Welcome back. In this lesson, I'm going to give you a real-world example of some questions to answer. At first, it might seem frustrating to get a question that doesn't necessarily have a right answer. But by the end, you should be able to understand the problem and present not just the data, but insight. So let's suppose you get asked some or all of these questions. Let's go through an example where you get asked the SQL question. Let's suppose you get asked some or all of these questions. How many times do users re-order a specific item? How many times do users re-order an item from a specific category? How long between orders or re-orders? So, we could go through the process of scaffolding each of these queries, and in the exercise I'll walk you through some of them. But what I want you to be asking back is, why do you want to know? These aren't maybe exactly the words and tone I would use with my boss, but my point is that you should understand the context. So imagine, I'll give you some context, imagine that the real reason you're being asked these questions is that, we are considering a few ways to improve recommendations. One idea is to suggest that users re-order items that they've previously purchased. There are a lot of variations of this idea, and none have been decided on yet. Maybe we could also recommend items from the same category of items. Maybe we should do that while the items are in your cart, or maybe we should do that after you finish an order. We could scrap that plan and focus on recommending our bestsellers instead to. Someone is asking you how often users re-order an item because they want to know if it's an important behavior. So imagine if I told you that one percent of users re-ordered an item. Would that help you figure out if this is a good idea? How are you even supposed to feel about that? Is it a lot? I just made that number up, but we'll have to run some actual queries to see what the numbers are. My point here is that some context is missing. It seems like what we really want to do is to compare these numbers to something. I'm much better at telling you which number is bigger than telling you how to feel. So let's turn these questions into some slightly better questions. First, let's think about the size of the audience for this recommendation. The number of people who could ever re-order is somewhere between the number of people who do re-order and the number of users who've ever ordered anything. Then, let's think about the impact. Is there evidence that previous purchases affects your chance of ordering an item in the future? And depending on your role, depending on the situation, you may have the opportunity to turn this into something of a more exploratory analysis. For something like this, we are essentially scoping different project ideas, and you might choose to scope additional variations of the project. Here's some questions you could ask. Perhaps we could recommend adding items from the same category. If you're in the middle of buying an industrial strength contraption, perhaps you also want to buy a fuzzy contraption warmer as part of the same order. What's the audience size for that? Or what if categories and previous purchases don't look like important factors at all compared to straight up popularity? We should find a way to ask the question so we can compare these results to the audience size and the impact, and compare that to results from our previous analysis. So to recap, here's the analysis framework that I'm suggesting. Figure out what decisions are being made, ask for context, see if you can refine the data questions to be more helpful. This can really save you time by preempting follow-up questions. We haven't also really talked about this. But as part of the refinement process, think about what assumptions you're making and whether that interferes with your ability to answer the question, and then finally present the data in a way that clearly informs the decision. Communicate both the results and the level of certainty. Here are some questions related to this scenario that don't exactly have right answers, but I'd like for you to think about them and discuss.
