Richmond Startup Predicts Derby Exacta. But Can It Deliver A Triple Crown?

[via PRNewswire] AI Certain Inc., a Vermont-based tech startup, correctly predicted this year's winning exacta pick at the Kentucky Derby using its new, patent-pending AI technology. The AI narrowed the field of 20 horses down to just two - Justify and Good Magic - and its deep learning neural network system predicted how close the race would be with win-odds separated by only one percent.

AI Certain Inc. was founded in January by father-son duo Tim and Nate Kenney. They launched AI Horse Racing — the company's first venture into sports — in April of this year.

"We were really excited as we watched and saw that the AI had correctly predicted the exacta for the race," Nate said.

The AI's top exacta prediction paid out at $69.60 on a $2 bet.

AI Horse Racing's deep learning neural networks use hundreds of data elements about each horse, jockey and trainer, along with specific information about the race, to predict the likely finishing position of every horse. The system produces a forecast called AI Fair Odds™ that includes odds for first, second, third, fourth, place, show and top exacta and trifecta picks.

The win odds are calibrated to within a 0.5 percent accuracy by analyzing thousands of past races. The odds for major tracks and most small venues in North America are posted every day for free at The company also uses the technology to create AI narrated videos which are posted on their YouTube channel.

"We train the neural networks with tens of thousands of real races. The neural networks learn by seeing real-world results from past events," Tim said. "Training involves quadrillions of calculations for the networks to learn how to predict the races."

"It really learns in a very similar way to how someone who has become an expert in horse racing does, just at a much larger scale," Nate added. "It looks at the same types of data: past performances, workout times and distances, equipment, records on different surfaces, records of trainers and jockeys, times since last workouts and races, how each horse ran in different types of races and much more. What's interesting is that we found that our accuracy is better if we don't allow the AI to see other people's picks and predictions. We let the AI do its job of learning the most important factors from raw race data."

Horses and racing are nothing new to the Kenney family. Nate's sister rode horses and he fondly remembers traveling to races at Saratoga as a kid.

"I remember my sister and I were just enthralled with the spectacle of it all. That's why we chose horse racing as the first sport to focus our efforts on," Nate said. "The reality of pari-mutuel wagering in horse racing is that it is harder than ever for the average handicapper to compete with complex algorithms. We're hoping that our AI can help race fans even the playing field."

The company aims to give players an accurate forecast on the race, allowing them to make smart choices at whatever risk factor they are comfortable with.

"No system can call every race correctly every time," Nate said. "There will always be randomness involved when it comes to horses, so the goal is to reduce that uncertainty. We don't tell people how to bet, we just give them the probabilities to do it better. Wagering involves a mix of prediction, final odds and risk management. Some players are more comfortable with a lower-risk 60 percent place or show prediction with a low payout, while some want the thrill of a long-shot overlay that becomes available minutes before the bell. The goal is to provide information to both types of players."

"My tech friends want to know how it works," Tim said. "A 20-horse race can have a finishing order of more than 2.4 x 1018combinations. That's quintillions of possibilities. I'm always careful how I answer because we are protecting our intellectual property.  We've put two years of research into this AI with statistical modeling. The technology we developed has applications to other sporting areas as well as financial markets, healthcare and business. The system can look at N-dimensional data from very diverse sources - even ones that may be missing data. The training of the neural networks takes some serious computational power and time. What's astonishing is that the predictions for a new scenario are ready in seconds. Our predictions are generally more accurate than other methods because they include such a wide field of information."

"We knew the technology had many applications," said Tim. "While we do find profitable wagers, like the one we placed on the Kentucky Derby, we realize there is more to be gained in becoming a hub for sports prediction. We are focused on rolling this idea into many other market opportunities."

Media Contact:

Nate Kenney  Phone: (802) 318-7227 Email:


Recent Posts

See All