How Healthcare Led the Way in Causal Inference (And The Sector That’s Next)

Causal inference is steadily becoming a standard part of data science practice. Some industries are slow to adapt, while others are almost forced to move ahead of the curve. In the case of causal inference, healthcare has long been the sector that quietly led the way. Not by coincidence, and certainly not without reason. The real question is: why has healthcare, out of all industries, been so far ahead? And just as importantly, which sector will be next to follow?

Healthcare’s Fundamental Need for Causal Inference

The fundamental goal of healthcare has always been to make people healthier. But long before modern medicine, it was already clear that simply noticing associations wasn’t enough to achieve this goal.

Seeing that patients who took a certain herb seemed to recover faster than others did not mean the herb caused their recovery. Sure, it might make people suspect the herb has a causal effect on recovering, but healthcare could never afford to act based on suspicions alone. In this field, wrong decisions do not just cost money or reputation. They cost human lives.

Very early on, healthcare recognized this and began adopting Randomized Controlled Trials (RCTs) on a wide scale.

RCTs offered something that patterns in observational data could never guarantee: the ability to isolate causation from association. By randomly assigning treatments to patients, researchers could ensure that the groups being compared were statistically similar in every way except for the treatment itself. This meant that differences in outcomes could be attributed to the treatment rather than to underlying differences between patients. It provided the industry with something it desperately needed: a reasonable level of certainty about what actually works and what does not.

Over time, this experimental mindset became deeply rooted in healthcare. It changed how treatments were evaluated, how therapies were approved, and how medical knowledge advanced. Clinical trials became the gold standard not just for new drugs, but also for surgical procedures, prevention programs, and even for public health interventions like vaccination campaigns.

However, there are many questions in healthcare where running a randomized trial is not possible, either for ethical reasons or practical limitations. You cannot, for instance, randomly assign people to smoke for a lifetime just to study its effects, nor can you randomly expose people to high levels of air pollution.

In such cases, healthcare did not abandon its need for causality. Instead, it adapted. The field began to adopt sophisticated methods for working with observational data. Techniques such as propensity score matching, instrumental variables, regression discontinuity designs, and modern causal graphs were increasingly used to isolate causation from association even outside of a randomized setting.

Healthcare today uses causal inference across many areas. It is used to evaluate the long-term effects of medications after they reach the market. It helps determine whether early interventions can prevent chronic diseases later in life. It informs guidelines for diet and lifestyle changes based on observational cohorts like the Framingham Heart Study. It even plays a role in policy decisions, such as understanding the impact of new health regulations or insurance reforms on patient outcomes. Causal Inference is the way how evidence is gathered, how risks and benefits are weighed, and how life-saving decisions are made every day.

Why are other industries behind?

Outside of healthcare, many sectors have long been far more forgiving. A new supply chain strategy would be rolled out, profits would increase, and success would be claimed without looking too deeply into the true drivers behind the results. Financial institutions would tweak credit models and assume falling default rates meant their adjustments worked. Retailers would change pricing strategies based on historical patterns, rarely putting efforts to understand whether their interventions genuinely caused improvements or simply coincided with favorable market trends.

For a while, this approach might have been sufficient. The risks of being wrong were often manageable, and systems were relatively stable. Acting on associations worked well enough to help a company grow, because the fact that you were using data was on itself already something that gave an advantage over competitors. But this tolerance is starting to fade. Markets are becoming more volatile. Competition is more intense. Margins are thinner. Everyone has data. In today’s environment, the price of settling for associations instead of causation is rising quickly.

In many fields, there is a growing discomfort; a sense that past methods are no longer enough. That understanding associations is not the same as controlling and manipulating outcomes. Slowly but surely, the need for causal inference is expanding beyond the boundaries of healthcare.

The question now is: which sector will be the next to embrace causal inference at scale?

Who’s next?

At the 2024 Causal AI conference hosted by CausaLens, this exact question was raised to Judea Pearl: which sector would be next to adopt causal inference at scale? His answer was something along the lines of “healthcare – oh wait, no, healthcare is already far – then marketing!” And honestly, I couldn’t agree more.

There are a few reasons why I think marketing is the next in line.

First, marketing is causal in almost every aspect. Whenever a company launches a campaign, changes pricing, tweaks a product description, or tries out some rewarding program, it is performing an intervention. When businesses ask how much each channel (like paid ads, social media, or email) contributed to a conversion, they are asking a counterfactual causal question. They want to know how much the outcome would have changed if a particular channel had not been present.

This causal nature of marketing is already implicitly recognized in the field through practices that are called driver analysis, root cause analysis and uplift modeling. But contradictory as it seems, these approaches have always been purely association-focused, and association-focused methods are fundamentally unable to identify causal relationships.

The lack of true causal understanding has led to strategies that often perform less reliably than hoped. And note the word hoped, because association-focused models do not help us understand the effect our decisions have on outcomes. Any decision made using these models is, at best, a hopeful guess. Applying real causal inference methods here could completely change how marketing performance is evaluated and optimized.

As a second argument, marketing is everywhere. Every company does marketing. Whether it is a global brand managing dozens of channels or a small business running localized campaigns, marketing decisions are always present. And competition in marketing is fiercer than ever. In highly saturated markets, simply reacting to surface-level associations based on gut feelings and hope is no longer enough. Businesses need to know with far more precision which actions genuinely affect customer behavior, and where they are wasting money.

A third reason is that marketing already operates in very data-rich environments. Massive amounts of information are collected about customer interactions across web visits, mobile apps, email opens, ad clicks, purchases, and more. Since observational data alone will never reveal causal effects, causal inference is needed to map this abundance of data into truly meaningful recommendations.

And indeed, we already see this shift happening. If you look at where most causal inference consultancies, tools, and software companies are positioning themselves, it is often in marketing and customer analytics. The demand is growing, and these businesses are sensing it.

However, I believe that over time, causal inference will become so fundamental to marketing that companies will not want to rely heavily on external vendors for it. Instead, I expect that more in-house data science teams will invest in training and educating themselves in causal inference. It will become an essential internal capability rather than something outsourced. Not because external companies do not provide value, but because understanding the causal mechanisms behind customer behavior is simply too central to leave in the hands of others.

Conclusions

Causal inference is no longer a luxury reserved for healthcare or academia. It is starting to make its way into core business processes, and my guess is that marketing will be the first field for widespread adoption.