Can AI help you become a successful punter?
After three years of using it almost daily, I see enormous opportunities, but also traps.
My thoughts on where the real value lies, why your thinking still matters, and how to focus on the right opportunities. https://x.com/i/article/2115604926915379200
AI in horse racing: your thinking is still the edge
The number of enquiries I receive from people wanting racing data to experiment with AI has exploded. It’s changing how people work across almost every industry, and racing will be no different.
Huge opportunities lie ahead for punters and form analysts. But I think a significant gap exists between what many people hope AI will do for their betting and where its lasting practical value lies.
I started using AI about three years ago and have used it almost daily since, working on one racing project or another. That work has been backed by access to our full database, ratings and other intellectual property. I’ve also spent decades working with racing data and developing models, both personally and professionally with other groups.
My view is that if you expect AI to “auto-solve” racing and produce something you can run automatically and make a meaningful profit, you’re almost certain to end up frustrated and disappointed.
That doesn’t make those projects pointless. You can learn a lot about racing, data, technology, and your own assumptions. You might develop something useful along the way.
But the biggest benefit of AI is its ability to help you develop ideas, test your thinking, and build knowledge far faster than before. What you do with that knowledge is where the real opportunity lies.
The biggest change is the speed of turning ideas into something useful
For me, AI's most immediate benefit has been reducing development time.
I can have an idea, specify what I need, build and test a working version, and put it into a workflow faster than it takes me to write a detailed email explaining the job to my programmer.
That changes what’s practical. A research question that once required hours of work, or a minor project that never made it far enough up the priority list, can now be explored in minutes, or at most a few hours. Larger projects that previously took weeks can now be completed in a day or two.
You can build metrics, test form angles, automate repetitive work and create tools that make your analysis more efficient, consistent and effective.
But faster development doesn’t automatically mean better ideas or a guaranteed edge.
The quality of what you get from AI still depends heavily on the quality of your thinking, the questions you ask and the data you provide. People who can identify worthwhile questions and interpret the answers intelligently are best positioned to benefit.
A profitable backtest doesn’t mean you’ve solved racing
For an individual starting with standard form data, I would put the chances of building a sustainable, fully automatic profit model at worse than 1000/1. That reflects how difficult I believe the task is with standard data.
If you have your own intellectual property, such as ratings or other metrics that add insight beyond standard form data, your chances improve. The more insightful that intelligence is, in terms of being less widely used in the market, the stronger your position.
The further challenge is that you’re competing against highly sophisticated groups already extracting edges, with far more money, expertise and resources than the average individual or small group of punters can commit.
Even if you develop a profitable model, the margins are likely to be small. Turning that advantage into meaningful profit depends heavily on how well you execute your betting and the prices you secure. A larger edge, more realistically achieved through a non-automated approach, gives you more room to move. A high-volume model operating on thin margins needs a very sharp betting operation.
For some individuals and groups, that ability to consistently secure better prices is what separates a useful model from a profitable betting business.
Even if you build a model with a historical edge, you still have to operate with it through the ups and downs of betting, including the losing runs that test your confidence when real money is on the line.
Losing runs in racing can be far longer and more severe than people expect. When real money is disappearing, confidence in a model that looked terrific on historical data can vanish just as quickly.
If you don’t understand why it selects particular horses, where its edge comes from, or what results are reasonably possible, how do you distinguish an ordinary losing run from evidence that it doesn’t work?
That uncertainty creates a familiar cycle.
You build a model. It looks good on historical data. You start betting. A losing run arrives, doubt takes over, and you investigate, adjust or rebuild it.
The revised version looks good on the updated history. Confidence returns. Eventually another losing run arrives, and the process starts again.
You can spend years on that merry-go-round without establishing whether the original approach had merit.
Understanding why you’re backing particular horses and where your edge comes from won’t prevent losing runs. But it gives you a firmer basis for judging whether those losses are part of the natural variation in betting or a reason to rethink your approach. Without that understanding, it’s much harder to stay confident when you're losing real money.
Use AI to build your understanding of racing
One of AI's most valuable uses is exploring the questions that come up while you're doing form.
We all carry theories about what helps or hurts a horse’s chances. Some come from experience. Others come from something we’ve heard repeatedly or a few memorable results.
AI makes it much easier to test those beliefs.
You can investigate form angles, track patterns, barriers, trainers, jockeys, distance changes, race pace, or other ideas that you believe may offer an edge. More importantly, you can examine how the betting market treats those factors.
A factor can matter to a horse’s performance without offering any betting advantage. If the market already accounts for it, knowing about it may not help you find value.
As an example, now that my AI setup understands my database, I was able to ask:
Show me every three-year-old that has run in the Cox Plate over the past 10 years, their lead-up ratings, relevant preparation metrics and their rating in the Cox Plate.
That research took about five minutes and highlighted how few had matched or exceeded their previous peak in the race. Previously, the same exercise would have taken me about an hour. With AI, it was roughly 12 times faster.
That finding doesn’t mean a three-year-old can’t win or run well. It gives me useful context for assessing what a particular horse needs to produce and how realistic that expectation is.
I still assess the individual horse and race, but with better information and a firmer understanding of the historical context.
Repeat that process across hundreds of worthwhile questions, and you can build knowledge far more quickly than before. You can challenge theories, refine principles and deepen your understanding of individual trainers, jockeys, tracks and race circumstances.
That knowledge stays with you. Its value extends well beyond the next selection a model produces.
When everyone has the tools, where is the edge?
People have been modelling racing for decades. AI has now dramatically lowered the barriers to doing it.
Previously, you needed data, programming skills and considerable expertise in quantitative analysis, either yourself or through others. AI now makes much of that work accessible to a far wider group.
That’s exciting, but it also creates a problem for anyone who thinks access to these tools will provide a lasting advantage.
As more people use similar data and tools to identify the same market inefficiencies and back the same horses, the edge will erode. I’ve seen this happen many times over more than two decades of betting and market analysis.
Time and sectional ratings are one example. Years ago, I was fortunate to have access to data that very few people had. I developed my own figures, and they gave me a real advantage in the horses they identified.
As businesses entered the racing industry and made that information more widely available, the market became better at recognising those performances. The figures remained useful, but the best betting opportunities became fewer and harder to find.
I expect the same process to apply to many AI-assisted approaches.
Having more data and better tools is valuable. Access to the same data and tools as everyone else is not, by itself, a sustainable edge.
Think three, four or five years ahead, with far more people using AI to pursue the same opportunities. How likely is it that everyone else's AI use will overlook the edges your model finds?
And do you want to stake your betting future on a black box when you don’t understand why it selects particular horses or where its advantage comes from?
For all but the most sophisticated and well-resourced groups, I don’t believe that’s a sustainable approach. Even those groups have to keep developing their intelligence and improving how they bet. Finding an edge once doesn’t mean you’ll still have it in five years.
The value of “special insight”
I’ve always believed in balancing the art and science of racing.
Data has been central to developing my knowledge, ratings and other intellectual property. At the same time, every race is unique, and every horse is an individual. General patterns provide context, but they don’t explain everything about the race in front of you.
As more of the market relies on data and AI-assisted models, I believe there will be increasing opportunities for punters who understand individual horses and race circumstances in ways those models don’t adequately capture.
Combine that insight with the knowledge you've built about racing and form by using data and AI, and you’ll be well positioned to find an edge in the betting markets of the future.
I call that extra understanding, beyond what the data alone reveals, “the special sauce.”
The question I ask myself is: What do I understand about this race or these horses that might not be reflected in the market? Or, more simply: What’s the special sauce I’m adding here?
That might mean asking:
- Is there something in this horse’s rating profile, based on what I’ve learned about how those profiles translate to future performance (using data and AI), that the market hasn’t reflected in its pricing?
- Do I understand a trainer’s patterns, how the jockey might ride the horse in this particular scenario, or how the track is playing today in a way the market doesn't seem to recognise yet?
- Have I picked up something in a video replay that, combined with my understanding of this particular horse, gives me reason to expect a better or worse performance than the market anticipates?
Historically, it's this type of thinking that gave punters an edge because it wasn't a data-rich game. Very few people had access to the tools, systems, and processes needed to maintain good data. Those principles will always be valuable, but data-driven punting has taken over the market, and people have proven it offers an edge. It's not the only edge, but it's been exploitable for years.
I believe that, with the explosion of AI and what it will mean for betting market behaviour over the next decade, we'll see a reversal.
I think the ability to return to more fundamental thinking about the individual horse and race, backed by deeper knowledge developed through data and AI, will become increasingly valuable in finding sustainable betting edges.
In other words, if more money starts to follow similar data- and AI-driven signals, people who can spot when those signals are misleading or less relevant to a particular race will have more opportunities to profit.
Wagering providers are increasingly using AI to improve their markets, and will do so more in the future. That will make some opportunities harder to find, while reinforcing the value of insight that goes beyond the obvious data.
My advice for getting the most out of AI as a punter
The goal is to develop insights that others miss. As more people use AI, the quality of your questions and how you apply the answers will become increasingly important.
Here’s where I would focus:
- Ask questions that deepen your understanding. Use data to test your assumptions about racing and form analysis. Find out which factors matter, under what circumstances, and whether the market already accounts for them. Can you ask a more insightful question than others working with the same data?
- Build your own intelligence. Develop metrics that reveal something beyond standard form data. You don’t always need an entirely new idea. A thoughtful variation on a familiar measure or technique might uncover something the market overlooks.
- Turn your research into a practical assessment process. Build tools that help you identify the horses, races, and betting scenarios most likely to offer an edge. The faster you can find those starting points, the more time you have to assess the individual circumstances and apply your judgement.
- Shorten the time between an idea and an answer. Use AI to speed up research, handle calculations, and automate repetitive work. Removing those barriers allows you to explore more worthwhile ideas and become a more effective punter.
- Understand the bets you make. Your edge may come from AI-driven data and analysis, but you still need to understand why a particular bet represents value. That understanding gives you a firmer basis for maintaining confidence and discipline through losing runs, and for recognising when your approach needs to change.
If building automated models interests you, pursue it. Those projects can be rewarding, and the lessons can improve your punting even if the model never becomes something you bet with. Just be realistic about the chances of developing a sustainable, fully automated approach. For most individuals, I believe those chances are slim to none.
AI offers enormous opportunities for punters and form analysts to learn faster, work more efficiently and turn good ideas into practical tools and insights.
But it also makes it easier to convince yourself that a sophisticated approach and profitable historical results mean you’ve solved the game. Neither guarantees a sustainable betting edge. That false confidence can cost you real money.
AI makes it easier to find answers. Your edge will still depend on asking better questions, understanding what the answers mean, and applying that knowledge with judgement.