Football Data Analytics Platform

AI Ball

Your Data Treasure

Every match quantifiable, verifiable, explainable

AI Ball AnalystDemo chat
For Man City vs Arsenal, what should I look at in the data?

Looking up this match…

Tool callfind_match("Man City")

Tool callget_match_analysis(match_id)

I pulled three things: the recent meetings between these two sides, each team's latest results, and the current injury and suspension list. Every line says which match it came from, so you can open them one by one and check.

Ask about a match, or have me explain any number…
matches
20,241
leagues
96
prediction records
121,428
1X2 accuracy Historical backtest · 2022–2026
60%
Free today

Today's Free Match Analysis

3 of today's matches are open to everyone. Sign up free to see the rest.

AI Analyst

More than a dashboard: an AI football analyst you can talk to

Built on the platform's data set (head-to-head, recent form, injuries & suspensions, fixtures and results), our football-data Agent explains what the numbers mean and where each one comes from.

AI Ball AnalystDemo chat
For Man City vs Arsenal, what should I look at in the data?

Looking up this match…

Tool callfind_match("Man City")

Tool callget_match_analysis(match_id)

I pulled three things: the recent meetings between these two sides, each team's latest results, and the current injury and suspension list. Every line says which match it came from, so you can open them one by one and check.

Ask about a match, or have me explain any number…

Full-dimension deep analysis

One sentence triggers a complete analysis of any match using 6 specialist tools: match search, match lookup, full analysis, six-mode comparison, evaluation stats and date range.

Hidden signal discovery

Cross-checks 50+ fields across 8 categories (head-to-head, upcoming fixtures, injury lists, recent form and more) to surface cross-signals ordinary dashboards can't show.

Explainable, confidence-rated advice

Every answer shows how sure the model is and what data it is based on. The risk score opens up into the factors underneath it.

Try asking

Capability tiers

TierCapabilitiesAccess
FreeBasic single-match Q&A, today's match overviewAfter sign-up
Pro (subscription)Multi-dimensional deep dives, line movement analysis, six-mode comparison, historical backtest queriesPaid users
Foundation6-tool chain · 14-module architecture · 3 safety layers · end-to-end auditable Trace
AI Football AnalystAsk in plain language · Every tool call visible · Six mode perspectives · Review anytimeStart chatting
Verifiable accuracy

Backtest report card: 120K samples

Historical backtest · 2022–2026

January 2022 to September 2026: 20,241 matches and 121,428 prediction records, each checked against the actual result.

60%1X2 accuracyRandom baseline 33.3%, +26.7pp uplift
49.8%Top-3 score hit rateActual score is among the three most likely scores
1.64ppCalibration error MAECEAll 13 confidence bands within ±5pp
0.578Brier scoreBaseline 0.667, lower is better
Confidence calibration: 70% confident means 70% rightR² ≈ 0.99 · High-confidence band (80%+) actual accuracy 86.8%
Cross-season stabilityOnly ±3pp variation over five years, no degradation
Higher risk score, lower accuracyBy risk-score band, accuracy falls from 90.2% to 28.9%
How it works

Six-path fusion + 9-factor risk control

First gather all the data, then cross-validate it through six independent paths, and finally use a risk score to tell you how reliable the conclusion is.

Head-to-headRecent formInjuries & suspensionsUpcoming fixtures
01

Inputs: 50+ fields in 8 categories

Every match analysis starts from this data, covering the teams, their history and the fixture calendar.

02

Six independent analysis paths

Six independent analysis paths each produce their own win/draw/loss probabilities and confidence level, which are then fused using mode-specific weights. Switching mode changes which path carries the most weight.

BalancedMarket-ledData-DrivenChange TrackingConservativeValue-focused
Confidence penaltyLeague adjustmentInjuries
03

9-factor risk score

Factors such as the model's own confidence, league characteristics and injury load are weighted into a single 0–1 risk score, with every reason spelled out.

90.2%28.9%61pp gap between low risk and highest risk

Not just a wrapper

Full Agent architecture · 14 modules · 3 safety layers

From question to reply, every step is handled by a dedicated module, with safety gates guarding the critical points.

  • End-to-end auditable Trace
  • Step cap MAX_STEPS=5 prevents runaway loops
  • Every answer carries its tool-call log, checkable step by step
  1. 1User question
  2. 2Red-line filterSafety gate
  3. 3Intent recognition & planning
  4. 4Tool selection & calls
  5. 5Content policy trimmingSafety gate
  6. 6Governance layerSafety gate
  7. 7Structured reply
agent_loopMain execution loop
governanceContent policy & red-line filter
memorySession & preference memory
plannerIntent recognition & planning
runtimeTool runtime
narrativeStructured narrative
Coverage

Leagues & fixtures covered

96
Leagues covered
5 years
Data span
99.8%
Completion rate
Premier LeagueLa LigaSerie ABundesligaLigue 1UEFA Champions LeagueUEFA Europa LeagueSaudi Pro LeagueJ1 LeagueK League 1MLSCopa LibertadoresEredivisiePrimeira LigaChampionshipScottish PremiershipAllsvenskanArgentine Primera División
Plans

Choose the plan that fits you

Cancel anytime · Full features on every plan · Public track record

Free

US$0

Ready on sign-up

  • 3 free match analyses daily
  • Basic AI chat Q&A
Sign up free
Most popular

Pro

  • Analysis for all matches
  • In-depth AI Analyst chat
  • Line movement analysis
  • Historical backtest queries
View membership plans

Team

Contact sales

For teams & businesses

  • API access
  • Batch analysis
  • Dedicated support
Contact us
FAQ

FAQ

How is this different from a stats site?

The AI Analyst understands your question, calls tools such as match search, full analysis and six-mode comparison on the spot to check the data, and shows every call and piece of evidence, so you can ask where any number came from.

How was the 60% accuracy verified?

By backtesting 20,241 matches from January 2022 to September 2026, 121,428 prediction records in total, comparing each pre-match prediction with the actual result.

What does the system mean by “70% confidence”?

It means that, historically, matches the system rated at around 70% confidence were actually right about 70% of the time. In backtesting, all 13 confidence bands deviated by no more than ±5pp, with an average calibration error of 1.64pp.

Does the system warn about high-risk matches?

Yes. Every match has a 0–1 risk score and risk level, with each reason explained. Very high-risk matches are clearly flagged to remind you to treat them with caution.

Where does the data come from, and how often is it updated?

Match schedules, results, form, head-to-head and injuries, updated daily.

Your data companion

Open AI Ball now and start with today's 3 free analyses.

Sign up free