Oops, we did it again.
After our March Madness experiment earlier this year, we couldn’t resist running the same question back for the biggest race in Louisville’s backyard: could AI help us build a smarter Kentucky Derby betting strategy?
Short answer? Yes. Here’s the play-by-play of how we used AI to handicap the 152nd Kentucky Derby.
The Gameplan
The Derby is a unique beast. Twenty horses, 1 1/4 miles, run at a pace that punishes front-runners, with a field that’s never faced each other at this distance. No single factor wins it. You need a model that weighs multiple variables at once, and that’s exactly where AI earns its keep.
We broke it into five steps:
- Build the data foundation. Collect structured performance data on every contender.
- Use AI for pattern recognition. Identify what winning profiles look like across 20+ years of results.
- Simulate the race. Project pace scenarios and identify which running styles benefit.
- Build a weighted scoring system. Score every horse and find value against the odds.
- Apply the human override. Let the data narrow the field, then use real eyes for the final call.
Step 1: Build the Data Foundation
We asked Claude to gather speed figures across the Beyer and Brisnet rating systems, pace figures, trainer and jockey stats at Churchill Downs, pedigree and stamina data, workout sharpness, and post-position history for all 20 horses. Claude built us a nine-tab research workbook.
Then we fed it some additional research PDFs to compare and fill gaps. Three big findings emerged: Chief Wallabee’s speed figures were declining with each start (133, then 123, then 121). Commandment fit a pattern where the last two Derby winners both broke their maiden at Churchill before returning to win the Derby. And Fulleffort’s prep path mirrored Rich Strike’s 2022 upset run almost identically, with Fulleffort actually outperforming Rich Strike in each of those same races.
Step 2: Pattern Recognition
With the data locked in, Claude found patterns from two decades of Derby results. Horses improving into the Derby outperform those who posted a career-best in their final prep (the “bounce” effect). Post positions 1-3 and 17-20 historically underperform. The ideal gap between final prep and Derby day is 21-28 days.
The weather variable was massive this year. A 90% chance of showers meant the track could come up sloppy, and six of the last 21 Derby winners ran on a sloppy track. Five of those six paid 5-1 or higher.
Step 3: Simulate the Race
Claude projected a HOT early pace. Six Speed and Pavlovian were both confirmed frontrunners who would ensure contested speed through the first three-quarters. Neither figured to win. Their role was to set the table for closers.
In horse racing, “fractions” are the split times at each quarter-mile marker. The model projected the first quarter-mile in about 22.5 seconds and the half-mile in about 45.8 seconds. Those are fast. When the early pace is that aggressive, the horses who sprint to the front tend to burn out before the finish line.
That race style historically rewards the horses who sit in the middle of the pack or further back and finish with a strong late kick. The last four Derbys followed that exact pattern, and the model said this year would too.
Step 4: The Weighted Scoring System
This was the step where AI really flexed. We weighted six factors based on historical Derby winner profiles, scored every horse 1-10, and generated composite scores. Then we checked against morning line odds to find overlays: horses whose win probability exceeds what the odds suggest.
| Rank | PP | Horse | ML Odds | Tier | Speed Trend (25%) | Late Pace (20%) | Pedigree (15%) | Trainer/Jockey (15%) | Trip Quality (15%) | Days Off (10%) | WEIGHTED SCORE | Overlay? |
| 1 | 6 | Commandment | 6-1 | A | 9 | 9 | 9 | 9 | 10 | 6 | 8.85 | YES |
| 2 | 18 | Further Ado | 5-1 | A | 8 | 7 | 9 | 10 | 8 | 8 | 8.15 | |
| 3 | 1 | Renegade | 9/2 | A | 7 | 10 | 9 | 8 | 9 | 6 | 8.15 | |
| 4 | 9 | The Puma | 10-1 | B | 7 | 9 | 8 | 9 | 9 | 6 | 8.00 | YES |
| 5 | 12 | Chief Wallabee | 10-1 | B | 5 | 8 | 8 | 10 | 8 | 6 | 7.35 | |
| 6 | 15 | Emerging Market | 20-1 | B | 8 | 7 | 10 | 8 | 7 | 4 | 7.45 | YES |
| 7 | 8 | So Happy | 15-1 | B | 8 | 5 | 5 | 8 | 8 | 8 | 6.95 | |
| 8 | 14 | Potente | 15-1 | B | 7 | 5 | 9 | 8 | 6 | 8 | 6.90 | |
| 9 | 20 | Fulleffort | 20-1 | B | 7 | 7 | 7 | 8 | 7 | 4 | 6.80 | YES |
| 10 | 11 | Incredibolt | 25-1 | B | 6 | 8 | 7 | 5 | 7 | 8 | 6.75 | |
| 11 | 13 | Silent Tactic | 30-1 | C | 5 | 5 | 8 | 4 | 7 | 6 | 5.75 | |
| 12 | 2 | Albus | 40-1 | C | 5 | 4 | 7 | 4 | 6 | 8 | 5.40 | |
| 13 | 19 | Golden Tempo | 30-1 | C | 5 | 5 | 8 | 5 | 5 | 4 | 5.30 | |
| 14 | 10 | Wonder Dean | 25-1 | C | 4 | 4 | 7 | 3 | 5 | 6 | 4.60 | |
| 15 | 16 | Pavlovian | 40-1 | C | 6 | 3 | 6 | 5 | 7 | 4 | 5.15 | |
| 16 | 5 | Right to Party | 50-1 | C | 4 | 5 | 6 | 4 | 5 | 8 | 5.00 | |
| 17 | 7 | Danon Bourbon | 50-1 | C | 5 | 4 | 7 | 2 | 4 | 6 | 4.55 | |
| 18 | 3 | Intrepido | 50-1 | C | 4 | 3 | 5 | 2 | 3 | 8 | 3.95 | |
| 19 | 17 | Six Speed | 40-1 | C | 4 | 1 | 5 | 3 | 5 | 6 | 3.70 | |
| 20 | 4 | Litmus Test | 30-1 | C | 3 | 3 | 8 | 6 | 2 | 1 | 3.75 |
Each factor scored 1-10 based on 20+ years of Derby winner profiles. Tier A = Win contenders. Tier B = Live longshots/exotic keys. Tier C = Fades. “Overlay” = Horses whose score suggests higher win probability than their odds imply.
So Who Does AI Like?
Win/Place/Show (Fast Track): Commandment (#6, 6-1) to win. The model’s top-ranked horse at 8.85 and the best overlay in the field. Four-win streak, two triple-digit Beyers (nobody else has that), and a Derby-winner pedigree. Further Ado (#18, 5-1) to place with field-best speed figures. The Puma (#9, 10-1) to show, the second-best overlay, whose connections won the 2023 Derby with Mage using the same stalking style.
Exacta: Box Commandment, Further Ado, and Renegade. For value, throw The Puma in.
Trifecta: Key Commandment on top, with Renegade, Further Ado, The Puma, and Chief Wallabee in second, plus So Happy and Emerging Market in third.
Rain pivot: If the track comes up sloppy, swap your win bet to Emerging Market (#15, 20-1). Best distance pedigree in the field, and PP15 has produced three Derby winners since 2013. Fulleffort (#20, 20-1) becomes a mandatory exotic inclusion in the mud.
The fade: Don’t put morning-line favorite Renegade (#1, 9/2) on top. Post position 1 hasn’t won the Derby since 1986; he has zero wet-track form, and his Brisnet speed is sub-100. Use him underneath, not on top.
For the record, this is purely for entertainment purposes. We’re a software company, not a sportsbook. Please bet responsibly and never wager more than you can afford to lose.
Step 5: The Human Override
AI finds the pattern. But you still need eyes on the horses: the paddock walk, the warmup, the demeanor. You can’t quantify a horse who looks like he wants to eat the starting gate.
The model narrowed 20 horses to six. From there, the plan was to watch replays, monitor the odds board for sharp money on Derby morning, and make the final call. AI built the map. The human still has to drive.
What We Learned
Just like with March Madness, the real takeaway isn’t about horse racing. It’s about what happens when you give AI a clear framework, structured data, and a specific problem to solve.
The model didn’t replace expertise. It organized it. It cross-referenced data faster than any human could, flagged patterns we would have missed, and turned an unruly 20-horse field into a ranked, actionable betting strategy in hours rather than days.
AI doesn’t replace thinking. It accelerates it. Whether you’re picking Derby horses or solving a business problem, the principle is the same: start with a clear game plan, feed it the right data, and let the model do the heavy lifting so you can focus on the decisions that matter.
Now, if you’ll excuse us, we have a race to watch.
Looking for past Derby Content?
Written by: Savannah Cherry
Savannah is our one-woman marketing department. She posts, writes, and creates all things Slingshot. While she may not be making software for you, she does have a minor in Computer Information Systems. We’d call her the opposite of a procrastinator: she can’t rest until all her work is done. She loves playing her switch and meal-prepping.
Frequently Asked Questions
AI can't guarantee a winner, but it can process far more data than a human can in a fraction of the time. By scoring horses across speed figures, pedigree, pace projections, post-position history, and trainer stats, AI identifies patterns and value that manual handicapping often misses.
The model used speed figures from Beyer and Brisnet, late pace ratings, pedigree and stamina scores, trainer and jockey win percentages at Churchill Downs, post-position history, trip notes from final prep races, workout data, and weather impact analysis.
An overlay is a horse whose actual win probability is higher than what the odds suggest. If a model scores a horse as a realistic 8-1 chance but the odds board lists them at 20-1, that gap represents value. Overlay bets are where smart money finds an edge over time.
A sloppy track changes everything. Six of the last 21 Derby winners ran on a wet surface, and five of those six paid 5-1 or higher. Horses with proven wet-track pedigrees and strong stamina breeding get a significant upgrade, while speed-oriented horses and those without any off-track experience become riskier bets.
In a 20-horse field, where you break from the gate directly impacts your trip. Posts 1-3 risk getting trapped along the rail in traffic, and posts 17-20 force horses to cover extra ground wide. Middle posts (5-12) historically produce the most winners because they offer cleaner paths to the first turn.
