← Back to dashboard
Betfair Data Scientists
Tipster dossier — settled-bet performance through 2026-07-14 20:15 UTC
Verdict
STRUCTURALLY LOSING
Small sample (4 bets) — anything below ~100 settled bets means CIs are wide enough to be near-useless. Calibration is negative: actual win rate 25.0% trails market-implied 31.4% by -6.4pp — picks have negative skill relative to fair odds. Bayesian-shrunk ROI (regressed toward -13% population baseline by sample size) is -13.4% — a more honest expectation than the raw -19.0%.
Bio & context
Anonymous internal quantitative team at Betfair Australia that produces automated prediction models published on the Betfair Hub. These are not traditional tipsters but machine learning models covering thoroughbred, harness, and greyhound racing across Australia. The models generate rated prices and value percentages rather than explicit win selections, designed to help customers identify market inefficiencies.
Australia (Betfair Australia internal team)Internal quantitative modelling team at Betfair Australia
⚠ Conflict / insider flags
- This is not a traditional tipster but an algorithmic model—no personal bias or ownership conflicts
- Model outputs are rated prices/value percentages, not explicit tips—customers must interpret and apply
- Betfair has commercial interest in customer betting activity but as an exchange benefits from accurate pricing
- Educational materials encourage customers to build their own models rather than blindly follow predictions
Methodology: Data-driven machine learning models (stacked ML algorithms) that generate rated prices for value identification. Horse racing model uses Punting Form data including sectionals, past performance, trainer/jockey statistics, track conditions, and Betfair Exchange BSP data. Harness model (partnership with Rise Racing) uses driver/trainer stats, sectionals, margins, and competition strength. Greyhound model uses box number, sectionals, and market prices. Focus is on identifying value by comparing model-rated prices to Exchange market prices rather than picking winners.
Specialty: All three codes (thoroughbred, harness, greyhound) across all Australian states. Thoroughbred coverage extends to UK racing. No specific track specialization—national coverage.
Public footprint
- Betfair Hub website (primary distribution channel for ratings and predictions)
- GitHub repository (betfair-datascientists) with open-source predictive model tutorials and educational content
- The Automation Hub (betfair-datascientists.github.io) - educational resource on model building
- Data modelling workshops held in Melbourne, Brisbane, Gold Coast, and Sydney (e.g., AFL sessions April 2019)
- Email contact: automation@betfair.com.au
- Betfair Quants Discord server for model/automation discussion
Industry connections
- Partnership with Rise Racing (harness racing data provider)
- Uses Punting Form as exclusive data source for thoroughbred racing
Source: anthropic_web_search · last researched 2026-06-10
Since last weekly snapshot: ROI: -106.4% (+87.4% → -19.0%) · Shrunk ROI: -60.7% (+47.3% → -13.4%) · Verdict: ROI MISLEADING → STRUCTURALLY LOSING
Charts
Cumulative P/L (theoretical $1 stakes)
A straight upward line = consistent profitability. Step-jumps = single fortunate bets carrying the curve.
Monthly ROI %
Top 10 P/L outliers
ROI by odds bracket
Calibration — actual vs implied win rate by odds
Blue bars under orange = picks lose to market expectations.
ROI by day of week
At a glance
Total P/L ($1 stakes)
$-0.76
Bayesian-shrunk ROI
-13.4%
Implied Win % (1/odds)
31.4%
What's driving the ROI?
Strip away the top winning bets and see how the picture changes — a stable record degrades gracefully,
a lucky record collapses.
(insufficient data)
Recent trend
(insufficient sample to compare recent vs prior)
Monthly performance
| Month | N | Win % | Avg Odds | P/L | ROI |
|---|
| 2026-05 | 4 | 25.0% | 3.25 | $-0.76 | -19.0% |
By odds bracket
| Odds | N | Win % | P/L | ROI |
|---|
| 2.0-3.0 | 1 | 0.0% | $-1.00 | -100.0% |
| 3.0-5.0 | 3 | 33.3% | +$0.24 | +8.0% |
By pre-race rank
No pre-race rank data yet for this tipster.
Strategy Zone (rank ≤ 3 AND odds ≥ 5)
No rank data yet.
By source
| Source | N | Win % | P/L | ROI |
|---|
| payload | 4 | 25.0% | $-0.76 | -19.0% |
By venue
(no venue data with sufficient samples)
By day of week
| Day | N | Win % | P/L | ROI |
|---|
| Mon | 4 | 25.0% | $-0.76 | -19.0% |
Place market
No place outcomes captured yet for this tipster.
Biggest wins
| Date | Venue | Selection | Odds | P/L |
| 2026-05-25 | mornington | Meisho | 3.24 | +$2.24 |
Biggest losses
| Date | Venue | Selection | Odds | P/L |
| 2026-05-04 | alice springs | Bender McGee | 4.00 | $-1.00 |
| 2026-05-11 | echuca | Zuppa Inglese | 3.12 | $-1.00 |
| 2026-05-11 | corowa | Under Wraps | 2.65 | $-1.00 |
Generated 2026-07-14 20:15 UTC. Stats based on theoretical $1 BACK stakes settled on Betfair. Place data uses captured Betfair Starting Price.
ROI = total P/L divided by total bets — high variance at small samples.