Experiments are the gold standard for figuring out what actually causes what. The logic is straightforward. Randomly assign who gets the treatment, control everything else, and the data tells you whether the effect is real. No prior belief required. No guesswork about competing explanations. Just evidence.
Companies like Amazon and Booking.com have built entire decision-making cultures around this idea. Run the test. Read the result. Move on. It works remarkably well for the class of decisions it fits — product changes, pricing tweaks, marketing campaigns. When the intervention is reversible, affordable, and fast, experimentation is exactly the right tool.
But consider an acquisition.
You have identified a target. The strategic logic is compelling. The revenue projections look strong. Your team has a clear thesis about why this business, combined with yours, will produce something neither could produce alone.
You cannot run an experiment. You cannot acquire half the company and see what happens. You cannot randomize a subset of your customers to the post-acquisition world and measure the effect. By the time any evidence exists, you have already signed the documents, paid the price, and begun the integration. The bet is made.
And yet the questions you need to answer are exactly the causal ones. Will the customer bases actually combine the way the model assumes? Will the sales team execute in a structure they have never operated in? Will the competitor respond in a way that changes the strategic logic entirely? These are not prediction questions. They are questions about mechanisms — about what will actually cause the outcome you are projecting.
Experiments would answer them perfectly. You just cannot run one.
The answer isn't more data. It's a model of the mechanisms — one that makes your assumptions explicit before you act on them.
← Back to Writing