Why We Publish Performance Data
The UK horse racing tipster industry has a credibility problem. Selective reporting of results — publishing only winning weeks, cherry-picking starting dates, using misleading strike rate calculations — is common enough that most experienced bettors treat tipster performance claims with deep scepticism. This scepticism is entirely rational: if you can't independently verify a tipster's record, claimed performance figures mean nothing.
RaceEdge X takes a different approach. All AI pick results are tracked automatically, updated in real time, and displayed in full on the Performance Dashboard — accessible to all members, not filtered for good periods, not curated to show only the best months. The figures below represent the actual performance record of the AI selection system, warts and all.
This transparency serves a purpose beyond credibility. Understanding how the AI performs — which race types generate the best returns, which courses are most reliable, how performance varies by going and season — allows you to make better decisions about when and how to act on AI selections. A blanket "follow all picks" approach will always underperform a selective, informed approach based on understanding the system's strengths.
Overall Performance Metrics
The Performance Dashboard shows rolling statistics across 30, 60, and 90-day windows, and seasonally from the start of the current flat or jump season. Key metrics displayed:
- Win Rate — percentage of AI top picks (highest-confidence selection per race) finishing 1st. The industry benchmark for a profitable tipster is typically 14–18% win rate across a broad selection of races. RaceEdge X's current-season win rate across all picks is in the 18–22% range — above the benchmark but consistent with the confidence levels assigned.
- Place Rate — percentage of picks finishing in the top 2, 3, or 4 (depending on race). Current season: approximately 45–52%. This is the most relevant figure for each-way bettors.
- ROI (Return on Investment) — profit or loss per £1 staked to win at advised prices, expressed as a percentage. A positive ROI means the system generates profit over the sample. Current ROI varies by race type (see below).
- Strike Rate Trend — whether win rate is improving, stable, or declining over recent weeks. A declining trend that persists beyond normal variance suggests reviewing which selections to prioritise.
Performance by Race Type
The AI's performance is not uniform across all race types. Understanding the variation is essential for deploying your bank effectively.
Non-handicap flat races (Class 1–3) consistently produce the AI's strongest win rate — typically 24–28% in this category. These races have smaller fields (often 6–12 runners), more established form, and less randomness than large-field handicaps. The AI's probability estimates are most reliable here because the data is richest and the outcomes are most predictable from historical patterns.
Novice and maiden races are inherently less predictable — horses with limited or no career data are harder to model than those with extensive form records. Win rates in maiden races are lower (typically 15–18%), but so is the market's accuracy. The AI identifies certain maiden runners — well-bred, trainer-backed horses with strong trial form — where the information advantage is largest.
Large-field handicaps (16+ runners) produce lower win rates (typically 10–14%) by the nature of the format — with 20 runners, a genuinely 20% chance of winning produces a 20% win rate only over large samples. However, place rates in large-field handicaps remain high (45–55%), and each-way ROI on these selections has been consistently positive. These are primarily each-way propositions, not win bets.
National Hunt championship races (Grade 1 hurdles and chases at Cheltenham, Sandown, Aintree) produce competitive win rates (18–22%) and are the AI's strongest jump racing category. The Festival in particular shows strong historical performance — course experience and going suitability are weighted heavily by the model, and the model's confidence ratings have correlated well with actual outcomes over multiple Festivals.
Performance by Course
The AI performs most consistently at tracks with reliable, transparent form. The top-performing courses for AI pick accuracy this season:
- Newmarket — The fairest flat track in Britain produces the most reliable AI performance. Without track-specific quirks to distort results, the model's data-based analysis translates directly into outcomes. Win rate here is consistently above overall average.
- York — Similarly wide and fair. Form translates well, market prices are accurate, and the AI's probability estimates have shown strong calibration at York over multiple seasons.
- Haydock — Strong AI performance, particularly in sprint races where draw bias (significant at Haydock on soft going) is a factor the model has learned to weight appropriately.
- Cheltenham — The AI's jump racing stronghold. Course experience is weighted heavily, and the model's performance at the Festival in particular has been consistently above average.
Courses where AI performance is more variable:
- Chester and Epsom — Both highly specialist tracks where draw and course suitability factors can override data-based form analysis. The model applies confidence discounts here, reflected in lower confidence ratings for these course picks.
- Windsor and Brighton — Smaller meetings with less historical data density. Performance is adequate but less consistently reliable than at major tracks.
How to Use the Performance Dashboard
The Performance Dashboard is designed to help you make informed decisions about which AI picks to act on, not simply confirm that the system works. Here's how to get the most from it:
Check the 30-day trend. If the recent 30-day win rate is significantly below the 90-day average, the system may be in a cold patch. This is normal variance, not a reason to abandon the approach — but it may be a reason to reduce stake sizes temporarily until the trend normalises. Conversely, a strong 30-day patch is not a reason to increase stakes aggressively.
Filter by race type. If you primarily bet flat races, filter the dashboard to flat race performance only. The overall figures include jump racing performance which may not be relevant to your betting. Similarly, if you focus on large-field handicaps for each-way value, filter to see the place rate and each-way ROI for those specific selections.
Compare confidence tiers. The dashboard shows performance broken down by confidence level (high, medium, standard). Historically, high-confidence picks outperform standard-confidence picks in terms of win rate and ROI. If you're managing a limited betting budget, concentrating on high-confidence picks only will typically produce better returns per bet than spreading across all confidence levels.
Track your own record alongside the AI's. The most valuable use of the Performance Dashboard is comparing your actual results against the AI's advisory performance. If you're underperforming the AI's returns, you may be taking worse-than-advised prices, deviating from recommended staking, or selectively following picks in a way that removes the statistical benefit. If you're outperforming, identify what you're doing differently — it may be a genuine refinement worth systematising.
Continuous Improvement
The RaceEdge X model is retrained periodically as new race data accumulates, typically at the end of each racing season and at the mid-season point. Each retraining incorporates the full history of recent results, allowing the model to adapt to shifting patterns: changes in how jockeys are booked, new trainers coming to prominence, courses altering their going maintenance, or shifts in which breeding lines are producing the most competitive horses at particular distances.
The most recent model update date is always shown in the Performance Dashboard. When a model update occurs, a brief description of what changed in the retraining is provided — which factors were reweighted, whether confidence thresholds were adjusted, and what the validation set performance showed before and after the update. This transparency is part of our commitment to treating members as partners in understanding the system, not passive recipients of unexplained tips.