Before any machine learning or AI or, more generally, financial analytics exercise, we always want to have a process for how we attack a problem or question. I think the most tried and true method is what most people refer to as the scientific method. I love this quote by Richard Feynman, the great physicist, which summarizes the scientific method very succinctly and eloquently. If it, and by it, he meant theory, disagrees with experiment, it's wrong. That simple statement is the key to science, and that characterization of the scientific approach to tackling problems, I think, is incredibly useful for tackling business problems as much as it is for tackling scientific problems. Let's be a little more explicit here. What is the scientific method? There's many ways to describe it. The one I find quite useful is the following. First, you start by clearly articulating a specific question. What problem are you trying to solve? Then you guess an answer. Statisticians would say you hypothesize an answer. Same idea. You come up with some potential answers or hypotheses, solutions, proposed solutions to the problem . Step 3, we want to then identify the empirical implications of those hypotheses or those guesses to the answers to the question. In other words, if the answer or hypothesis is correct, what should we see in the data? Then finally, we want to compare the empirical implications of our guesses with what we actually see in the data. In other words, if our hypothesis is correct, this is what we should see in the data, but do we actually see it in the data? That's really the last step. Let's consider an example and apply the scientific method. Here's a question one might have, why is our revenue growth slowing? Common question many business leaders might ask of their company? Now, seems like a clear specific question, but, in fact, I have a lot of questions about that question. It's not quite as clear as you might think if we want to go to the data to try and extract an answer to that question. For example, what revenue? Are we talking about consolidated, gross, net of returns, discounts, etc? Over what time period? Are we talking about revenue growth just over the last year, over the last two years, five years ? If we want to take to the data, so to speak, this question, we need to be very specific and clear about the question. Let's imagine that what we're really interested in here is why is revenue growth over the last three years? Apologies for typing. Let's say it's consolidated at the aggregate level, not for a particular product. Why is consolidated revenue growth over the last three years slowing? Eventually, I'll learn how to type time here. This is our question. Remember that step 2 was to guess an answer. What are some potential answers? In fact, pause the video and try and come up with some answers. The fact that you don't know what specific company we're discussing here is not relevant. Just use your general business sense and understanding of what drives revenue, a very general level to try and come up with some answers to this question. Hopefully you're back. If you came up with some, what are some potential answers here? Some hypotheses as to why revenue growth is slowing. Maybe one thing that's going on is that product demand is slowing. Maybe something else that could be going on is, actually demand is not slowing, but there's a supply problem. There's a lot of demand for our products, but we just can't get product to the shelf for some reason. Maybe what's going on is there's been a change in pricing strategy. Remember, revenue is price times quantity. A reduction in revenue over time or a slowing of growth, more precisely, could come through either the quantity or price channel. There's a few hypotheses that we can test. Then the third step was to identify the empirical implications of our guesses. Well, the empirical implications of these different hypotheses are actually relatively straightforward. If there's a change in pricing strategy, we can look at the data, at the pricing of our products, to see if we have been reducing prices perhaps too aggressively to maintain demand. We should see reductions in product prices. Maybe what's going on is that instead, we're seeing a slowdown in products sold, quantities. What's interesting is, this is the same implication we would see for a supply problem. Both will affect quantities. If it turns out to be the case that we're just selling less units, we're going to have to probe a little bit more deeply later on. But there are some potential hypotheses behind our revenue growth slowing question as well as some empirical implications of those hypotheses. Then the last step is to compare the implications with the data. That's our last step. We'll have to, actually, look at some data on product sales, on product prices to see what or which, if any, of these hypotheses are actually playing true in the data. Now, that's a very stylized example, but hopefully, an illustrative one to emphasize the importance of, One, clearly articulating a specific question that you can take to the data. Two, guessing some answers to the question or hypothesizing. Three, identifying the empirical implications of those hypotheses. That is what we should see in the data. Four, eventually taking those implications to the data to see which, if any, are in fact true.