Breaking

Sozu's Live Football AI Assigned Just 10.7% Average Goal Probability Across 15 Tracked Strikes

July 15, 2026 at 20:15 EDT

AI Agents
Foundation Models

A live football-prediction system built by Sozu assigned an average pre-goal probability of just 10.7% across 15 tracked scoring moments, illustrating how cautious real-time AI forecasting can be even seconds before the ball hits the net.

Live Football AI · Sozu Prediction System

The model says the goal probably won't happen — right up until it does

Across 15 tracked scoring moments, a real-time football-prediction system assigned an average pre-goal probability of just 10.7% — a sign of cautious, realistic calibration rather than inflated hype.

10.7%

Average probability assigned just before each goal

Scoring moments tracked by the system

18-19%

Peak readings moments before a goal

Pre-goal probability · France v Spain

Even the goals that actually arrive were rated unlikely at the moment they were scored.

19.5%

Spain 31'

18.3%

France 60'

10.7%

15-goal avg

Each block ≈ 5 percentage points of assigned probability

Read as a strength

Modest numbers reflect realistic calibration — the model does not inflate the likelihood of rare scoring events, and observers point to strong recent hit rates.

Open questions

Architecture, training data, update frequency and platform are undocumented. No confirmed benchmark against xG, Elo or XGBoost methods exists yet.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.