I’m old enough to remember when statistical analysis in sports pretty much meant checking the box scores in the morning newspaper.
Of course, in recent years statistical analysis has exploded in sports. If you’re a bettor or a DFS player or an interested party and don’t use, subscribe, and understand things like wOBA, xwOBA, xFIP, SIERA, DVOA, EPA per play, CPOE, RAPTOR, LEBRON, DARKO, true shooting percentage, zone-entry efficiency, high-danger chances, spin-axis deviation, bat-path attack angle, opponent-adjusted garbage-time-neutral success rate, and a proprietary model that measures how well a left-handed reliever performs on humid Tuesdays after his catcher eats chicken parm, then frankly, I don’t know how you’re even allowed to bet on sports.
But here’s the thing: With the rise of AI, the world of sports analysis — and figuring out what to do — is no longer the sole domain of code jockeys and computer scientists.
Truth is, just about anyone now can create a proprietary model that measures how well a left-handed reliever performs on humid Tuesdays after his catcher eats chicken parm.
For further proof, there’s Tom Crosby.
Crosby is 38. He lives in Massachusetts with his wife and an 11-year-old Cavalier King Charles named Louie. He studied journalism in college. He spent a couple of years in a band making electronic music, right up until his bandmate bolted for Los Angeles on a whim. He has spent the last dozen or so years in the software industry, not as a programmer but as a product manager.
In other words, he is not the guy you’d picture building a sports betting model from scratch.
But that’s what Crosby did. Working alone, doing the whole thing himself, he built TheHomeRuns.org, a research platform for betting home run props. It went live about a week before we talked. It has 28 paying customers. Nobody has asked for a refund.
A baseball guy
Crosby grew up playing ball. He played in high school, dreamed of making it to the MLB, and, in his own estimation, “probably wasn’t good enough” to play in college. He didn’t.
But the love stuck around, and when Massachusetts legalized sports betting, it found an outlet.
“I just got really into the home run angle,” he said.
The model he eventually built pulls in 17 or 18 different signals — things such as the ballpark, the weather, the matchup, the pitcher’s arsenal, the batter’s recent numbers, even how lucky or unlucky a given pitcher has been lately. The idea is that no single number tells you much, but stack enough of them on top of one another and something real starts to emerge from the noise.
“If you look at any one stat, it’s not a very good way to predict anything,” he said. “But my theory is that if you layer enough stats on top of each other, you actually can find signal.”
He started doing this by hand, scoring the day’s best matchups himself, then automating a little more each year before launching the site a few weeks back.
A year ago, Crosby said, none of this was possible. He knows because he tried. He could get the tools to produce individual pieces, but he couldn’t get them to cohere into anything that worked.
“It was basically just still a workbench at that point,” he said.
This year, with the right tools, it clicked. He leaned on Claude and Codex.
“I wrote no code,” he said. “It’s doing all of that.”
But — and this is the part Crosby is insistent about — the AI didn’t erase the barrier to entry. It just moved it.
“You have to have a real vision,” he said. “It’ll do whatever you tell it to, but it’s not magic. I mean, well, it is magic, but it’s not just going to go out and figure out the appropriate way to put things together. You still have to have a plan.”
So the new barrier isn’t knowing how to code. It’s knowing what you want, building a plan to get there, and then grinding it out.
“While it would have been impossible to do it at all a year ago,” he said, “there’s still a lot of nose to the grindstone that’s involved.”
That’s the shift. The grindstone — not the code — is the barrier now. If you build it, etc.
Don’t pay for picks
Here’s a weird thing about a guy selling a betting product: Crosby doesn’t actually want to sell you picks.
“I heard long ago that you should never pay for picks, and I agree,” he said.
What he’s building instead, he says, is “an empowering research platform,” a place to do your own digging, slice the data a bunch of different ways, and keep the fun in it. He’s transparent to a fault about how it all works. The model’s weights are published on the site. So are the results.
“I’m public with everything,” he said. “Transparency is basically my only mode at this point.”
The pitch isn’t a magic list of winners, and he’ll be the first to tell you that.
“If I just gave you a list of 10 picks every day and they all hit, sure, that’s incredibly valuable,” he said. “Also not really possible.”
It costs $10 a month, or $40 for the rest of the season, a number he planned to knock down to $30. As for whether it works: On most days, Crosby says, at least 30% of his top 10 picks hit. On June 12, his top four picks all homered. So did Nos. 7, 9, 12, and 14.
The trenches
Crosby isn’t pretending he’s cracked the books, though he’d like to. He talks about wanting to hit big someday on a crazy home run parlay. But he’s clear-eyed that he’s a bettor building a tool for bettors, himself very much included.
“I kind of consider myself in the trenches with everybody that takes the journey with me,” he said.
And that, in the end, is the story — not whether Crosby’s model wins. The story is that the thing exists at all, that it was built by a guy who can’t code, working by himself with a Claude subscription.
The barrier used to be the code. Now the barrier is just whether you’re willing to sit down and do the work.
So that proprietary model measuring how a left-handed reliever performs on humid Tuesdays after his catcher eats chicken parm? I ran the numbers. The xFIP is through the roof.
