Every few decades, the so-called quants invade another industry. They arrive with data, models, and a (sometimes irritating) conviction that the self-described experts don’t know as much as they think. Sometimes they’re wrong. But when they’re right, they don’t just improve how the industry works. They also change who wins.
I’ve spent much of my career at the intersection of quantitative methods (data, machine learning, AI, model-driven businesses) and finance. I’ve also been fascinated by other industries that have “gone quant.” Looking back over the past several decades, it’s been one of the most important recurring patterns in business and one of the great ways to make money.
I think we’re now on the cusp of another wave. AI is making it possible for industries that have historically resisted quantification to undergo the same transformation we’ve seen elsewhere.
Here are three of my favorite examples of industries that have gone quant over the past few decades, starting with the one I know best.
Finance. For much of its history, investing was dominated by experienced professionals making judgments about companies, markets, and securities. Then the mathematicians, physicists, and computer scientists arrived, bringing new data, statistical techniques, and the belief that markets contained patterns that systematic approaches could exploit. Firms like Renaissance Technologies demonstrated just how extraordinarily profitable this could be. (The story is wonderfully told in Gregory Zuckerman’s The Man Who Solved the Market.)
Today, some of the world’s most successful investment firms—including Citadel, Two Sigma, and D. E. Shaw—have substantial systematic elements, although their approaches differ considerably (an article for another day). In some cases, models make the investment decisions themselves. In others, human investors make the bets, but systematic approaches to portfolio construction, risk management, and capital allocation play an enormous role. The common thread is that investing has become much more systematic, and the firms that figured out how to do this well have captured enormous value.
Politics. Political campaigns were traditionally run by experienced operatives who relied heavily on instinct and conventional wisdom to decide which voters to target, which messages would resonate, and where to spend campaign resources.
That began changing well before Barack Obama, but his presidential campaigns—particularly 2012—brought quantitative campaigning into the mainstream. As Sasha Issenberg chronicles in The Victory Lab, campaigns increasingly used data, modeling, and controlled experiments to determine which voters to contact, which messages worked, and how to allocate resources. There were two particularly important dimensions: deciding where to invest campaign resources and deciding what to say.
That process has only accelerated. I recently watched a media advisor to a major U.S. Senate campaign take a ten-minute interview with a voter at his front door about his support for the candidate, turn it into 200 possible 20-second ads, and distribute them across multiple social media platforms to test which ones deserved more investment. That’s a long way from the consultant who knows from decades of experience what will work.
Sports. This is probably the most famous example, thanks to Michael Lewis’s Moneyball. For generations, baseball teams relied on scouts who evaluated players through experience and intuition. Then Billy Beane and the Oakland A’s began systematically exploiting the difference between what traditional baseball experts valued and what actually contributed to winning games.
On-base percentage is the famous example. The A’s recognized that the market undervalued players who were good at getting on base. That insight allowed a relatively poor team to compete with much richer teams by making better decisions about how to spend its money (or at least it did until others learned that secret). Today, quantitative analysis is embedded throughout professional sports, from player recruitment to game strategy. Major League Baseball teams now employ substantial staffs of analysts, data scientists, and quantitative researchers—roles that barely existed in their offices a generation ago.
There are plenty of other examples: digital advertising, revenue management in hotels and airlines, insurance underwriting, freight routing, and more. I’m not trying to write a full history of the trend, just providing enough examples to think concretely about what might come next.
I’d define it this way:
A field goes quant when systematic, data-driven prediction and optimization begin to outperform traditional human judgment and become a primary source of competitive advantage.
I actually prefer the word systematic to quantitative. It’s broader and, I think, more accurate. But “quant” is the term that has entered the popular vocabulary, so I use it too.
In each of these examples, experts had developed systems of judgment based on experience, intuition, and conventional wisdom. Then people armed with data and models began asking whether those judgments were actually right. Sometimes they discovered that conventional wisdom was wrong. Sometimes they found patterns humans couldn’t detect. And sometimes they could simply make decisions more consistently.
The math didn’t have to replace human judgment entirely (though sometimes it did). In some cases, it complemented or constrained human judgment. Sometimes the quants developed an entirely new approach alongside the traditional one; that’s how I would describe what happened in finance. In my mind, the determinant of whether a field has “gone quant” is whether the resulting decisions are, at least in many cases, better.
So what makes an industry susceptible to quantification? I think there are five ingredients:
Capturable information. You need to capture and process information about the inputs to a decision. In baseball, that’s player statistics. In finance, it’s prices, financial statements, and other market data.
A prediction or optimization problem. There needs to be something you’re trying to predict or optimize. Will this player contribute to winning? Will this stock outperform? Which voters should a campaign contact?
Measurable outcomes. You need to know what happened. Did the player perform? Did the investment make money? Did the voter turn out? This allows models to be tested and improved.
Something valuable at stake. Better decisions need to matter economically—or, in politics, in terms of winning elections.
Lots of decisions. It helps to have lots of at-bats. The more decisions you make, the more opportunities you have to test predictions, learn from outcomes, and improve your models.
These ingredients help us think about what might come next.
Historically, quantification has been easiest where the relevant information was already structured: stock prices, baseball statistics, advertising clicks, and voter records. But much of the world operates on information that has been far harder to capture and process—conversations, documents, images, videos, and human interactions.
That’s changing. More of this information is being recorded digitally, and AI can increasingly interpret it at scale. The result is that vastly more information can be incorporated into systematic prediction and optimization. What’s changing is the quantity and variety of information that can be brought into the prediction process.
There’s a second development that’s potentially just as important: generative AI can create new things to test. Instead of simply analyzing existing information, models can generate arguments, videos, designs, and other alternatives that can be evaluated against one another. Experimentation isn’t new, but the range and volume of things that can be generated and tested may expand dramatically.
There’s still a catch. Having information about a decision isn’t the same as knowing whether the decision was right. In finance, sports, and advertising, outcomes are relatively easy to measure. In other fields, they can take years to emerge and be much harder to attribute to any particular decision. AI may be removing a major constraint on the input side of quantification without necessarily removing the constraints on the outcome side.
Which industries are about to discover that the way they’ve organized human judgment is fundamentally inferior to a quantitative alternative? Here are three I think are interesting bets, and I’d love to hear other ideas in the comments.
Trial law strikes me as a promising candidate because so much of it involves prediction. Will a case succeed? How will a judge rule? Which arguments will be persuasive? How much might a jury award? When should a case settle? Today, lawyers answer these questions primarily through experience, research, and judgment.
Imagine preparing to argue before an appellate court. You could practice before AI versions of the actual judges, each informed by that judge’s prior opinions, questions during oral arguments, and other relevant materials. They could challenge your arguments and identify weaknesses. Before submitting a brief, you might have a model of the judge evaluate it. Of course, the value of this approach depends on how well the models actually predict the judges they’re attempting to simulate.
But there’s another dimension that could be even more consequential: litigation finance.
Litigation is expensive. Many potentially meritorious cases are never brought because plaintiffs cannot afford to pursue them, particularly against well-financed defendants. What happens if we get much better at predicting the probability of winning, the likely damages, the cost of pursuing a case, and the time required to resolve it?
Legal claims could increasingly be evaluated and priced like financial assets. That could produce a substantial expansion in litigation financing, allowing investors to fund cases that otherwise would never be pursued. It could also mean more justice by giving people access to legal remedies they couldn’t previously afford. Or perhaps it produces a flood of litigation as the economics of bringing cases improve.
Either way, a more systematic approach to predicting legal outcomes could have enormous consequences.
Entertainment is another industry built around consequential exercises of human judgment. A movie studio might spend $100 million or $200 million producing a film, then another enormous sum marketing it. Will people want to watch it? Will they like the story? Will the actors resonate? Will the ending work? These decisions are made through a combination of creative judgment, experience, audience research, and instinct.
Of course, entertainment has already begun going quant. Look at TikTok and YouTube. Creators publish content, measure audience response, experiment with formats, and adjust what they produce. Distribution algorithms determine which content gets shown to whom, while the resulting audience data influences what creators make next. Creation and distribution feed off one another in an increasingly systematic process.
But what happens when that approach moves into higher-budget entertainment?
Imagine a studio that produces hundreds of relatively inexpensive pilots, measures audience response, and then concentrates its resources on the winners. Or one that uses synthetic audiences to evaluate scripts, casting choices, dialogue, and alternative endings before committing hundreds of millions of dollars.
AI may make it possible to experiment with the creative product itself, not just its distribution. If you can cheaply produce dozens of versions of a scene, character, or ending, you can begin to test creative decisions that historically required enormous commitments of time and money.
I suspect we’re still in the early stages of understanding how systematic approaches will change what gets created, not just how it gets distributed. But there’s a risk here too. Once baseball went quant, teams optimized their way into a game of strikeouts, walks, and home runs that fans found less fun to watch, and MLB eventually had to change the rules. Entertainment optimized for what’s measurable could hit the same wall: more content people click on, less that they remember.
This is probably the most ambitious prediction of the three. There are approximately 11.1 million managers in the United States, earning an average of $145,000 annually. That’s roughly $1.6 trillion a year in management wages alone, according to Bureau of Labor Statistics data.
What do managers do? At the most fundamental level, they decide how to create value given an objective, build an organization of people, processes, and technology to deliver that value, and acquire and manage the inputs that the organization requires. Each of these activities involves consequential decisions about what will work, what won’t, and where to invest resources. Yet most of those decisions still rely heavily on human judgment.
Consider employee performance. Almost everyone has worked with someone the boss thinks is terrific but whose colleagues know isn’t contributing much. The reverse happens too: highly productive people who don’t advertise their accomplishments get overlooked. Managers are subject to all sorts of biases, and they typically observe only a fraction of what their employees actually do.
Today, much more of that work is observable. Meetings can be recorded, communications are digital, and work products can be evaluated. AI makes it possible to process much more of that information systematically. The interesting opportunity isn’t having AI write a better performance review. It’s developing models that can actually predict which employees will perform well, which deserve promotion, and which are likely to succeed in different roles. You could then imagine taking another step and developing a customized coach for employees that helps them develop in the most important areas.
Or consider how companies allocate resources. Imagine a business with 50 potential projects but enough capital and people to pursue only 10. Today, that decision is often made through executive presentations, financial projections, and the judgment of senior managers. What if models could systematically predict which projects were most likely to succeed, estimate their expected returns, and recommend the best portfolio given the company’s constraints? The opportunity isn’t to automate the budgeting process. It’s to make better investment decisions about what the company should actually do.
Of course, managing employee performance and choosing where to allocate resources are difficult. Outcomes can take years to emerge, and it’s often hard to separate individual contributions from those of the team. There’s also the uncomfortable prospect of having an algorithm evaluate your performance, although human managers are hardly free of bias themselves. Better measurement doesn’t automatically mean fairer measurement.
By the framework above, management is the hardest of the three: decisions are fewer, outcomes take longer to show up, and they’re harder to attribute. That’s exactly why I think it’s the most interesting bet. If someone cracks it, the prize isn’t a better tool for managers; it’s a different kind of company.
In every industry that has gone quant, the experts weren’t wrong about everything. They were wrong about something specific and valuable, and they didn’t know which thing it was.
AI is expanding the information the quants can use and the alternatives they can test, which means more industries are about to face the same question. Lawyers, studio executives, and managers should all be asking it now, before someone else answers it for them: what’s our on-base percentage?
I’d love to hear from readers: What companies are already building in these areas? And what other industries do you think are next to go quant? Post in the comments.