Case study
WorldCup Edge 2026 performance report.
A post-tournament review of an AI football prediction model: what it forecast correctly, where it failed, and how it should evolve into a serious analytics product.
Model ledger
Tournament snapshot
Total matches analyzed
104
Correct model picks
72
Full-model accuracy
69.2%
Estimated bankroll
€1,261.54
The model's strongest signal was prediction, not fully automated betting. Its best next form is a transparent prediction ledger and performance-reporting engine.
Financial scenarios
Three ways to read the same model.
The strict betting layer was protective. The broader prediction layer showed the real promise.
Strict value system
€1,000.00
0 official bets. The risk layer protected capital but was too conservative to monetize the model.
Group-stage prediction mode
€1,162.50
47 correct picks across 72 group matches using saved odds and equal stake sizing.
Full tournament simulation
€1,261.54
104 model picks across group stage and knockout pass-through scenarios.
Bankroll curve
The upside was hiding behind the safety layer.
Strict value system
No official plays triggered
€1,000
Group-stage prediction
47/72 correct picks
€1,162
Full tournament simulation
72/104 correct picks
€1,261
Knockout pass-through
Indicative odds model
€1,484
Knockout profit is indicative because pass-through odds were reconstructed after the tournament. Accuracy is real; simulated profit is a product-design signal, not financial advice.
Accuracy profile
Prediction quality improved when the format changed.
Group stage
47 correct / 72 matches
65.3%
Knockout pass-through
25 correct / 32 matches
78.1%
Full tournament
72 correct / 104 matches
69.2%
Tournament timeline
From experiment to evidence.
01
Pre-match engine
Built ratings, probabilities, confidence bands and edge checks before the tournament started.
02
Live tournament ledger
Tracked model picks against real results and separated prediction quality from betting triggers.
03
Post-tournament audit
Measured accuracy, bankroll simulations, missed calls and the limits of the first risk layer.
04
Product direction
Move from a one-off dashboard into a reusable analytics lab with timestamped predictions.
Final verdict
Strong enough to continue. Not clean enough to automate.
WorldCup Edge should move forward as a prediction and reporting product. The next milestone is timestamped forecasts, quote history and model-vs-market evidence before events happen.
Key learnings
Prediction worked better than betting automation
The model showed useful forecasting power, while the value-betting rules were too strict to trigger official plays.
Knockout needs its own engine
Pass-through probability, extra-time risk and penalties risk were more useful than 90-minute 1X2 alone.
The product opportunity is reporting
The strongest asset is a transparent prediction ledger with model-vs-reality reporting, not hype-driven tips.
Model failure analysis
The misses were not random.
Most failures clustered around penalties, extra time and elimination-match volatility.
What I would build next
Turn the model into a public analytics lab.
The next product should not ask people to trust a black box. It should show the forecast before the event, preserve every version and report the outcome after the final whistle.
Prediction ledger
Every forecast should be timestamped, locked and replayable after the match.
Quote history
Track market movement so the model can explain whether value improved or disappeared.
Data pipeline
Separate team strength, injuries, venue, form and market inputs into auditable modules.
Season mode
Apply the same evidence loop to leagues, cups and repeatable betting markets.