Firms invest in new strategies, campaigns, and operational changes constantly—but without a clear view of what’s working and why, they risk misallocating resources. When you make a decision for your firm, you want to know how it’s going to change your bottom line. Causal inference offers the tools to isolate impact and measure what matters.
What is causal inference?
Imagine a retail chain launches a new pricing strategy. A month later, revenue is up 10%. Was it the pricing change? Seasonal demand? A competitor’s exit?
Relying on correlation can lead to repeating ineffective strategies or abandoning promising ones. Causal inference offers a structured approach to disentangle overlapping factors, identify the true drivers of change, and quantify their effects.
Causal inference is all about estimating a “counterfactual,” i.e. understanding what would have happened without our intervention of interest. In our retail example above, the counterfactual we want to estimate is “what would sales have been in absence of the pricing change?”
So how do you estimate the counterfactual?
If you can, you should run an experiment. But unlike in the natural sciences, where practitioners can run tests in controlled environments with hundreds of samples, this is often very difficult in a business setting. If you want to test a radio advertising campaign, you have little control over who hears it and who doesn’t. If you want to experiment with a pricing discount at one brick-and-mortar location, that location might have some qualities that make it a bad comparison to your other locations. If you want to run a geographic lift test (“geo-test”) with a marketing campaign, there are special considerations you need to take to actually detect an effect.
Sometimes it’s not just hard, but impossible to run an experiment. You may only have one location. You may not want to experiment with lowering prices, especially if your margins are already thin. It may be hard to get buy-in to shake things up from other stakeholders in the firm. It may even be unethical to run an experiment, which is of particular concern for companies that work in public health, education, and medical spaces.
How does King Street Economics approach these problems?
Economists are used to these problems. You can’t run experiments on an economy, or pass laws just to see what will happen, but understanding cause and effect in these settings is exactly why economists exist. Sometimes you can run experiments with organizations, but it’s difficult to randomize and there are few units that could serve as a test group. We developed methods like synthetic controls, differences-in-differences estimation, matching, regression discontinuity design, double machine learning, and more to answer causal questions in non-experimental settings.
These tools have been immensely useful in the private sector, too. Today, Amazon employs more economists than anyone except for the Federal Reserve, and many other large firms (including Walmart, Apple, and Wayfair) employ economists to study how to better market, price, and run their business.
Your firm probably isn’t a hundred-billion dollar corporation (if it is, don’t worry, we’d still love to have you as a client), and if so, it also probably doesn’t make sense to have even one full time economist, let alone a full team. King Street Economics exists to provide the services to companies like yours, letting you buy the same expertise in causal inference, analytics, and economics that the big guys use, on demand, in exactly the amounts you need. If this sounds like a good fit for you, please reach out to [email protected].
