This story was originally supposed to be something like “How AI has changed the sports betting landscape.”
I may as well have set out to write something like “How oxygen has changed the breathing landscape.”
In short: Yep.
I started interviewing people. Gamblers. Business owners. Business owners who are gamblers. The answers were, to quote a Thai phrase, “same same but different.” Everyone is using it, everyone is using it in distinct ways, and the biggest takeaway from them all was basically this, as put by Captain Jack Andrews in a text to me as I was ideating this piece: “We’re on the precipice where there will only be two types of bettors. Those that can use AI and those who lose.”
Top down
“I’ve been using it to build out top-down models for prediction markets and sports betting,” said Justin Herzig, best ball champ and founder of PredictQ.market, a tool to help people find the sharp sides in prediction markets. “The beauty here is what used to take weeks for a research project, I can now have an idea and test it in hours.”
As a “for instance,” Herzig had an idea around golf, what he termed a “theory around potential correlation that may not be showing up in current models.”
He said he used AI to test it on nearly 120,000 player rounds with over 1,800 golfers. He took the results, backtested it against a subset of recent rounds, and …
“Saw a signal among five to eight percent of players, which is enough to give me around three to five solid plays a tournament,” he said.
If you’re asking yourself something along the lines of “Huh?” you’re in good company (well, at least you’re in my company). Because to be clear, there is no magic sauce here, and figuring this stuff out isn’t simply saying abracadabra. There is work to be done, and a learning curve to scale.
“My number one piece of advice is to download Claude Cowork and just start talking into the microphone about what you do for a living, what types of things you like to do on the side, what your goals are and what you’re trying to do more of, and ask the AI how it can help you,” Herzig said.
In other words, figure it out yourself.
Which was echoed by everyone, in one way or another.
Edward Golden, the founder of Right Angle Sports, widely considered one of the sharpest handicapping groups in the business and who was eager once to discuss how sharp bettors and sportsbooks are engaged in their dance, demurred when it came to spilling tea on AI.
“I’d prefer not to get into specifics, but we’re definitely using it,” Golden told me. “It’s made us more efficient, more organized, and able to cover more ground than ever before.”
Too valuable to ignore
Brian Hooper is a longtime pro gambler and co-founder of Endgame Syndicate, which consists of four of the top minds in daily fantasy sports teaching DFS players how to fish (metaphorically speaking). Hooper is unequivocal from the business side: Use AI, or fail.
“I almost feel like we’re getting close to the point where we can separate the people who are using it and the people who aren’t. And that’s also who’s making money and who’s not. There’s too much promise in it to not get familiar with it,” Hooper said. “I’m telling my buddies, you guys have got to do something.”
Hooper notes that pre-AI, the cost of getting into business — gambling or otherwise — could be prohibitive.
“If they were going to start a company, they’d say, ‘I’ve got to hire a coder, I’ve got to hire a website guy, I don’t know how to do any of that stuff,’” he said.
But today?
“For small businesses, those barriers are gone,” he said. “It’s a very good time to be doing all this stuff. It’s very early, but also accessible.”
He then started explaining what he’s done, using the website Run The Sims (which his company bought a little over a year ago) as an example.
“The way I think to use it right now is you have it build the code for you, and then you run the code either on your PC or in the cloud. For Run The Sims, we’re on Amazon AWS, and it writes the code for the changes,” he explained.
“It can look through your code and tell you what’s there. I said, ‘Look at the Run The Sims code, what do we do?’ And it said, ‘This is a DFS tools and props tools company that uses simulations to produce sports betting projections’ — just from reading the code that’s already there. So we didn’t have to redo the code base or hire someone new. It could take what you already have and improve it, which just couldn’t be done before.”
So what did he do?
“I redid our entire model for every sport with AI: How much do offensive linemen matter, how much does aDOT matter versus YAC versus weather concerns, all these things,” he said. “It ingested all of it.”
While Hooper comes off as a bit of an AI bull, he does note the limitations — basically, even people as plugged in as he is will eventually run into the brick wall that is, basically, WTF is this thing doing?
“It’s so black boxy, it’s hard to tell specifically what it’s doing,” he noted.
Enter the dummy
So here are some things you should know about me and AI: I like it. I also have no idea how to use it correctly when it comes to high-level stuff. I don’t have Claude Cowork, for example.
But I wanted to play in the sandbox, even though while the tools may be available to everyone, it doesn’t mean everyone knows what to ask, what data to provide, or when you’re getting garbage back as a result.
Whatever, though.
So I went to ChatGPT, the regular ol’ site, with three separate questions: What happens in MLB games when the home team goes down 1-2-3 in the bottom of the first inning? What happens in NFL games when teams that rely heavy on play-action go up against teams that blitz often? And what happens in NFL games when one team is down 21 points or more at halftime?
The prompts I gave were basically those with a few added bells and whistles. Each prompt took about an hour or so to run. It started telling me things like, “I built a cross-validated classifier using information available at halftime” and “Brier score: 0.197, versus 0.230 for simply predicting 36% every time” and “points over expectation: p = .212.”
Obviously, I’ve presented those sentences without context, but even with context, I would not be able to tell you what the hell it was telling me.
But it did spit out some answers after going through “4,856 regular-season games from 2021–22 and identified 1,426 strict 1-2-3 innings,” and “2,586 offense-versus-defense observations across 1,326 games” and “178 games in which a team trailed by at least 21 points at halftime” over the last 10 years.
The results? Well, mostly unactionable, but did you know …
“When the game was 0-0 entering the bottom of the first, going down 1-2-3 had essentially no negative effect whatsoever on the home team” and “Frequent play action appears to give an offense some protection against aggressive blitzing, but the data does not show that a high-PA-versus-high-blitz matchup by itself produces a dependable ATS advantage” and — OK, maybe this one is actionable, and I’ll paraphrase for brevity’s sake, but if a team is down 21-24 points and receives the second half kickoff, 61% of the time they end up losing by less than 20 points.
Now you know.
The captain speaks
Captain Jack Andrews of Unabated has basically been my own personal gambling AI since I started covering this industry. And he is also — as noted in the very beginning of this story — an AI evangelist.
“FOMO is good in this new world we live in,” he said. “If you don’t think everybody else is 10 steps ahead of you, you’re probably not even thinking on the right track. That should be the driving factor right there, thinking ‘I need to do this to keep up with the Joneses.’ It’s the greatest equalizer in the meritocracy of sports betting that I’ve ever seen. It’s making the people who weren’t quite good enough to be at the upper echelon and boosting them up.”
Andrews also sees the rise of prediction markets, coupled with AI, as something that is about to burst.
“AI is democratizing the science of sports betting,” he said. “And when you work with that in conjunction with prediction markets, which are democratizing the art of sports betting, it’s a revolution. The whole paradigm of what’s possible as a full-time sports bettor has shifted even further upward. It’s probably as big a gold rush as 2018 was with sports betting, or the early 2000s with internet casinos.”
Andrews is cautious, though, in how he uses it, and how he thinks others should use it.
“As you approach AI, it’s not an oracle, it’s an intern,” he said. “It’s going to do what you tell it to, so you need to give it tasks you know can be done. Don’t give it a problem where you don’t already know what the answer should look like, because that’s when it gets it wrong and doesn’t tell you it’s wrong.”
But again: Not magic.
“You can’t expect AI to find data for you. It can suggest where the data might be, but you need to find the data and give it to the AI. Now you have the intern. You’re telling the intern, look over all this data and let me know whether WOBA or WRC+ is better,” he said.
“If you were doing this on your own, it kind of breaks down. You’d say, ‘I need to write a script that parses this data and knows the score when each person comes up to bat.’ You’re talking a couple of days, maybe a week of programming. This is where AI thinks for 30 minutes and comes back with, ‘I’ve come up with a script that’ll parse all the data. Do you want me to run it?’ And you say yes. Magic happens from there.”
OK, fine, so there is a little magic.
Time is now
So the intern has — ahem — AIrrived. It works fast, it never sleeps, doesn’t have a stupid haircut, and it might spit out falsehoods once in a while.
But the question is no longer whether or not you hire it. The question is whether you know enough to supervise it.
