In this lecture, we're going to talk about ML and business processes. When we think about how are we going to get from a no ML to ML Solution in our organization, that's path. I really want us to think about the evolution of a business process. Here when I say a business process I'm talking about any set of activities a company must do directly or indirectly to serve customers. Organizations must continually improve these processes, which they do through a feedback loop and this is really critical. Oftentimes when we think of business processes, we forget that almost every one of them has a feedback loop and understanding that loop becomes really important to understanding the role ML plays inside a large organization. Let me give you a concrete example. Let's say we have a call center and the call center takes customer questions and then produces answered questions. You call your favorite local telecom and you say, there's a problem with my Wi-Fi router, alright, the call center answers that for you, but the interaction doesn't end there. We all know that after, the call center hangs up, they say, would you like to answer the survey or they email you? What they do, if you actually answer the surveys is they take all those answers, and they extract new insights from them. They get new canned answers, new product promotions or decision trees on how to handle future calls and then those new insights, they get fed into HR or training reference manuals and then employees are retrained and those retrained employees may now go answer new customer questions. This is this feedback loop that allows us to convert operational expertise and to better future outcomes. Here I'm showing you a more general view of this. Here we have some input and some output to a process. If this looks similar to our example of what ML is compared to software, that's not an accident. Now what we do is we take this output and we generate new insights from it. These insights are going to give us some new operational parameters, such as a new canned answer or a new product promotion and then we're going to tune the original process with updated instructions. Here again, we have this flow within a much more general way, that could be applied to almost any business process in any organization. When we think about the path to ML, I want you to think about how we're going to automate each one of these boxes, each one of these cornerstones in the business process tuning. In the first step before anything and I'm going to explain more of these steps on later slide, but the first step of any new business processes, just individual contributor, one person doing it, and you get multiple people doing it, and then you digitize that process and these steps: 1, 2, and 3 they all affect the core process itself. But as you know, in the last decade or so, big data and analytics, and machine learning has become very popular and very impactful. What happens there is we're trying to automate the insight generation phases and the tuning phases thus, we have automated the entire feedback loop. I'm going to go ahead and define what I mean by these in a little bit more detail. Business processes that eventually end in ML typically go through five phases. Now they don't have to spend the same amount of time in each of these phases, but skipping these phases, as we'll see later, usually isn't a good idea. The first one, what do I mean when I say individual contributor? A task or business process that's in the individual contributor phase is performed by a single person and a great example is like a receptionist inside the office building. This person can answer the phone, maybe points people towards the bathroom, just one of them and the task is not paralyzed or scaled at all. Usually it's very informal. Then what happens is as the business process becomes more important to the company, usually we start to delegate we get multiple people who are all performing the same task in parallel and a good example would be like a store checker. What happens when we start to delegate is we have to start to formalize the role and like put in rules so that each store checker starts to behave a little bit more like the others. There's some repeatability in a task. Then we get to digitization, a little bit of a marketing buzzword. What I mean is that we take the core repeatable part of a task or business process and we automate it with computers. Great examples is an ATM. ATMs can't do everything, right if you can't open a mortgage through an ATM, but you can withdraw cash and because that cash withdrawal part of that business process where the interaction with the user is so repeatable and so well automated, customers get a very high quality of service using ATMs. How many of us actually would walk into a bank to extract $40? Almost no one. But after we digitize what happens next? Now we move into big data and analytics. The idea here is, we're going to use a lot of data to build operational and user insights. So maybe when i say operational. A good example would be like Toyota manufacturing. Toyota is famous for their lean manufacturing philosophy with a kind of measure everything about their construction of their facilities, and then they use that to tune each little knob in the process to get better and better outcomes and faster and faster time to delivery. You could do this for your internal operations or you could do this to learn about your external users. This would be like marketing research on steroids. And then of course we get to machine learning, which we're going to represent as the last phase and the path to ML. Here we're going to do is we're going to use all these data that we had from the previous step, but we're going to automatically start to improve these computer processes. A big example here is YouTube recommendations. As you click through YouTube and you watch different videos and you like them or you don't like them, or you watch to the end or not. The algorithm is learning in the background what are good videos, what kind of videos you like how you are different or similar to other users. What I want to do is we think about this path to ML is I want you to take a moment. I want you to sketch this diagram for a specific example from your organization. It doesn't have to be an ML example. Maybe you have digitized part of this business process, but not all of it. What phase of the path to ML is your example in? Do you have another example two, that's in a different phase.