← Back to list
Colossus (Invest Like the Best / Business Breakdowns)Podcast10 Dec 2019Source: investlikethebest.libsyn.comHost: Patrick O'Shaughnessy

Jeff Ma – Making Decisions with Data - [Invest Like the Best, EP.151]

In plain words

Jeff Ma, a former MIT blackjack team member and data scientist, discusses using data for decisions in sports betting, hiring, and social media. He says less than 1% of sports bettors consistently profit, and their edge is fragile because rules and players change yearly, forcing constant model updates. He credits Twitter's recovery to shifting focus from monthly to daily active users. He warns Netflix and similar streamers haven't truly evolved—they just make old TV content prettier and shorter.

AI SummaryAI-generated · may contain errors · verify against the original

At a Glance In Invest Like the Best EP.151, Jeff Ma explores the application of data in sports and business decision-making. The core argument is that quantitative analysis has expanded from sports betting to areas such as talent evaluation and social media, yet human judgment remains indispensable.

~9 min full read · 8 sections
Deep Analysis

At a Glance

Jeff Ma is a former member of the MIT Blackjack Team and the real-life inspiration for 21. He later founded the people analytics firm 10X and served as Vice President of Data Science at Twitter. This episode’s main thread is the application of data in decision-making across sports betting, human capital assessment, and social platform governance. The most impactful takeaway from the entire episode: Jeff Ma believes that fewer than 1% of sports bettors are consistently profitable, and their moat is very shallow—top bettors must constantly rebuild their models because the rules of the game, schedules, and player rotation patterns change every year; static models are doomed to fail.


Theme 1: Sports Analytics Has Shifted from "Acceptance" to "Data Quality"

Jeff Ma points out that the biggest evolution in sports analytics is "acceptance"—now, virtually every professional team must use analytics to compete.

  • Basketball: The efficiency of three-point shots (especially corner threes) has fundamentally changed gameplay—teams aggressively shoot threes, while defenses scramble to block corner threes. This is a classic example of analytics driving game changes.
  • Baseball: Traditional stats only record "hits," but a hit could be a ground ball that slips through a gap or a line drive that hits the wall. Now, through camera capture of underlying data such as exit angle, spin rate, and exit velocity, more accurate predictive models can be built.
  • Football: Jeff considers this the biggest blue ocean—on-field strategies (such as punt/field goal decisions) remain poorly understood. Even the Patriots often opt for field goals when analytics indicate a "very poor decision" (e.g., attempting a field goal from inside the 5-yard line is almost always a bad idea, as it forfeits possession and loses the chance for a touchdown).

Key analogy: Jeff's conversation with legendary coach Bill Walsh—Walsh intuitively stated that "a first-and-10 needs about 4-plus yards to be considered successful," which was nearly identical to the 4.5 yards Jeff's team calculated using Excel. "Geniuses don't need models, but everyone else does. The value of analytics lies in unlocking the patterns within the genius's mind."


Theme 2: Gambling Profits Are Highly Concentrated, but the Moat Is Very Shallow

Jeff Ma believes that fewer than 1% of professional gamblers can consistently beat the market, and their edge is difficult to sustain.

  • Concentration: In any sport, typically only one or a very small number of individuals capture over 100% of the profits (because a large amount of "dumb money" incurs losses).
  • Moat: Very shallow. Top bettors must constantly adjust their strategies—using the NBA as an example: changes in the schedule (more player rest days), rule changes, and the evolution of playing styles (the three-point revolution) render static models based on historical data ineffective.
  • Non-quantitative players: A minority profit through market arbitrage (exploiting the price differences between the black market and legal markets), without relying on quantitative models.
  • Psychological resilience: The most prominent trait of successful gamblers is not analytical ability, but the discipline to overcome loss aversion and short-term thinking, and to adhere to the process rather than the outcome. This is entirely consistent with investing.

Jeff's comparison: In blackjack, the model is stationary (the probabilities of the cards remain unchanged), but in sports betting, the data is non-stationary—rules, players, and the competitive environment are all changing, exactly like the public stock market.


Theme 3: Human Capital Analytics — Great Data, but People Don’t Want to Be Quantified

The 10X company founded by Jeff Ma attempts to evaluate software engineer performance using digital footprints (Jira, GitHub, calendars, etc.), but faces two major obstacles.

  • Obstacle 1: Self-motivation is overestimated. Professional athletes enjoy reviewing their own statistics and setting goals, but ordinary knowledge workers are far less proactive — the tool must become a top-down mandate from management.
  • Obstacle 2: Poor data hygiene. For “non-active work” tools like Jira, employees must first be taught proper usage workflows before the data can be used for analysis — which is a nightmare in itself.

The irony of hiring: Jeff points out that almost no predictive models or data are used in the hiring process, relying entirely on interviews (“Would you want to have a beer with this person?” — the ultimate bias). He recommends focusing on intellectual curiosity and creative thinking, rather than skill matching, because startup environments require constant adaptation.

Building a data science team: Jeff prefers PhD candidates with scientific backgrounds (chemistry, economics, etc.) over those with purely computer science backgrounds. Reason: they are trained in rigorous scientific methods and can formulate hypotheses and design experiments. However, such individuals often underestimate themselves — they need someone to give them a push.


Theme 4: Data Science at Twitter — From User Growth to Election Interference

Jeff Ma led the data science team at Twitter, tackling complex issues ranging from the restructuring of user growth metrics to Russian interference in elections.

  • Metric Restructuring: Twitter once faced stagnant user growth, with its stock price falling from $50 to $14. The team used a Hidden Markov Model and found that the healthiest users were daily active users (DAU), rather than monthly active users (MAU). Shifting the target from MAU to DAU was a key factor in Twitter's subsequent recovery.
  • Election Interference: The team conducted foundational work for a congressional report. They discovered that the Russian accounts' strategy was "amplification, not creation" — heavy retweeting and coordinated attacks, with behavioral patterns similar to accounts already flagged by Twitter, but operating in a gray area (not violating terms of service, yet clearly manipulative).
  • Algorithmic Gaming: Manipulators understood the recommendation algorithm and used coordinated attacks to push disinformation to the surface in areas like search. This prompted Twitter to change the ranking algorithm in certain areas and introduce credibility scores.

On Platform Responsibility: Jeff supported Jack Dorsey's decision to ban political ads ("influence should be earned, not bought"), but acknowledged the complexity of the issue — echo chamber effects (who you follow determines what you see) make the platform susceptible to becoming a self-fulfilling prophecy.


Theme 5: Future Investment Directions — Media Evolution Is the Biggest Blind Spot

Jeff Ma argues that current media companies (Netflix, Hulu, Apple TV+, Quibi) have not truly "evolved" — they have merely made the television content of 25 years ago look better and shorter.

  • True evolution should be interactivity — the convergence of video games and traditional media. Millennials and Gen Z will not sit through a 3.5-hour football game; they are already on their phones.
  • Misconception: People often conflate esports with traditional sports — "Just because I'm a sports fan doesn't mean I'll watch a League of Legends match." The two are fundamentally different species.
  • Other directions: Blockchain (asset tokenization), the sharing economy (only scratching the surface), and urban function optimization.

Mentioned Positions

Position Guest Sentiment Key Data
Twitter Bullish (as former VP, affirms its data science transformation) Stock recovered from $50 to $14 low; DAU replaced MAU as core metric
10X (his founded company) Neutral (admits failure—people not as self-driven as expected) Software engineer digital footprints (Jira/GitHub) have poor data hygiene
Netflix/Hulu/Apple TV+/Quibi Risk Warning (not truly evolved media) Content remains linear narrative from 25 years ago, only better picture quality and shorter duration
San Francisco 49ers Bullish (innovative contract structure) Leads in guaranteed amounts and salary cap management for NFL contracts
Boston Red Sox Neutral (as analysis case) Traditional data only records "hits"; new data (exit angle/spin rate/exit velocity) can build underlying models
Stats Bomb (Ted Knutson's company) Bullish (football analytics innovation) Manually reviews matches to create new datasets, predicts scoring opportunities rather than goals themselves

Judgments Worth Remembering

1. “In sports betting, less than 1% of participants are consistently profitable, and the moat is very shallow—top bettors must constantly rebuild their models because the rules, schedules, and player rotation patterns change every year.” (Jeff Ma) — Static models are doomed to fail, exactly mirroring the non-stationarity in investing.

2. “Geniuses don’t need models, but everyone else does. The value of analysis lies in unlocking the patterns within a genius’s mind.” (Jeff Ma) — Bill Walsh intuitively identified the 4.5-yard success line, consistent with Excel calculations, but ordinary people need data.

3. “The biggest bias in hiring is ‘Would you want to have a beer with this person?’—This describes whether the other person resembles you, not whether they can do the job.” (Jeff Ma) — Interviews are essentially bias amplifiers, yet almost no company uses predictive models to replace them.

4. “Data science teams should prioritize hiring science PhDs (chemistry, economics) over pure CS backgrounds—they understand the scientific method and can design experiments, but often underestimate themselves.” (Jeff Ma) — This is an arbitrage opportunity in the talent market.

5. “Twitter is the only tech company that has recovered from user growth stagnation—the key was shifting the target from MAU to DAU, and through HMM modeling, discovering that the healthiest users are daily active users.” (Jeff Ma) — Metric selection determines company behavior.

6. “Russia’s election interference strategy was ‘amplify, not create’—massive retweets, coordinated attacks, behavioral patterns similar to already-penalized fake accounts, but operating in a gray area.” (Jeff Ma) — Gray-area manipulation is harder to counter than outright violations.

7. “Current media companies (Netflix, Hulu, Quibi) haven’t truly evolved—they’ve just made TV content from 25 years ago look better and shorter. True evolution should be interactivity, such as the fusion of video games and traditional media.” (Jeff Ma) — Millennials won’t sit through a 3.5-hour football game, but esports ≠ sports.

8. “The blackjack model is stationary (card probabilities don’t change), but in sports betting and investing, data is non-stationary—rules, players, and competitive environments all change, requiring constant questioning of the model.” (Jeff Ma) — This is the core cognitive leap from the MIT blackjack team to a data scientist.