How Goalii Works

Goalii is an AI football analysis app for iOS and Android that turns 40+ live signals into daily match predictions. This page explains the five stages of Goalii's pipeline — from data ingestion to recalibration — so you can understand exactly what powers every prediction the app shows you.

Step 1 — Goalii ingests live football data

Goalii pulls match data, lineups, injuries, weather, referee assignments, and league standings from licensed sports data providers. The data refresh runs every 90 seconds in the run-up to kickoff, so the model has the most recent signals before the whistle.

Step 2 — The model derives 40+ tactical features

For each fixture, Goalii computes expected goals (xG and xGA) trends, recent form weighted by opponent strength, home/away splits, set-piece efficiency, rest days, travel distance, and head-to-head momentum. These features feed the prediction model.

Step 3 — Predictions are generated with confidence scores

Goalii's ensemble model produces probability estimates for Over/Under 1.5 and 2.5 goals, Both Teams to Score (BTTS), and the 1X2 outcome. Each prediction comes with a confidence grade (A/B/C) so users can filter by certainty.

Step 4 — Every analysis is logged and audited

Goalii records each prediction with timestamp, market, and confidence. After the match, the result is matched against the prediction. The rolling 90-day hit-rate is published in-app for full transparency, including misses.

Step 5 — The model recalibrates daily

Last night's results feed back into tomorrow's model. Drift is monitored automatically; if a league or market starts underperforming, Goalii reduces its confidence rather than hide the issue.

What this means for users

Goalii's transparency principle is simple: if you can't audit a prediction system, you can't trust it. Every analysis Goalii has ever published is visible in the app, with the confidence level it was issued at and what actually happened. The model is allowed to be wrong; what it cannot be is hidden.

Where to go next

Last updated