NZBIO.ORG.NZ //
New Zealand has always punched above its weight in sport. An All Blacks jersey carries weight far beyond a population of five million. But when it comes to the technology behind sport — the data science, the machine learning models, the predictive analytics — NZ has largely been a consumer, not a creator.
That’s starting to change. And one of the most unexpected frontiers is horse racing, with a new AI innovation, Winning Post, leading the way.
For as long as New Zealanders have bet on horses, the process of analysing a race has stayed remarkably consistent. You grab the form guide. You look at recent results, track conditions, barrier draws, and jockey bookings. You remember how a horse ran last time at this track. You factor in the weather. You make a call.
This is pattern recognition — and the human brain is genuinely good at it. The problem is scale. A single Saturday card across three or four NZ meetings might feature 80 to 100 horses, each with its own history of 10, 20, or 50 starts. Multiply that by thousands of variables per runner — sectional times, track condition splits, jockey and trainer history at specific venues, barrier performance at specific distances — and the information load becomes insurmountable.
Even the most dedicated form student is working with a fraction of the available data.
Machine learning flips the problem on its head. Instead of a human trying to absorb more information, a model is trained on tens of thousands of historical race outcomes and left to discover the patterns on its own.
This is called supervised learning. You feed the model a massive dataset of past NZ races — every runner, every barrier, every jockey, every track condition, every finishing position — and let it find the combinations of factors that correlate with winning. The model doesn’t have preconceptions. It doesn’t assume that barrier 1 is always good, or that the favourite is always over bet. It simply analyses the data and learns what matters.
The results are often surprising. Factors that human handicappers consider important — recent finishing position, for example — can be less predictive than things like sectional finishing speed relative to the field, or how a horse has performed on similar going at a similar distance. The model surfaces patterns that are real but invisible to the naked eye.
New Zealand thoroughbred racing is a fascinating dataset for machine learning. It’s large enough to train meaningful models — thousands of races over the past decades — but contained enough that a focused model can genuinely master it.
Compare that to a jurisdiction like Australia or the UK, where the volume of racing is an order of magnitude larger. A single model covering all Australian thoroughbreds would need to account for vastly different track types, prize money levels, training regimes, and racing patterns across Victoria, New South Wales, Queensland, and beyond. The complexity grows exponentially.
NZ racing, by contrast, has a consistent structure. A small number of trainers dominate the trainer standings. The jockey pool is tight and stable. Track conditions follow predictable seasonal patterns. And the TAB betting market, while efficient, is not so deep that value opportunities are immediately arbitraged away.
This makes NZ an ideal sandbox for AI-powered racing analysis — and a proving ground for techniques that could scale to larger markets later.
Building a prediction engine for NZ thoroughbreds involves several layers of technology:
Data collection. Raw racing data is pulled from multiple sources — TAB NZ for race cards, results, and fixed-odds markets, along with sectional times, profiles, and historical results; stewards’ reports for gear changes and track conditions. This data is collected daily and stored in a structured database.
Feature engineering. Raw data is transformed into predictive features. A finishing time becomes a relative sectional speed. A barrier draw becomes a track-and-distance-specific win rate. A jockey’s booking becomes a strike rate at this specific venue over the past 50 rides. A well-built feature set is what separates a mediocre model from a strong one.
Model training. A machine learning model — using an algorithm that consistently wins machine learning competitions — is trained on the historical dataset. The model uses ranking-based objectives to learn not just whether a horse will win, but how it should be ordered relative to the other runners in the race.
Calibration and probability estimation. Raw model outputs are calibrated into genuine probabilities — a horse marked at 25% should win roughly one in four races at that mark. This is critical for the next step.
Expected value calculation. Model probabilities are compared against market odds to identify value bets. This is where the punter’s edge lives.
This isn’t just about picking winners. It’s about New Zealand establishing itself as a player in the global sports analytics industry. Racing analytics is a multi-billion-dollar market worldwide, and the techniques that work in NZ — small-field prediction, condition-specific modelling, and combined human-AI workflows — are directly applicable to racing jurisdictions in Asia, Europe, and North America.
The same models that predict Saturday’s card at Riccarton could, with retraining, predict Group 1 races at Sha Tin or Royal Ascot.
One of the first NZ-built AI racing platforms to reach the public is Winning Post, a prediction service trained exclusively on New Zealand gallop data. It generates win and place probabilities for every NZ thoroughbred meeting, surfaces value opportunities by comparing its probabilities against TAB fixed odds and updates its model after every race day.
Winning Post is currently in beta mode and will be available via free trial — a chance for NZ punters and racing fans to see what machine learning looks like when it’s built for their own backyard.
The form guide isn’t going away. But the future of racing analysis belongs to the data.
(This article is for informational purposes only. Horse race betting involves financial risk. Please gamble responsibly.)
NZBIO Website: Empowering New Zealand Businesses The NZBIO website serves as a valuable platform that showcases success stories of New Zealand businesses across diverse industries, offering inspiration and promoting entrepreneurship. The website provides essential resources and information to help aspiring entrepreneurs navigate the intricacies of establishing and growing a business. Moreover, it facilitates connections with potential partners, investors, and collaborators, thereby contributing to the growth and success of New Zealand’s business landscape. To promote your business with NZBio.org.nz contact us now.