For James Sun, CEO of Kapnova, quantitative thinking starts with a simple question: what would have happened if a business had not made a particular decision?
“I start with a very simple question: what would have happened if we had not done the thing?” Sun says. Correlation tells you that two things moved together, while causation asks whether changing one actually changed the other. In business, where many things happen at the same time, that distinction can be difficult to establish. “If sales rise after we increase marketing, it is tempting to credit the marketing. But maybe it was Black Friday, a competitor went out of stock, our reviews improved, or demand was already rising.”
Sun says he tries to identify the intervention, the outcome, what else could have caused the outcome, and what a credible counterfactual looks like. But he also says that businesses do not always have enough evidence to establish causality. “Sometimes the right answer is, ‘We see a relationship, but we need an experiment before we can say it caused the outcome.’”
Discounting is one example of how quantitative analysis can challenge an obvious conclusion. A company may see that promotions consistently increase unit sales and conclude that promotions work. But over a longer time horizon, customers may begin waiting for discounts, the reference price can change, and full-price demand can decline. A promotion that looks profitable over one week could destroy value over six or twelve months. “That changes the decision from, ‘Do promotions work?’ to, ‘What promotion depth and frequency maximize total contribution over time?’”
For Sun, this is what makes quantitative thinking valuable. “Frequently it does not tell you that your intuition was completely wrong. It tells you that the question you were asking was too simple.”
He also distinguishes being quantitative from simply being good with numbers. “Being quantitative is not about being good at Excel or being able to calculate something quickly. It is about how you think about uncertainty.” That means asking what assumptions are being made, what else could explain an outcome, how confident you are, what evidence would change your mind, and whether you are measuring what you actually care about.
“The best quantitative people I have worked with are often less certain, not more certain,” Sun says. “They understand that a number can look incredibly precise and still be wrong because the model behind it is wrong.” To Sun, quantitative thinking means being disciplined about what you know, what you think you know, and what you still need to test.
Establishing causality is particularly difficult in business because “businesses are not laboratories.” Companies can change price while changing advertising, while competitors launch products, inventory changes, consumer sentiment shifts and the economy changes. Management can also make decisions based on information that may never appear in the dataset.
The most important number, Sun says, is one that cannot be directly observed: what would have happened if a different decision had been made. “That is the counterfactual. Causal methods are essentially different ways of trying to estimate that missing world.” Sometimes historical variation or a natural experiment can provide enough information. Other times, the right answer is to run an experiment. “In that case, we would rather recommend an experiment…perhaps across geographies, customers or products to create a credible counterfactual than pretend the historical data proves something it does not.”
As a founder, Sun does not see intuition as a bad thing. “I don’t think intuition and quantitative evidence are opposites.” He describes intuition as accumulated experience compressed into a judgment, but says the problem comes when intuition is treated as fact. “I think about intuition almost like a starting probability. It gives me an initial view. Then the data should update that view.”
Kapnova itself provided an example of this process. Sun initially thought the company should lead heavily with causal inference because it was one of its most differentiated capabilities. But after spending more time with operators and mapping the decisions they actually make, his perspective changed. “Executives do not wake up in the morning wanting causal inference. They want to know which decision will create more revenue or profit.”
That changed how Kapnova was built and positioned. “Causal science is still underneath the product, but we now start with the economics of the business, identify the decisions that can materially change those economics, rank what is worth investigating, and then apply the appropriate models.” The lesson, Sun says, was that “the technology should serve the decision, rather than making the decision serve the technology.”
For someone starting from scratch, Sun recommends learning probability and basic statistics, regression, experimental design, causal inference and the counterfactual. He would also study economics before spending significant time on machine learning and AI.
“The tools are becoming easier every year. The harder skill is framing the right question.” AI can now write code, clean data and run sophisticated models quickly, but “if you ask the wrong question, control for the wrong variable or mistake prediction for causation, faster computation just gives you the wrong answer faster.”
“That is why I think causal reasoning is becoming more important in the AI era, not less.”