Hours before Dana Holgorsen’s American Athletic Conference debut, Michael Schwimer made a bold proclamation on ESPN’s Daily Wager, a new gambling program focused on the rapidly expanding legalized sports betting market.
Led by Holgorsen, Houston opened American Athletic conference play on Sept. 19 as a 5-point underdog at Tulane, an AAC West foe that fell to the Cougars 48-17 a year earlier. While the Green Wave defense can unsettle quarterbacks with frequent blitzes and stunts, Houston countered with a veteran quarterback in D’Eriq King who has the acumen to handle exotic looks, Schwimer explained. Schwimer took Houston plus the points.
For the majority of three quarters, Schwimer’s pick appeared correct. The Cougars raced out to a 28-7 lead on a 35-yard touchdown run by King, one of three touchdowns he accounted for on the night. Unfazed, Tulane methodically erased the deficit with a strong ground attack of its own. With the game tied at 31 in the waning seconds, Tulane wide receiver Jalen McCleskey caught a pass in traffic, eluded three defenders and scored on a 53-yard touchdown to give the Green Wave a 38-31 lead and the victory.
One play earlier, Tulane advanced near the 50-yard line on a fake kneel down, leaving 12 seconds left, enough time for McCleskey to notch the improbable, game-winning touchdown.
Tulane beat Houston yesterday on this crazy last-second touchdown 😮 @GreenWaveFB
— The Checkdown (@thecheckdown) September 20, 2019
A revolutionary idea or risky concept?
The fluke ending underscores the inherent challenges of predictive modeling in football betting, the danger of “locks” and guarantees. The most advanced algorithms can forecast an expected outcome based on nearly every matchup that could transpire on the field: short-yard running situations, special teams coverage on punts above 50 yards and overload blitzes on 3rd-and-long. But can an algorithm predict that two defensive backs will miss a tackle, while a third slips in coverage? What models have the ability to determine if a defense can stop wild, unconventional gimmick plays? The proposition seems nearly impossible.
Schwimer, a former MLB pitcher with the Philadelphia Phillies, is putting his models to work. Last December, he launched JAMBOS Picks, a machine learning-based sports picks subscription service that aims to outsmart sports betting markets. Of Jambos’ staff of 25 full-time employees, the vast majority is comprised of data scientists, Schwimer said, along with data engineers and computer programmers.
Retired MLB pitcher Michael Schwimer is betting on analytics https://t.co/8DPZ2ZwjAU
— Phil Sanchez (@Phil_Sanchez) September 17, 2019
“Each member of our team goes through an extensive assessment process,” Schwimer told Sports Handle. “We have had countless Ivy League-educated statisticians apply, but are completely unable to do the work we need. In predictive modeling for sports you need to be able to deal with messy data and figure out answers to questions that you do not learn in school.” He continued, “It takes creativity and figuring out ways to get the best predictions with incomplete information in order to work for us. Those people are very hard to find.”
Ahead of the football season, Jambos began offering subscribers a 17-week plan comprised of picks from NFL, NCAA Football, MLB and NCAA men’s basketball games. For a $3,000 fee, subscribers will receive a minimum of 1,000 picks on the four sports. Jambos also offers smaller packages on a 1-week and 4-week basis. Subscribers are guaranteed a minimum of 50 picks with the weekly package and 200 recommendations in the latter. Jambos regularly sends out email alerts that new picks are posted on the website
“I care about showing the world that we can predict outcomes of games in sports better than any person or group that has ever been around,” Schwimer told Sports Handle. “That is the goal.”
If Jambos can meet those lofty expectations, he expects the company to flourish.
“I firmly believe we will turn Jambos into a multi-billion company,” Schwimer said.
But to get there, or anywhere close, an industry source explained, Jambos will have to find a way to overcome a fundamental challenge: scalability.
“The fundamental issue endemic to pick selling is scalability,” the source explained. “If you’re successful, you move markets and your customers can’t get down at the release price. The only way to have sustained success as a tout is to market yourself well and not have any market influence. If your picks are winning picks, you will have market influence. No guarantee will change that.”
The Jambos business model
Besides its reliance on machine learning techniques to guide its sports picks, Jambos also differentiates itself from other tout services with “a first-of-its-kind financial guarantee,” the company’s brochure says.
In the event that Jambos’ picks finish in negative territory at the completion of the 17-week package, the company has promised to deposit $10,000 back into a subscriber’s account. The company’s pick structure is based on a calculation of 1 unit as the equivalent of $300 when making a wager at a sportsbook. The other weekly packages also offer guaranteed payback features of $300 and $2,000, respectively.
At the very least, those interested in the service should read the fine print before making a substantial investment. Once a subscriber receives the weekly picks, their chance of receiving the guarantee is out of their hands. The premise of Jambos’ guarantee is based on the performance of picks made at the time of their release, not according to whether, or to what extent, subscribers can bet the number in any particular amount at a sportsbook.
After weighing several options, Jambos selected Costa Rican-headquartered BetCris (Bookmaker) as the sportsbook for providing the market consensus line. Jambos currently uses the BetCris line near the time of its release to grade all of its picks. Each recommendation is time stamped with the corresponding betting odds.
On Sept. 22, Jambos released a 2-team, 6.5 point teaser (Vikings -2.5 and Chargers +3.5) at 11:00 am ET, two hours before kickoff at U.S. Bank Stadium. Four nights earlier, the website released a spread bet on the Bears -3.5 over the Redskins on Monday Night Football.
When examining the relationship between touts and customers historically, it is one that can shatter instantly if the subscriber perceives the slightest hint of impropriety.
“Any tout that is going to have any ethics or credibility at all, needs to have transparency,” said Keith Miller, a gaming law professor at Drake University Law School.
The Jambos’ system also factors in “juice,” or the commission that a sportsbook receives on losing bets. Priced at -130 at BetCris, the teaser lost when the Texans prevailed over the Chargers 27-20. While each win with Jambos is worth 1 unit, a defeat accounts for the vig. In the case of the Chargers’ loss, the defeat represented the equivalent of -1.3 units.
If the Jambos experiences a modicum of success during the NFL regular season, their subscribers may reap the dividends. At the end of the 17-week period if Jambos is up 25 units, a subscriber could make a profit of $4,500. On the other hand, if Jambos takes a bath and ends the season down 25 units, a subscriber could suffer a $500 loss even with the guaranteed $10,000 return. At $300 a unit, a subscriber would drop $7,500. Tack on the $3,000 subscription fee, then the customer will be in the red.
A customer will only lose money in the 17-week plan if Jambos is down at least 23.33 units, according to the company’s website. After the picks are graded, the results are available on the Jambos’ Picks website free of charge. As of this writing, Jambos’ picks are down 3.17 units through Week 3 of the NFL season.
It is important to note that the picks, in numerous cases, are based on “soft openers” shortly after BetCris releases its initial lines, not the closing price on pregame markets. With MLB and NCAA basketball selections, Jambos releases their picks on the day of the game, Schwimer said. Typically, Jambos publishes NFL and NCAA football picks at least 12 hours after the lines come out, he added. Jambos usually posts college football picks on Tuesdays, one day before its NFL picks are released.
In the case of betting openers, according to an industry source, the entire problem — by design — is that bettors can’t put that much money down. Possibly not even the recommended minimum $300.
“It’s almost certain there will be instantaneous line movements,” a source said.
To illustrate this dynamic: if Jambos releases a pick on the New York Giants -2.5 (-110), and quickly after release or even 30 minutes, only -3 is available, or -2.5 with a higher vig, the subscriber would not be able to deploy the pick.
“If [Jambos] offered a guarantee where the guarantee was based on the closing line price that could be a solution,” one source explained. “But when you release the picks nobody knows what the closing line will be.”
Jambos addresses the conundrum on its website, even advising customers to discontinue their subscriptions if they become frustrated by the recurrence of volatile line movements.
We cannot predict how future line movements will happen because that is entirely out of our hands. However, if you are consistently seeing lines that are drastically different from the lines we post when you go to place wagers, then we recommend you do not continue buying packages with us. Our goal is to provide our subscribers with a winning formula and if lines are moving drastically that will be much harder to achieve.
Subscribers will have to reconcile the value of a guarantee connected to such unpredictability.
The liquidity of the company
Last month, Jambos received $23 million in seed funding, according to an SEC filing. Among Jambos’ investors include: business executives Bill Miller and David Ganek, Marvin Bush, the youngest son of former U.S. President George H. W. Bush and former New York Racing Association Chairman Steven Duncker. The company has enough funds to cover their guarantees in case the Jambos’ picks model loses, Schwimer said.
1/2 Big News! JAMBOS set up an escrow account with Capital One. They will independently verify that we have enough funds to cover our total liability, updated weekly as our liabilities change, and we will post a bank letter from Capital One to our website, under the FAQs section.
— Michael Schwimer (@mschwimer) September 4, 2019
Jambos also posted a letter from Capital One Bank on its website to assuage concerns that the company might lack the funding to cover their liabilities. According to the letter, the company maintains an escrow account with Capital One with a balance equal to or greater than the JAMBOS total subscription liability amount, as of Sept. 25, 2019. Jambos updates the letter each week on Wednesdays after package sales for that week close.
Schwimer declined to release any figures related to the company’s subscription base.
There are indications that Jambos received less than 100 subscribers for their first 17-week package. The subscribers are mainly high-worth individuals from the financial industry, according to a source.
Jambos is now offering another 17-week plan set to begin on Oct. 8. The subscription totals for Jambos’ 1-week and 4-week plans have not been publicly disclosed. It is not inconceivable for Jambos to look to add hundreds of thousands of subscribers in a short period of time.
Big League Advance
— Sports Illustrated (@SInow) September 4, 2018
Schwimer, a former pitcher at the University of Virginia, received third-team All-American honors from the National Collegiate Baseball Writers Association in 2008, a season when he went 3-1 with a 1.72 ERA. After Schwimer was selected by the Phillies in the 14th Round of the 2008 MLB First-Year Player Draft, the 6-foot-8 pitcher spent time with four minor league teams before he received a call up to the majors in 2011. Schwimer made 47 appearances as a reliever with the Phillies over two seasons, but retired prematurely after suffering a labrum tear in his shoulder.
Following his retirement, Schwimer founded Big League Advance (BLA), an analytics company that uses predictive modeling to forecast the probability that a minor league prospect could eventually receive a promotion to the majors. In essence, BLA provides the minor league players with capital to help further their careers in exchange for a cut of their major league earnings upon their promotion. Since its launch in October 2016, 38 of 77 players from BLA’s first fund have spent at least one day with an MLB team, including Fernando Tatis Jr., a standout rookie with the San Diego Padres.
Duncker, a retired executive at Goldman Sachs & Company, invested in BLA’s second investment fund before becoming a general partner of the company. The BLA models have produced a success rate of around 50%, a rate that far exceeds the average minor league call-up rate. It compelled Duncker to also invest in the sports picks subscription service.
“I have pretty high expectations for Jambos,” Duncker said.
Some of Jambos’ data analysts also split their time between both of Schwimer’s companies. Their credentials include previous stops at STATS LLC, TB12 and the NBA. One analyst conducted research on novel concepts of machine learning during an internship at the Air Force Research Lab in Upstate New York.
During a soft launch earlier this year, Jambos exceeded Duncker’s expectations in college basketball. Over a five-month period spanning December 2018 through April 2019, Jambos claims a record of 850-601-25, finishing up 189.5 units. For the season, Jambos says its recommended picks performed at a winning clip of 57.5%.
However, a complete log of picks only begins on Feb. 13, at which time Jambos indicates it was already up 137 units. The website says a law firm examined its picks:
CohnReznick was engaged to perform an examination of JAMBOS, LLC’s (JAMBOS) assertion that its Overall Total Percentage of correct predictions of for the period December 8, 2018 to February 12, 2019 (the Period) was correct. JAMBOS’ Overall Total Percentage was based on the types and number of predictions made during the Period and the Overall Total Percentage 59.38% of correct predictions was calculated as the number of correct predictions made relative to the total number of predictions made.
The company also notes a potential course change between the college basketball selections and future picks, adding, “The type and number of predictions made in the future by JAMBOS may be materially different than the type and number of predictions included in the examination.”
If a picks service has high-quality data that is markedly better than its competitors, it is not difficult to beat soft openers, an industry source said. The task also becomes less complicated when you are capitalizing on market inefficiencies borne from obscure contests. Any oddsmaker worth their salt can set a fair opening line for Duke-North Carolina, but only the most savvy, informed ones can do the same for the first half total of Savannah State-North Carolina A&T. And, even if you are able to lock in a bet at the early release there is a good chance your limits will be drastically lower in comparison with the amount you can put down just before the tip.
Speaking of the Tigers, Jambos recommended picks featuring Savannah State at least seven times last season. Over the final three months, the subscription service also touted picks on Lipscomb, Belmont and Hofstra seven times each. By comparison, Duke, Kentucky, Michigan State and Virginia were featured in 40 picks, led by the 2019 national champions with 13 selections. There appears to be a good mix between offerings on Top 25 matchups along with recommendations on less-heralded mid-majors.
Current results through Week 3 of NFL season
Week 4 recap = 108 recs. +2.99 units
4 week review
Week 1 = +8.49
Week 2 = -19.77 (worst week in JAMBOS history)
Week 3 = +10.83
Week 4 = +2.99
Total = +2.54
Subscribers profitable after paying our fee in 4 of the 5 total packages that have closed. 4 (1 week) and 1 (4 week)
— Jambos (@JambosPicks) September 24, 2019
As of Sept. 23, Jambos went 27-27-1 in 55 NFL games, or -3.17 units through three weeks of the NFL regular season. The mark includes a horrendous opening week when Jambos hit on only 4 of 13 picks during the first Sunday of action (-5.92 units). Overall, Jambos finished the week down -19.33 units. Subscribers of the 1-week package were refunded $300 each.
In college football, Jambos enters this week with a 61-54-3 record (+1.43 units). Jambos has been in positive territory since Sept. 13 when it hit on all four of its picks on the day. On Sept. 21, Jambos went 19-17 to finish the daily slate up 0.15 units.
Football is more challenging to model than other sports, according to Schwimer, namely college basketball. The models are unable to account for unexpected strategic changes made by a coaching staff. In the second half of a Week 2 loss at Detroit, the Chargers’ staff surprised Schwimer by running different play types than the Jambos’ models anticipated based on the defensive formations and coverages presented by the Lions. The algorithms can also be remarkably precise, yet still predict a losing outcome. Schwimer said Jambos handicapped the Packers-Bears’ season opener 90% correctly, but underestimated the capabilities of the Packers’ defensive line in pass rushing situations.
“If you have one area wrong it can affect the whole game,” Schwimer said.
Nevertheless, the vast data sets at Schwimer’s fingertips enable Jambos to come up with predictive outcomes at an extremely granular level. When asked if Jambos tracks Saquon Barkley’s yards per carry versus an 8-man box, Schwimer responded that the data is much more advanced. Eventually, an A.I.-based sports picking model may be able to incorporate data on how quickly a corner flips his hips when beaten on a fly route, along with his recovery time when he enters into a full sprint to break up a pass, Schwimer said. The company’s models could be infused by an incredible amount of data.
By the same token, sports betting companies with advanced predictive modeling could incorporate data on how long a corner can hold a receiver in press-man situations into various bets. Down the road, they may be able to use player tracking technology to determine how quickly a receiver can get in and out of breaks. Already, MGM Resorts uses sophisticated analytics to develop win probability models for certain high-profile NFL games. This week, the Cowboys have a 55% probability to defeat the Saints on the moneyline, according to BetMGM. The forecast is based in part on Pro Football Focus’ signature stats ranging from defensive pressure rate to passer rating from a clean pocket, as well as play action rate and yards per attempt on play action plays.
Potential long-term ramifications for the market
9/29 Record: 10-11-3, -1.52 U
MLB: 11 plays, 5-3-3, +2.66 U
NFL: 13 plays, 5-8, -4.18 U
Cumulative Records (Public Picks only vs. BetCris): +70.95 U
MLB: 1491 plays, +33.14 U
NCAAB: 527 plays, +39.05 U
NCAAF: 148 plays, +4.11 U
NFL: 70 plays, -5.35 U
— Jambos (@JambosPicks) September 30, 2019
Schwimer’s foray into sports betting has not come without detractors and skeptics. One of the primary concerns among several industry experts interviewed for this story stems from the cascading effect that could ensue if a bevy of sharp bettors begin to place large, six-figure wagers from information derived by Jambos’ models or the picks themselves.
“My feeling from looking at touts over the years is, if they are good then they will be figured out by sportsbooks very quickly and people will not be able to get the betting line,” professor Keith Miller said. “The line will change so quickly that most people will not be able to get down on the game at that line.”
Miller points to the travails endured by Right Angle Sports (RAS) owner Edward Golden, a sports handicapper whom he describes as one of the “most honest, legitimate touts,” in the profession. In recent years, the handicapping service has been forced to constantly regroup on the fly as sportsbooks adjusted lines almost immediately after RAS released a play, Miller said.
Miller, a co-author of The Law of Gambling and Regulated Gaming, a leading textbook on gaming law, is also a visiting professor at UNLV where he teaches a course on sports betting and fantasy sports.
Over the last 10 years, Golden’s Right Angle Sports service is up 255.6 units, winning at a rate of 55.6%. Last season, he topped 58% in both football and basketball. Based on a data set of 1,000 picks, a solid win rate for a person betting openers is somewhere in the mid 50s.
It may require a sample size of 5,000 to 10,000 picks before Jambos can be properly evaluated, Schwimer indicated.
In some markets, the line movements will be instantaneous, another source explained. For more opaque ones, it may take simply one bettor or a single betting syndicate to move the line considerably.
Companies that attempt to scale their touting services may also fall into a trap of acting as if certain markets are more liquid than they actually are. Not only does this apply for markets with early lines, but for most sports beyond the major U.S. professional leagues.
Schwimer is acutely aware of the tectonic shifts that could ensue across the industry from the rapid line movements.
“We have been doing it for free for everyone for eight months and it hasn’t happened yet,” Schwimer said. “However, when it does happen, we will be in an incredible position as a company…we will be the worldwide market mover for a $500 billion industry.”
“I will use the harshest language possible: The industry is disgusting, it’s heartbreaking, it’s just awful,” he said.
Will it end in handsome profits for Jambos subscribers, or, as in the case of Houston backers at +5, heartbreak?
When it comes to covering the spread, there are no guarantees.
Brett Smiley contributed to this report.