the actual final score. It ranked 6th in the distribution. The model had it in the right neighbourhood even if it wasn't the top prediction.

How to read distribution shapes in general:

• 0-1 leading (like this match): The model sees a dominant away side that wins without conceding much. The Under is favoured, and the match is expected to be tight.
• Tight cluster (top 6 all between 8%–13%): The model has no strong opinion. Goals could go either way. This is a genuinely open match.
• 0-0 in the top 2: Both teams have weak attacks or strong defences. Under 2.5 is heavily favoured.
• 2-1 and 1-2 both above 8%: The model sees goals from both sides but one team slightly ahead. Classic BTTS Yes territory.

The score spread is a fingerprint. Two matches can have the same win probability but completely different score distributions — one expects 1-0 or 0-1, the other expects 2-1 or 1-2. The spread tells you which type of match the model is predicting.

---

Step 4: Edge Percentages — The Scouting Endpoint

This is where scouting ends and decision-making begins.

On certain markets, OddsFlow displays an edge percentage — a number like +3.1% or +6.2% next to a line. This represents the difference between what our model thinks the true probability is and what the bookmaker's odds imply.

Here's the logic, using numbers from the Sao Paulo match:

• 1The Dixon-Coles model says an away win has a 46.1% probability.
• 2A bookmaker offers Atletico Paranaense to win at odds of 2.50.
• 3The bookmaker's odds imply a probability of 1 / 2.50 = 40.0%.
• 4Our model says 46.1%, the bookmaker says 40.0%. The difference is +6.1% in favour of the away win.

That +6.1% is the edge — our model's disagreement with the bookmaker on this specific line.

What the edge percentage tells you:

• Positive edge (+): Our model thinks this outcome is more likely than the bookmaker's price implies. The bigger the number, the bigger the disagreement.
• Negative edge (−) or no display: Our model agrees with the bookmaker, or thinks the outcome is less likely than the price implies.
• Edge above +5%: A meaningful disagreement. Worth serious investigation.
• Edge between +1% and +3%: A small disagreement that could easily be noise. Don't act on thin edges alone.

Important reality check: An edge is not a guarantee. The model can be wrong. The bookmaker can have information our model doesn't (injuries, lineup changes, weather). Edge tells you where to look, not where to bet blindly.

Think of edge percentages as the endpoint of scouting. You started by scanning the match list, then narrowed down using win probability and expected goals, then read the score spread for match shape. The edge percentage is the final filter: "Is there a line here where our model disagrees enough with the bookmaker to warrant a closer look?"

If the answer is yes, you've found something worth investigating. That investigation — reading the Asian Handicap lines, understanding settlement, comparing models — is EXECUTION, covered in Part 2 of this series.

---

Step 5: Two Paths — SCOUTING or FOLLOW

Everything above describes the SCOUTING path. You manually scan matches, evaluate the numbers, and make your own decisions. This is the slower, more hands-on approach, and it rewards people who enjoy analysing matches.

But OddsFlow also offers a second path: FOLLOW.

The Live Signal Room generates real-time trading signals during live matches. These signals are produced by specialised models (like the Handicap Sniper and Active Trader) that monitor odds movements and generate entry points automatically.

If you follow the Signal Room, you skip the scouting phase entirely. The system tells you what to look at and when.

Both paths are valid. Many users start with FOLLOW (it's faster) and gradually move to SCOUTING as they get comfortable with the numbers. Others do both — scout before the match, then check signals during it.

---

Step 6: Proof — What Actually Happened (Sao Paulo vs Atletico Paranaense)

We've been using this match as our example. Now let's show you what happened after kickoff — because scouting is only useful if the numbers lead somewhere.

What the AI Preview said before kickoff

What the Live Signal Room did during the match

The Live Signal Room generated signals in real time as odds shifted during the match. Here's the timeline:

Minute 10 (score 0-0): The core_strategy model entered first. It placed an Asian Handicap bet on Atletico Paranaense -0.5 at odds of 2.04–2.08. The model saw value on the away side before anything had happened on the pitch. Result: WON.

Minute 19 (score 0-0): Still 0-0, and core_strategy doubled down — Atletico Paranaense Asian Handicap 0 at odds of 2.10. At this point, the AI had entered twice on the away side with zero goals scored. Result: WON.

Minute 40 (score 1-0 to Sao Paulo): This is where it gets interesting. Sao Paulo scored first. The home crowd at the MorumBIS erupted. The market shifted towards Sao Paulo.

But two of our models — Handicap Sniper and Active Trader — did the opposite. They identified that the market had *overreacted* to the home goal. While the odds on Atletico Paranaense ballooned, these models placed away handicap bets at inflated odds of 2.04 to 2.43. They also spotted Over 1.5 goals at 1.53 and Over 2.25 at 1.93.

The contrarian call was: Atletico Paranaense is still the better team. One goal doesn't change the fundamentals. Result: All away and over bets WON.

Minute 47 (score 1-1): Atletico Paranaense equalised. The models immediately spotted Over 2.5 at odds of 2.375 — the match had already hit 2 goals, momentum was shifting, and the model calculated a third goal was coming. Result: WON.

Final score: 1-2. Atletico Paranaense completed the comeback.

The scorecard

Not every signal won. Some models placed home and draw bets that lost. That's normal — multiple models run independently and sometimes disagree. The key is that the away conviction was strong enough, and the contrarian bets at minute 40 (when the market overreacted to Sao Paulo's goal) produced the largest wins.

What this proves about scouting

The AI Preview flagged Atletico Paranaense as the likely winner before kickoff (46.1% — almost double Sao Paulo's 22.7%). The score distribution put the most likely result at 0-1 (away win). The Live Signal Room then acted on that same conviction in real time, even doubling down when the score went against the prediction in the first half.

Scouting gave you the thesis. The signals executed it.

---

Putting It All Together: A 2-Minute Scouting Run

Let's walk through a real scouting session, start to finish.

0:00 — Open the Predictions page, click "Upcoming."

You see 28 matches across six leagues today. That's too many to analyse in detail. Time to filter.

0:15 — Scan win probability bars.

Most matches have a clear favourite (60%+ one way). Skip those unless you specifically follow that league. Look for matches where the model has a strong but not overwhelming opinion — something like 45% / 30% / 25%.

One match catches your eye: Sao Paulo vs Atletico Paranaense. The away team is favoured at 46.1%, which is unusual. Worth a click.

0:40 — Check expected goals and BTTS.

Expected goals: 1.99. Over 2.5 at 32.2%. BTTS at 38.1%. The model is predicting a tight match — the value is on the result, not on goals. This narrows the angle.

1:00 — Open the match detail. Read the score spread.

0-1 at 16.5% leads the distribution. Three of the top four predicted scores are away wins or draws. The model has a clear thesis: Atletico Paranaense wins a low-scoring match.

1:20 — Check edge percentages.

You see edge on the away side. The model's 46.1% away win probability exceeds what the bookmaker's odds imply. That's the disagreement you're looking for.

1:45 — Decision point.

You've found a match where the model has a specific, actionable thesis (away win) backed by conviction across win probability, score distribution, and edge. Star it. You'll check the Live Signal Room when the match goes live.

2:00 — Done.

In two minutes, you scanned 28 matches and found the one that deserved your attention. That's SCOUTING.

---

What Comes Next

This article covered the reading part — what each number on the predictions page means and how to use it for quick filtering. We also showed you what happened when both paths (SCOUTING and FOLLOW) converged on the same match.

But scouting only identifies interesting matches. It doesn't tell you what to do with them. For that, you need to understand Asian Handicap lines, Over/Under lines, quarter lines, half-won/half-lost settlement, and how different OddsFlow models approach the same match differently.

That's Part 2: EXECUTION — coming soon.

And once you've acted, you need to evaluate your results honestly over time, learn from losses, and adjust. That's Part 3: REVIEW.

SCOUTING → EXECUTION → REVIEW. The full loop.

---

*Have questions about reading the predictions page? Join the OddsFlow Community or message us on Telegram.*