Why Venture Capital Still Thinks It’s the Scout, Not the Statistician
Deen Khan and Alan Yang
Nov 26, 2025
Back to Insights
When Baseball Was Still About Gut, Not Graphs
For most of the 20th century, baseball judged players by the eye test and a few blunt stats. Scouts prized the “five tools” - batting average, power, speed, arm strength and fielding – and a grab bag of vague intangibles. Scouting reports were full of phrases that sounded precise, but weren’t: “slender body, loose actions, plays light”. They gave the impression of deep insight but pushed the hard part – deciding what those words mean – back on to the reader. Albert Pujols is an obvious counterexample. He dominated at Maple Woods Community College. Yet his pre-draft report led with “bulky body, future weight problem” before mentioning his skills. He fell to pick 402 in the 1999 draft. Two years later he was Rookie of the Year. Jeff Bagwell was another miss. He crushed records at Hartford, a small cold-weather school scouts rarely visited, and still slipped to the fourth round - before becoming a Hall of Famer.
The Moneyball Era
Bill James, a nighttime security guard at a Kansas pork and beans cannery, had been self-publishing his unconventional baseball stats for two decades when the Cardinals drafted Pujols. For years he was a voice in the wilderness: he had dedicated readers, but the baseball establishment ignored him. In the early ‘90s, Oakland A’s GM Sandy Alderson grew tired of gut feel and started reading Bill’s Baseball Abstract. He handed it to his young assistant, Billy Beane – a once promising prospect who never lived up to scouting expectations.
When Beane became GM, he was ready to usher in The Moneyball Era. He hired Paul DePodesta, a Harvard-trained analyst, who spoke the same statistical language. With no money, they had to buy what others overlooked, and Bill’s work pointed them to traits that produced wins but were mispriced. In the movie, Brad Pitt and Jonah Hill made one trait famous: getting on base. The logic was simple: teams that get on base score more runs, teams that score more runs win more games. Most teams still paid for what looked good: batting average, home runs, the old trophies. But getting on base produced runs more consistently than any of them.
The A’s changed how baseball front-offices operated. Analysts used objective data to narrow the field, scouts then went out to test those targets. It flipped the traditional process - stats came first, stories second. And it worked. In 2002, the A’s payroll was $41 million, the Yankees’ was $125 million. Both won 103 games. When Moneyball was published the next year, the league noticed. The Red Sox, led by 28-year-old Yale grad Theo Epstein, made analytics explicit and hired Bill James himself. A year later they won the World Series.
******
The Clubhouse Era: Venture Capital’s Exclusive Beginnings
Over the same period, a different game began on Sand Hill Road. As venture matured from a niche to an industry, it still looked like baseball scouting: small, relational and based on instinct. “Collaborative competitors” co-invested as much as they competed (no fund had enough capital to go it alone). Investor clubs formed – ‘The Group’ held lunch pitches, where founders waited on the sidewalk while investors decided their fate. Deals came through people you knew - professors, friends, ex-colleagues. When Jobs needed money for Apple, his former boss sent him to Don Valentine at Sequoia, who introduced him to Mike Markkula, a retired Intel executive. Markkula wrote a $250,000 cheque and rounded up the rest.
Still, many VCs missed great companies because they trusted archetypes and networks over facts. Jobs failed the archetype test. Tom Perkins and Gene Kleiner refused even to meet him. Don Valentine joked “Why’d you send me these renegades from the human race?” (though, once Apple had the Valley’s social proof, Valentine chased Markkula to get into the deal). After investing Markkula noted that “a lot of people wouldn’t invest… because Steve was so odd.” Pierre Omidyar failed the network test. He wasn’t in the Sand Hill club, and the idea looked low-status – an online flea market. Bessemer’s anti-portfolio records the instinctive reaction: “Stamps? Coins? Comic books? You’ve GOT to be kidding.” Benchmark ignored the stigma, led a $6.7 million round, and made 750x in two years.
Most deals back then began with a gut call. Venrock, the Rockefeller venture arm, was a cautious East Coast fund preferred real revenue. When Jobs and Markkula pitched them on a computer in every home, the partners didn’t even try to model it. The idea felt too speculative to take seriously. “We were flying blind,” one partner said later. They stepped into the hallway, looked at each other, shrugged, and said “what the hell.” Then they wrote a $300,000 cheque for ten percent of Apple. Peter Crisp later joked that people gave them far too much credit for being smart about the decision. The $300,000 cheque turned into roughly $100 million.
The same gut-first, facts-later habit also funded disasters. At the time the A’s systemised data analytics, venture doubled down on instinct. In the dot-com mania, capital was everywhere and competition for hot deals was intense. The pressure was to act fast and data analysis started to look like a tax on speed. With no true framework underlying deals, investors defaulted to gut instinct and a herd mentality. Take Webvan. Benchmark, Sequoia and a parade of late-stage funds piled in, terrified of missing the “next Amazon.” More than a billion dollars went into a grocery-delivery dream with unproven economics. The faster it grew, the faster it died.
“This Time it’s Different”
You could argue venture had its own ‘data’ boom around the Moneyball era. There were numbers and metrics everywhere (few that mattered or were used, but numbers nonetheless). When traditional (read ‘boring’) metrics – revenue, profit, even user growth – were thin, founders and VCs invented new ones: ‘eyeballs’ (slang for unique site visits), and user ‘stickiness’ (time on site) became common parlance. Never mind that few of those visitors became paying users, VCs suspended disbelief and real money followed faux metrics. Pets.com is the punching bag for the era of excess, but countless cautionary tales existed: Fashion e-tailer Boo.com burned through $100 million chasing ‘cool site’ page views, only to find its flashy web design was practically unusable on 1999-era dial-up. Near the end it had 300,000 monthly visitors and just £200,000 in sales.
******
Why Was Baseball the First Game You Could Hack with Numbers?
So why was baseball first to ride the data wave? The game is discrete. Each pitch starts from the same place. Each at-bat is an isolated trial with a clear outcome: ball, strike, hit, out. Roles separate cleanly too. A batter faces a pitcher, and fielders and runners have defined jobs – it’s often an individual contest disguised as a team sport, where success and failure can be assigned to a single player. The records are also deep: box scores and batting averages have been quantified since the 19th century. And, there’s just a lot of it: there are 2,430 regular-season games each year.
If baseball was ready for numbers, why did Bill James spend two decades in the wilderness? Because the culture wasn’t ready. Scouts waved off numbers, claiming heart or grit can’t be measured. Front-offices favoured self-preservation. Economist John Keynes wrote “Worldly wisdom teaches that it’s better for reputation to fail conventionally than to succeed unconventionally”. Try Bill James’ ideas and fail and you didn’t just lose your job – you were blacklisted as an oddball (the Red Sox weren’t the first club to hire Bill, he’d consulted for three MLB teams already – but was forbidden to acknowledge his employment).
Today every club runs an analytics group. Once the ideas spread, the competition moved upstream: proprietary data, quicker decisions, better player development. Baseball has now settled into a hybrid - models and scouts.
Will There Ever be a Patagonia Wearing Bill James?
Since the days of Atari’s hot-tub board meetings, venture has seen itself as more art than science. Even now, half of early-stage investors make decisions by gut, not detailed analysis and most deals still come through personal networks. So, will venture ever have a Moneyball moment?
A discrete game, defined roles and reams and reams of clean historical data were the bedrock for baseball’s embrace of Moneyball. In contrast, venture has historically been an industry of long cycles, unstructured and hard to access data and far from obvious connections between inputs and success.
The venture game is not discrete. The industries it backs change fast. Technology, behaviour and business models shift. Mental models trained on last decade’s data might miss the next wave (no amount of dial-up-era data could have predicted smartphone apps).
Feedback is painfully slow. It can take 10 years to know if an investment really worked. Even when feedback arrives, assigning praise and blame is hard. Each startup is its own case, the few that win do so in idiosyncratic ways and failure is often hard to distil into a single source (Peter Thiel warns against obsessing over losses - there are usually too many causes to learn much from any one of them).
Bill James had mountains of clean baseball stats. Venture rarely does. Most data is private or doesn’t exist. What does exist is messy - pitch decks, founder interviews. There is also just a small sample size, less than a thousand Australian startups are funded each year.
These structural challenges led venture to favour fast, narrative-led decisions even when hard data is thin. And the industry hasn’t been forced to change. Even with a high miss rate, venture’s power law has proven successful. Most VCs are very smart and used to being right, and these past wins taught them to trust their judgement over data.
******
Can Venture Capital Finally See Clearly?
Venture has started to chip away at its old barriers. For starters, there’s simply more data. PitchBook, Crunchbase and Dealroom have pooled a huge amount of private-market data. Startups throw off more ‘data exhaust’: web traffic, headcount, repo stars. It’s messy, but it exists and is scrapable. Enter LLMs, they can read code, reviews and interviews and they’re good at the unglamourous part –handling and cleaning messy, qualitative data – all while flagging insights and red flags.
The game is still fluid. But our ability to ingest and use timely data has improved. Venture’s slow, noisy feedback loop once made it hard to refine a model. Today, private markets are more liquid and transparent. Startups raise more often, and secondary markets now trade shares between rounds, creating earlier signals of direction and value. The result is a faster feedback loop (eventually, tokenised startups may trade continuously). And AI is pushing it further by shifting the speed-rigour trade-off: VCs can now screen hundreds of startups in the time it took to meet one.
It's still hard to assign blame and praise in venture, but pattern-finding is improving. Machine-learning models thrive on the growing pool of startup data. They can scan thousands of cases and surface patterns no human could. A single datapoint - education, early growth, team size - predicts little. But hundreds, combined well, can reveal signals hidden to intuition. And Australia is a great place for trainspotting: leading the world in unicorns per venture dollar.
Will Venture Soon Run on Autopilot?
In one generation, hedge funds taught markets to trade themselves. With AI moving fast, will venture follow? In the 80s, Jim Simons – an ex-codebreaker and geometry professor - decided to treat markets like maths. He hired physicists, not traders, and from a quiet Long Island office they sifted through old price data to see if were really random. As trading moved from pits to screens, data became continuous and machine readable, and Renaissance built a system around it: thousands of small, repeatable edges, each expressed by code. By the 2000s, the Medallion Fund was compounding over 60% a year.
But autopilot investing needs rails: dense, reliable data, fast feedback and stable rules. Venture is not there yet.
Man Meets Machine
Early venture lacked what Bill James had: the structure and data to turn instinct into numbers. We’ve closed some of that gap, but we’re still a long way from automated venture.
The future looks more like baseball today: models and scouts working together. Moneyball didn’t replace human judgment, it increased the amount of data that could be ingested to augment it. AI is doing the same - shifting where we use judgement, not removing it. In a world drowning in data, the scarce resource is still attention, and the point is to aim that attention where it matters most.
The art won’t disappear. Founders still need to be met, trust built, and shocks handled in real time. Done right, venture becomes faster and fairer - decisions based more on evidence than charisma - but you’ll still need foresight, empathy and a bit of luck.
Competition is for Losers
The industry can argue about whether a Moneyball shift is coming. The better question is who prepares first. Some VCs will keep leaning on warm introductions, familiar archetypes and gut feel. Those instincts still matter, but relying on them alone in a market moving this fast is its own risk. We’d rather extend them than defend them.
Moneyball’s lesson wasn’t just “pick winners.”. It was “find what others miss, and move before the crowd.” That’s the game we’re playing: see more, decide better, move faster. That’s what Lighthouse is built for. It widens our field of view far beyond any one network. Each day it pulls in millions of signals across Australia’s startup ecosystem and uses LLMs to clean and link them into a live map of where energy is building long before rounds get crowded. When something starts to move- a team finding its market or a product catching on - we see it early. We reach out not because someone made an introduction but because the data surfaced it.
We invest the same way. Lighthouse’s signals feed predictive models that generate probabilistic briefs. They don’t replace judgement; they anchor it. They help us tell real traction from noise and avoid the pull of hype when the next bubble inflates.
Speed is part of the edge. Founder briefs are ready before the first call. Diligence agents surface what matters. Term-sheet builders shorten the path from first conversation to committed capital. We don’t want to miss the next Jobs because he looks different, or the next Omidyar because he’s outside our circle. Humans make the final call, but machines clear the path so we can focus on what only humans can do: read ambition, build trust and make bold commitments.
Moneyball didn’t remove scouts. It redirected their judgement to where it mattered most. Venture is heading the same way: models broaden the search; people sharpen the decisions.
We’re not waiting for the industry to catch up. We’re running the experiment now.
******
“Adapt or Die” - Moneyball
When Baseball Was Still About Gut, Not Graphs
For most of the 20th century, baseball judged players by the eye test and a few blunt stats. Scouts prized the “five tools” - batting average, power, speed, arm strength and fielding – and a grab bag of vague intangibles. Scouting reports were full of phrases that sounded precise, but weren’t: “slender body, loose actions, plays light”. They gave the impression of deep insight but pushed the hard part – deciding what those words mean – back on to the reader. Albert Pujols is an obvious counterexample. He dominated at Maple Woods Community College. Yet his pre-draft report led with “bulky body, future weight problem” before mentioning his skills. He fell to pick 402 in the 1999 draft. Two years later he was Rookie of the Year. Jeff Bagwell was another miss. He crushed records at Hartford, a small cold-weather school scouts rarely visited, and still slipped to the fourth round - before becoming a Hall of Famer.
The Moneyball Era
Bill James, a nighttime security guard at a Kansas pork and beans cannery, had been self-publishing his unconventional baseball stats for two decades when the Cardinals drafted Pujols. For years he was a voice in the wilderness: he had dedicated readers, but the baseball establishment ignored him. In the early ‘90s, Oakland A’s GM Sandy Alderson grew tired of gut feel and started reading Bill’s Baseball Abstract. He handed it to his young assistant, Billy Beane – a once promising prospect who never lived up to scouting expectations.
When Beane became GM, he was ready to usher in The Moneyball Era. He hired Paul DePodesta, a Harvard-trained analyst, who spoke the same statistical language. With no money, they had to buy what others overlooked, and Bill’s work pointed them to traits that produced wins but were mispriced. In the movie, Brad Pitt and Jonah Hill made one trait famous: getting on base. The logic was simple: teams that get on base score more runs, teams that score more runs win more games. Most teams still paid for what looked good: batting average, home runs, the old trophies. But getting on base produced runs more consistently than any of them.
The A’s changed how baseball front-offices operated. Analysts used objective data to narrow the field, scouts then went out to test those targets. It flipped the traditional process - stats came first, stories second. And it worked. In 2002, the A’s payroll was $41 million, the Yankees’ was $125 million. Both won 103 games. When Moneyball was published the next year, the league noticed. The Red Sox, led by 28-year-old Yale grad Theo Epstein, made analytics explicit and hired Bill James himself. A year later they won the World Series.
******
The Clubhouse Era: Venture Capital’s Exclusive Beginnings
Over the same period, a different game began on Sand Hill Road. As venture matured from a niche to an industry, it still looked like baseball scouting: small, relational and based on instinct. “Collaborative competitors” co-invested as much as they competed (no fund had enough capital to go it alone). Investor clubs formed – ‘The Group’ held lunch pitches, where founders waited on the sidewalk while investors decided their fate. Deals came through people you knew - professors, friends, ex-colleagues. When Jobs needed money for Apple, his former boss sent him to Don Valentine at Sequoia, who introduced him to Mike Markkula, a retired Intel executive. Markkula wrote a $250,000 cheque and rounded up the rest.
Still, many VCs missed great companies because they trusted archetypes and networks over facts. Jobs failed the archetype test. Tom Perkins and Gene Kleiner refused even to meet him. Don Valentine joked “Why’d you send me these renegades from the human race?” (though, once Apple had the Valley’s social proof, Valentine chased Markkula to get into the deal). After investing Markkula noted that “a lot of people wouldn’t invest… because Steve was so odd.” Pierre Omidyar failed the network test. He wasn’t in the Sand Hill club, and the idea looked low-status – an online flea market. Bessemer’s anti-portfolio records the instinctive reaction: “Stamps? Coins? Comic books? You’ve GOT to be kidding.” Benchmark ignored the stigma, led a $6.7 million round, and made 750x in two years.
Most deals back then began with a gut call. Venrock, the Rockefeller venture arm, was a cautious East Coast fund preferred real revenue. When Jobs and Markkula pitched them on a computer in every home, the partners didn’t even try to model it. The idea felt too speculative to take seriously. “We were flying blind,” one partner said later. They stepped into the hallway, looked at each other, shrugged, and said “what the hell.” Then they wrote a $300,000 cheque for ten percent of Apple. Peter Crisp later joked that people gave them far too much credit for being smart about the decision. The $300,000 cheque turned into roughly $100 million.
The same gut-first, facts-later habit also funded disasters. At the time the A’s systemised data analytics, venture doubled down on instinct. In the dot-com mania, capital was everywhere and competition for hot deals was intense. The pressure was to act fast and data analysis started to look like a tax on speed. With no true framework underlying deals, investors defaulted to gut instinct and a herd mentality. Take Webvan. Benchmark, Sequoia and a parade of late-stage funds piled in, terrified of missing the “next Amazon.” More than a billion dollars went into a grocery-delivery dream with unproven economics. The faster it grew, the faster it died.
“This Time it’s Different”
You could argue venture had its own ‘data’ boom around the Moneyball era. There were numbers and metrics everywhere (few that mattered or were used, but numbers nonetheless). When traditional (read ‘boring’) metrics – revenue, profit, even user growth – were thin, founders and VCs invented new ones: ‘eyeballs’ (slang for unique site visits), and user ‘stickiness’ (time on site) became common parlance. Never mind that few of those visitors became paying users, VCs suspended disbelief and real money followed faux metrics. Pets.com is the punching bag for the era of excess, but countless cautionary tales existed: Fashion e-tailer Boo.com burned through $100 million chasing ‘cool site’ page views, only to find its flashy web design was practically unusable on 1999-era dial-up. Near the end it had 300,000 monthly visitors and just £200,000 in sales.
******
Why Was Baseball the First Game You Could Hack with Numbers?
So why was baseball first to ride the data wave? The game is discrete. Each pitch starts from the same place. Each at-bat is an isolated trial with a clear outcome: ball, strike, hit, out. Roles separate cleanly too. A batter faces a pitcher, and fielders and runners have defined jobs – it’s often an individual contest disguised as a team sport, where success and failure can be assigned to a single player. The records are also deep: box scores and batting averages have been quantified since the 19th century. And, there’s just a lot of it: there are 2,430 regular-season games each year.
If baseball was ready for numbers, why did Bill James spend two decades in the wilderness? Because the culture wasn’t ready. Scouts waved off numbers, claiming heart or grit can’t be measured. Front-offices favoured self-preservation. Economist John Keynes wrote “Worldly wisdom teaches that it’s better for reputation to fail conventionally than to succeed unconventionally”. Try Bill James’ ideas and fail and you didn’t just lose your job – you were blacklisted as an oddball (the Red Sox weren’t the first club to hire Bill, he’d consulted for three MLB teams already – but was forbidden to acknowledge his employment).
Today every club runs an analytics group. Once the ideas spread, the competition moved upstream: proprietary data, quicker decisions, better player development. Baseball has now settled into a hybrid - models and scouts.
Will There Ever be a Patagonia Wearing Bill James?
Since the days of Atari’s hot-tub board meetings, venture has seen itself as more art than science. Even now, half of early-stage investors make decisions by gut, not detailed analysis and most deals still come through personal networks. So, will venture ever have a Moneyball moment?
A discrete game, defined roles and reams and reams of clean historical data were the bedrock for baseball’s embrace of Moneyball. In contrast, venture has historically been an industry of long cycles, unstructured and hard to access data and far from obvious connections between inputs and success.
The venture game is not discrete. The industries it backs change fast. Technology, behaviour and business models shift. Mental models trained on last decade’s data might miss the next wave (no amount of dial-up-era data could have predicted smartphone apps).
Feedback is painfully slow. It can take 10 years to know if an investment really worked. Even when feedback arrives, assigning praise and blame is hard. Each startup is its own case, the few that win do so in idiosyncratic ways and failure is often hard to distil into a single source (Peter Thiel warns against obsessing over losses - there are usually too many causes to learn much from any one of them).
Bill James had mountains of clean baseball stats. Venture rarely does. Most data is private or doesn’t exist. What does exist is messy - pitch decks, founder interviews. There is also just a small sample size, less than a thousand Australian startups are funded each year.
These structural challenges led venture to favour fast, narrative-led decisions even when hard data is thin. And the industry hasn’t been forced to change. Even with a high miss rate, venture’s power law has proven successful. Most VCs are very smart and used to being right, and these past wins taught them to trust their judgement over data.
******
Can Venture Capital Finally See Clearly?
Venture has started to chip away at its old barriers. For starters, there’s simply more data. PitchBook, Crunchbase and Dealroom have pooled a huge amount of private-market data. Startups throw off more ‘data exhaust’: web traffic, headcount, repo stars. It’s messy, but it exists and is scrapable. Enter LLMs, they can read code, reviews and interviews and they’re good at the unglamourous part –handling and cleaning messy, qualitative data – all while flagging insights and red flags.
The game is still fluid. But our ability to ingest and use timely data has improved. Venture’s slow, noisy feedback loop once made it hard to refine a model. Today, private markets are more liquid and transparent. Startups raise more often, and secondary markets now trade shares between rounds, creating earlier signals of direction and value. The result is a faster feedback loop (eventually, tokenised startups may trade continuously). And AI is pushing it further by shifting the speed-rigour trade-off: VCs can now screen hundreds of startups in the time it took to meet one.
It's still hard to assign blame and praise in venture, but pattern-finding is improving. Machine-learning models thrive on the growing pool of startup data. They can scan thousands of cases and surface patterns no human could. A single datapoint - education, early growth, team size - predicts little. But hundreds, combined well, can reveal signals hidden to intuition. And Australia is a great place for trainspotting: leading the world in unicorns per venture dollar.
Will Venture Soon Run on Autopilot?
In one generation, hedge funds taught markets to trade themselves. With AI moving fast, will venture follow? In the 80s, Jim Simons – an ex-codebreaker and geometry professor - decided to treat markets like maths. He hired physicists, not traders, and from a quiet Long Island office they sifted through old price data to see if were really random. As trading moved from pits to screens, data became continuous and machine readable, and Renaissance built a system around it: thousands of small, repeatable edges, each expressed by code. By the 2000s, the Medallion Fund was compounding over 60% a year.
But autopilot investing needs rails: dense, reliable data, fast feedback and stable rules. Venture is not there yet.
Man Meets Machine
Early venture lacked what Bill James had: the structure and data to turn instinct into numbers. We’ve closed some of that gap, but we’re still a long way from automated venture.
The future looks more like baseball today: models and scouts working together. Moneyball didn’t replace human judgment, it increased the amount of data that could be ingested to augment it. AI is doing the same - shifting where we use judgement, not removing it. In a world drowning in data, the scarce resource is still attention, and the point is to aim that attention where it matters most.
The art won’t disappear. Founders still need to be met, trust built, and shocks handled in real time. Done right, venture becomes faster and fairer - decisions based more on evidence than charisma - but you’ll still need foresight, empathy and a bit of luck.
Competition is for Losers
The industry can argue about whether a Moneyball shift is coming. The better question is who prepares first. Some VCs will keep leaning on warm introductions, familiar archetypes and gut feel. Those instincts still matter, but relying on them alone in a market moving this fast is its own risk. We’d rather extend them than defend them.
Moneyball’s lesson wasn’t just “pick winners.”. It was “find what others miss, and move before the crowd.” That’s the game we’re playing: see more, decide better, move faster. That’s what Lighthouse is built for. It widens our field of view far beyond any one network. Each day it pulls in millions of signals across Australia’s startup ecosystem and uses LLMs to clean and link them into a live map of where energy is building long before rounds get crowded. When something starts to move- a team finding its market or a product catching on - we see it early. We reach out not because someone made an introduction but because the data surfaced it.
We invest the same way. Lighthouse’s signals feed predictive models that generate probabilistic briefs. They don’t replace judgement; they anchor it. They help us tell real traction from noise and avoid the pull of hype when the next bubble inflates.
Speed is part of the edge. Founder briefs are ready before the first call. Diligence agents surface what matters. Term-sheet builders shorten the path from first conversation to committed capital. We don’t want to miss the next Jobs because he looks different, or the next Omidyar because he’s outside our circle. Humans make the final call, but machines clear the path so we can focus on what only humans can do: read ambition, build trust and make bold commitments.
Moneyball didn’t remove scouts. It redirected their judgement to where it mattered most. Venture is heading the same way: models broaden the search; people sharpen the decisions.
We’re not waiting for the industry to catch up. We’re running the experiment now.