How we analyse sports matches, generate predictions, and ensure every tip is grounded in data — not guesswork. Full transparency on our process, tools, and limitations.
Every prediction on VoxSports is built on one question: does the bookmaker's implied probability diverge from our assessed true probability? When it does — and the divergence is meaningful — that's a value bet worth highlighting.
We do not predict winners. We identify value. A bet can be correctly identified as good value and still lose. Our methodology is designed to find long-term edge, not guarantee short-term results.
All predictions are built on verifiable, publicly available data. We do not use proprietary models or insider information. Our primary sources:
| Data Type | Source | Used For |
|---|---|---|
| Expected Goals (xG) | FBref, Understat | Attacking/defensive quality assessment |
| Team Form | Official league sites, BBC Sport | Recent performance weighting |
| Head-to-Head Records | Soccerway, 11v11 | Historical matchup context |
| Current Odds | Oddschecker, Betfair Exchange | Implied probability calculation |
| Injury & Lineup News | Official club sites, BBC Sport | Probability adjustment |
| NBA Stats | Basketball-Reference, NBA.com | ATS records, rest patterns, player props |
For each match we publish, our process follows these steps:
We start with historical win/draw/loss rates for the relevant competition and home/away context. Premier League home win rate is approximately 46% — that's the anchor before any match-specific data.
We weight last 6 home fixtures for the host and last 6 away fixtures for the visitor. A team with 5 wins in 6 home games shifts probability upward from the base rate.
Expected Goals better predicts future performance than actual goals. A team scoring 1.8 xG/game but only 1.2 actual goals is likely underperforming — we factor this into our probability estimate.
Injury to key players (typically ±3–8% probability shift), squad rotation risk, fixture congestion, European midweek fatigue, and motivational context (relegation battle, title race).
We compare our probability estimate to the best available odds. If our probability exceeds implied probability by 5% or more, we flag it as a value recommendation.
We select the specific market (1X2, Asian Handicap, Over/Under, BTTS) that offers the best value given our analysis. We never default to match winner if another market offers better edge.
Our primary coverage. We analyse Premier League, Champions League, and major European competitions. For high-profile matches, we use full xG-based modelling. For lower-priority fixtures, we use condensed form-based analysis.
We focus on spread (ATS) and totals markets. Key factors: rest patterns (teams on back-to-backs cover ATS at ~47% vs 52% for rested sides), player props based on matchup-specific data, and playoff vs regular season adjustments.
We never fabricate statistics, invent injury news, or guarantee outcomes. If data is insufficient for a confident recommendation, we say so explicitly rather than manufacturing false certainty.
We do not use: insider information, betting exchange trading patterns as a primary signal, social media sentiment analysis, or unverified team news from unofficial sources.
We maintain a public record of all predictions and their outcomes. Accuracy statistics are updated monthly. We do not retroactively remove or alter published predictions. Our track record — including losing runs — is part of our editorial commitment to transparency.
Prediction outcomes are recorded within 24 hours of match completion. Voided matches (abandoned fixtures, postponements) are excluded from accuracy calculations.
No prediction methodology is infallible. Known limitations of our approach:
xG models do not fully account for tactical changes mid-season. Injury news can be unreliable until official lineup confirmation. Market odds move rapidly and our analysis reflects a specific point in time. Historical data is a better predictor in stable competitions than in knockout formats with high variance.
We always recommend treating predictions as one input among many — not as financial advice.