How does AI pick a winner in a horse race?
An AI rating model evaluates every runner against a learned function of features: Official Rating, Racing Post Rating, top-speed figure, trainer 14-day form, days since last run, class, going preferences, draw, and (in the more advanced models) live market drift. Each horse gets a 0-100 score; the top-scoring horse in a race that passes the model's segment filter becomes the pick. The model isn't 'predicting the future' — it's quantifying which horse has the best combination of measurable signals at the off.
What's a realistic AI horse racing strike rate?
Around 18-25% for a model picking one selection per race across UK & Irish racing. Roughly 1 in 4 races. Higher than that on a small sample (50-100 picks) usually means small-sample noise; lower than that consistently means the model isn't beating the market favourite. Strike rate alone is misleading — a 20% strike rate at an average advised price of 6.0 is profitable; a 30% strike rate at 3.0 is breakeven.
Can AI beat professional tipsters?
On consistency, yes — AI models don't get tired, don't hold grudges against horses, don't fall in love with a stable. They publish before the off, every off, every day. On individual race nuance, a top professional with stable contacts can still outperform on specific races. The best modern setup is AI as the screening layer (covers every race, surfaces standouts) plus a human filter on top for the high-stakes picks.
Is AI horse racing profitable long term?
Only if it consistently produces positive Closing Line Value (CLV). CLV is the gap between the price you backed at and the SP — sustained positive CLV is the gold-standard evidence that the model has a real edge the market hasn't yet absorbed. Any AI tipster who can't show positive CLV across hundreds of picks is, on the maths, just expensively wrong. Racing Alpha publishes its CLV ledger live.
Can I see Racing Alpha's actual AI performance?
Yes — every pick from every model is logged before the off and settled publicly. The head-to-head dashboard tracks three picking strategies (v1.0 heuristic, v1.1 ML, Furlong manual) side by side with bankroll trajectories. The CLV page shows whether the advised prices actually beat the SP across the settled sample. No edits, no hiding losing runs, no retroactive selection.