Beating the Big Guys with Data: The $500,000 Win

June 2002, Oakland. In a windowless room at the Coliseum, the Athletics are running their draft, and the room is at war with itself. On one side: the scouts, men who've spent decades reading bodies and swings, who talk about "good faces" and can tell you which teenager looks like a big leaguer. On the other side: the general manager, Billy Beane; his assistant, Paul DePodesta; and a laptop. The laptop keeps naming players the scouts have barely watched, slow, awkward college hitters who share exactly one habit at the plate. Michael Lewis was in the building that spring, there for the book that would make this room famous. By his account, the scouts thought the laptop was wrecking the draft.

The fight was older than the room. Since the 1970s, a night watchman named Bill James had been self-publishing little heresies, the Baseball Abstracts, arguing the game measured the wrong things. Take batting average, baseball's sacred number. It ignores walks completely, as if the most dependable way to not make an out simply didn't count. James's writing was public, cheap, and adored by a small cult of readers. The people who ran baseball ignored it for twenty-five years.

Oakland couldn't afford to ignore it. Their payroll was about $40 million. The Yankees' was over $125 million. And here's the trap a poor team is in: if you judge talent the same way the rich teams do, you just become a worse version of them. Oakland's only shot was to find something the market had priced wrong.

The question they actually asked

Here's the move, and it's simpler than the legend makes it sound. Don't ask, "Who's a good player?"; that's the question everyone was already answering, with scouts and instinct. Ask two separate questions and compare the answers:

1. What actually produces wins? Regress team performance on the hitting skills, mainly on-base percentage (how often you avoid making an out) and slugging (how many bases you get when you hit).

2. What does the market pay for those same skills? Regress player salary on the exact same numbers.

If the market were smart, those two answers would line up: skills that win games would cost the most. The whole Oakland thesis was a bet that they didn't line up, that there was a gap between what a skill is worth and what it costs. Find the skill that wins games but is cheap to buy, and a poor team can compete on arithmetic instead of money.

Two economists, Jahn Hakes and Raymond Sauer, later ran exactly these regressions on the public record and published the results. This is the rare business legend you can actually check, so let's check it.

Strip the baseball away, and the A's found a market inefficiency in feature valuation, a gap between how much a feature drives the outcome and how much the market charges for it. Notice the structure, because it's reusable: you need two models, not one. A model of the target (what drives winning) and a model of the price (what the market pays). One model alone can't find a bargain; a bargain is a disagreement between two models. And the punchline: DePodesta didn't need secret data. Every walk was in the public box scores. The edge wasn't information nobody had; it was a valuation nobody had bothered to run.

What does that gap buy you in the standings? Oakland turned it into wins at roughly half a million dollars apiece, about the cheapest cost-per-win in baseball, and won 103 games in 2002, matching the mighty Yankees at a quarter of the payroll. The 2002 team also ripped off twenty straight wins, an American League record. Lewis's book landed in 2003, the movie in 2011, and "Moneyball" walked out of baseball to become the everyday word for beating the big guys with data.

But this story earns its place in a publication about rigor for what happened next, two catches the poster version skips.

First, the honest asterisk: economist Steven Levitt and others argued Oakland's real engine was its three ace starting pitchers, found, ironically, the old-fashioned way. So the team's wins had other explanations tangled in; what held up cleanly under the regressions was the narrower claim: the market underpaid for on-base skill. Second, and bigger: the bargain expired. Hakes and Sauer found that by 2004, right as the book spread through every front office, the market's on-base salary coefficient leapt (their estimates put it around 3.68), and the discount was gone, corrected almost overnight. Richer teams like Boston bought the same idea with deeper pockets and won titles. The edge died of fame. (Even this is argued over; later work using new-contract data questions how fully the market adjusted, which tells you how slippery "the market has learned" is to prove.)

There's one more quiet embarrassment in the record. That famous 2002 draft the whole room fought over, the laptop's hand-picked players, shown off in the book, mostly flopped. The system was right about the market and still often wrong about individual players. Being right on average and wrong on any given pick is exactly what riding a statistical edge feels like from the inside. It never stops being uncomfortable.

Twenty-four years on, every pricing team, every quant desk, and every growth team runs Beane's playbook and inherits its expiry date. This is the part I'd put on the wall: an insight starts dying the moment you publish it. The A/B-tested growth hack, the alpha factor in a paper, and the clever prompt in a viral thread are each on-base percentage in 2003, an edge that leaks value the second it's shared. What lasted wasn't the finding. It was the habit that kept producing findings — running the two-model check again and again, watching for the next place where value and price have drifted apart. That question never expires. And in our region, where whole industries still steer by proxy numbers, nobody has regressed against the real goal in years; the gaps are wide open and the laptops are cheap.

Bill James wrote the answer in a mimeographed pamphlet in 1977. The market took twenty-five years to price it in and about eighteen months to price it back out.

So here's this week's question for you: What's the batting average of your industry, the sacred metric everyone chases that nobody has actually regressed against the real goal lately?