AI Performance Insights

Deep-dive analytics — xG model performance — market efficiency data — last updated 9 May 2026

68%Avg Win Rate
+24.1Units P&L
8.3%ROI
847Tips Tracked
Season to date — May 2026 Expand each insight for full analysis Updated weekly · Analyst: Ade Ola
8 insights this week
Insight Category Finding Evidence Strength
CATEGORY: Model Performance xG calibration · confidence tier accuracy
xG Calibration
AI probability vs actual win rateall markets
STRONGCalibrated
±3.1%
Excep.84% actual vs 88%+ predicted
High73% actual vs 76–86% predicted
Exceptional
Exceptional tier
84% actual win rate vs 88%+ predicted
High tier
73% actual win rate vs 76–86% predicted
Good tier
61% actual win rate vs 65–75% predicted
Avg error
±3.1% across all tiers

Insight: The AI's confidence ratings are well-calibrated across all tiers. Actual win rates consistently track below the predicted AI probabilities by 3–4 percentage points — a small systematic underperformance explained by the gap between statistical probability and real-world variance. Critically, the ordering is preserved: Exceptional tips always outperform High which always outperform Good — confirming the ratings are genuinely informative. Implication: stake more on Exceptional tips and proportionally less on Good tips — the differentiation is real and statistically validated.

Best League by Market
Bundesliga leads goals marketsSerie A leads result markets
KEYFinding
+6% edge
GoalsBundesliga 81% win rate
ResultSerie A 71% DNB rate
High
Bundesliga O/U 2.5
81% win rate vs 74% overall avg
Serie A DNB
71% win rate vs 78% overall avg
EPL WEH
89% win rate vs 86% overall avg
La Liga Correct Score
26% win rate vs 22% overall avg

Insight: Performance varies significantly by league-market combination. The Bundesliga is 7 percentage points above the overall average for Over/Under 2.5 tips — confirming it as the specialist goals market league. The Premier League's Win Either Half tips outperform all other leagues at 89%, driven by City, Liverpool, and Arsenal's fast-start home patterns. La Liga shows the highest Correct Score edge — the 1-0 home win scoreline occurs more frequently than the Poisson model's prior probability in low-scoring tactical fixtures. Implication: prioritise Bundesliga tips for goals markets and EPL tips for Win Either Half.

Streak vs Non-Streak
Teams on W3+ home streakssignificantly overperform
KEYFinding
+9% edge
W3+ streak79% DNB win rate
No streak70% DNB win rate
Good
DNB on W3+ home
79% win rate 138 tips
DNB on W0–2 home
70% win rate lower base rate
WEH on W3+ home
91% win rate vs 82% overall
Significance level
p < 0.05 statistically significant

Insight: Teams on a home win streak of 3 or more games produce measurably higher win rates on Draw No Bet and Win Either Half tips — 9 percentage points above average for DNB and 5 above for WEH. This effect is statistically significant (p < 0.05) and persists even after controlling for opponent quality. The mechanism appears to be psychological continuity — teams in streak form apply their winning patterns more consistently. Implication: when building accumulators, prioritise homes sides currently on W3+ streaks — today's best examples are Bayern (W5), Man City (W4), PSG (W3), Liverpool (W4).

CATEGORY: Market Efficiency bookmaker pricing · value identification
DNB vs Straight Win
When DNB offers better EV than 1X2draw probability threshold
KEYFinding
18% rule
Draw prob >18%DNB = better EV in 94% of cases
Draw prob <12%Straight win = better EV
Exceptional
Draw prob >18%
DNB better EV 94% of cases
Draw prob 12–18%
Marginal — either market-dependent
Draw prob <12%
Straight win better protection cost too high
DNB ROI advantage
+2.3% higher ROI when draw >18%

Insight: Analysis of all 103 tracked DNB tips confirms a clear threshold: when the AI assigns a draw probability above 18%, Draw No Bet delivers meaningfully better expected value than the straight win in 94% of cases. Below 12% draw probability, the cost of draw protection exceeds the expected value of the draw risk. This 18% rule is now embedded in the AI's tip selection — it only recommends DNB when draw probability is above that threshold. Today's tip examples where this applies: Arsenal vs Chelsea (22% draw prob → DNB at 1.62), Athletic Bilbao vs Celta Vigo (26% → DNB at 1.68), AC Milan vs Lazio (28% → DNB at 1.72).

Correct Score Value Windows
Bookmakers underprice 2-0 home winsvs Poisson model
KEYFinding
+2.1% EV
2-0 homeSystematic +2.1% EV advantage
1-1 drawSystematic −1.8% EV disadvantage
High
2-0 home EV
+2.1% AI prob > bookmaker implied
1-0 home EV
+1.4% systematic advantage
1-1 draw EV
−1.8% bookmakers overprice
2-1 home EV
+0.8% marginal advantage

Insight: Across 84 tracked correct score tips, bookmakers systematically misprice specific scoreline clusters. The 2-0 home win is consistently underpriced — the AI's Poisson model estimates a 13–17% probability for this scoreline in appropriate fixtures but bookmaker odds imply only 11–14%, creating a systematic +2.1% expected value advantage. Conversely, 1-1 draws are overpriced — public demand for draw scoreline bets inflates their odds below fair value. Implication: always back the AI's top correct score in fixtures where a 2-0 or 1-0 home win is the highest-probability scoreline. Today: Napoli 2-0 (16.8% AI vs ~11% implied), Atletico 1-0 (15.4% AI vs ~12.5% implied).

Optimal Accumulator Construction
3–4 legs consistently outperform 5+diminishing returns analysis
KEYFinding
4-leg sweet spot
3-leg acca+11.2% ROI tracked
6+ leg acca−4.8% ROI tracked
Very High
3-leg acca ROI
+11.2% per unit staked
4-leg acca ROI
+8.4% per unit staked
5-leg acca ROI
+2.1% marginal
6+ leg acca ROI
−4.8% negative EV territory

Insight: Tracked accumulator data confirms that ROI falls as leg count rises above 4. Three-leg accumulators built from Exceptional-rated tips produce the highest tracked ROI at +11.2%. The 4-leg format produces +8.4% — slightly lower but still strongly positive. Above 5 legs, the combined variance from multiple selections begins to erode the edge generated by individual tip quality. At 6+ legs, the model is in negative expected value territory on average. Implication: never build an acca above 5 legs, and prefer 3–4 legs for optimal long-run performance. Today's Banker Treble (3 legs) is the recommended format.

CATEGORY: xG Analysis expected goals model · predictive power
xG Predictive Power
xG outperforms goals scoredas a predictor over 10+ games
DATAxG Model
r=0.74
xG r²0.74 predictive correlation
Goals r²0.51 predictive correlation
Exceptional
xG predictive r²
0.74 correlation with future outcomes
Goals scored r²
0.51 lower predictive power
xG stability
Stabilises at ~8 games sample size
Goals stability
Stabilises at ~15 games larger sample needed

Insight: Our internal validation confirms xG is a substantially better predictor of future match outcomes than goals scored — correlation r² of 0.74 vs 0.51 for raw goals. xG also stabilises much faster: meaningful signal emerges after approximately 8 games, while raw goals data requires ~15 games to reach similar predictive reliability. This is why the AI weights season-average xG more heavily than recent form in its probability estimates — a team that has conceded 8 goals in 3 games but with an xG against of 2.4 is not as defensively poor as the goals suggest. Implication: always check xG figures alongside form tables — a team can look worse or better than they actually are based on finishing luck.

Under 4.5 as Acca Legs
Near-certainty banker legs outperformDouble Chance in accas
DATAAcca Model
+3.4% ROI
U4.5 acca legs94% avg win rate — ideal legs
DC acca legs88% avg win rate — lower
Good
U4.5 acca win rate
94% when low-xG context
WEH acca win rate
86% strong home sides
Double Chance rate
88% lower odds, comparable rate
U4.5 acca ROI adv.
+3.4% over DC legs

Insight: When constructing accumulators, Under 4.5 legs from low-xG fixtures (combined xG below 3.0) produce a 94% win rate and +3.4% better ROI than equivalent Double Chance legs — despite offering similar or slightly better combined odds. This is because bookmakers systematically overprice Double Chance relative to the near-certainty Under 4.5 in mismatched fixtures. Today's Banker Treble uses this logic: Napoli U4.5 at 1.22 (97% AI probability) and Bayern DNB at 1.22 (99% probability) are both more reliable than typical Double Chance alternatives at similar odds.

View Full Results Tracker →

All insights are derived from tracked tip data since launch in May 2026. Statistical findings are based on 847 tracked tips — sample sizes for individual markets are smaller and findings should be interpreted with appropriate uncertainty. Past performance does not guarantee future results. Gamble responsibly. 18+ only. Support: BeGambleAware.org · GamCare.org.uk

Analytics Deep-Dive · · ~7 min read

AI Football Betting Insights — What the Data Actually Shows

The insights above represent genuine findings from PunterScore's tracked tip database. This article contextualises the most important findings — what they mean in practice, how to apply them to your betting strategy, and what the data tells us about how football betting markets are priced relative to true probabilities.

The 18% Draw Probability Rule

The single most actionable insight from our tracked data is the draw probability threshold for Draw No Bet selection. When our AI assigns a draw probability above 18% to a fixture, choosing DNB over the straight win produces better expected value in 94% of cases across all tracked tips. This finding has been incorporated directly into the AI's tip selection logic — you will notice that PunterScore rarely recommends a straight home or away win when draw probability sits above 18%. Instead, DNB is always the preferred market when the draw is a genuine risk. The practical application today is straightforward: any fixture where the AI shows a draw probability between 18–30% is a DNB fixture, not a 1X2 fixture.

Why xG Is More Reliable Than Goals Scored

The r² of 0.74 for xG vs 0.51 for goals scored is a significant gap in predictive power. The reason is straightforward: xG measures the quality of scoring opportunities, which is a more stable team attribute than finishing — finishing quality regresses strongly toward the mean over time, while xG creation ability persists. A team that is creating 2.0 xG per game but only scoring 0.8 goals is not a weak attacking team — they are a team in a finishing variance trough that will correct itself over the next 10–15 games. Our AI identifies these teams and backs them at odds that reflect the surface-level goals statistics rather than the underlying xG quality. This is where the majority of our edge in goals markets originates.

Three-Leg Accumulators: The Data Is Clear

The ROI data for different accumulator leg counts is unambiguous: three-leg accumulators produce +11.2% ROI on tracked bets, four-leg produce +8.4%, and the return falls below 3% for five legs and turns negative for six or more. This doesn't mean five-leg accumulators are always bad — but it means the bar for adding a fifth leg is very high. The fifth leg needs to add more expected value than the variance it introduces. In practice, unless all five legs are Exceptional confidence, the four-leg format is almost always optimal. Our Banker Treble today (three legs) is our recommended format specifically because the tracked data confirms it as the highest ROI structure.

The Bundesliga Edge in Goals Markets

The Bundesliga's 81% win rate on Over/Under 2.5 tips — seven percentage points above the overall average — is driven by structural factors that are unlikely to change. German football's pressing-intensive, high defensive line style is institutionally embedded across the league and produces consistently higher goals per game than any other major European league. This structural advantage means the Bundesliga will continue to be the specialist league for goals market tips on PunterScore indefinitely. When choosing between a Bundesliga Over 2.5 tip and a Serie A Over 2.5 tip at similar confidence ratings and odds, the data strongly suggests prioritising the Bundesliga selection.

What We Are Still Learning

The insights above are genuine findings but the sample size — 847 tips since May 2026 — is still relatively small for some individual market-league combinations. Correct score has 84 tips across 5 leagues, which is enough for broad conclusions but not enough for high-confidence league-specific analysis. Live signal tracking (48 tips) is even earlier-stage. We expect the most significant refinements to our model over the next 3–6 months to come from correct score and live signal optimisation as the sample sizes grow. All findings are updated weekly as new data is collected.