Welcome to Module 7. Now that you know the fundamentals of machine learning in this module, I'm going to focus on what you'll need to know to manage a machine learning project. You're already familiar with the various phases within an ML project, and the key considerations within each phase. Most importantly, though, you probably now know that this process is not linear. The end to end process is often iterative. To help you manage a project successfully, I'll cover a few considerations that extend beyond what you've already discussed. First, I'll provide you with a few questions to help you identify the business value for ML. Then I'll cover, how to formulate a data strategy. Next, I'll discuss how you can develop data governance for ML. Then, how to build successful ML teams and ways to create a culture of innovation. Finally, we'll complete this module with a hands-on lab, where you will evaluate an ML model using BigQuery ML. Let's begin. [MUSIC] To ensure you have a comprehensive view of managing an ML project, I've outlined a few areas that we'll discuss further. Business value, data strategy, governance team or the overall expertise, and culture or the individual and team mindset. The list of considerations here is not a comprehensive one, but the topics we'll cover will give you high level view of what to look for as you begin working on your own ML project. The first consideration, business value, is a combination of factors we've already discussed up to this point. Evaluating the cost and benefit of a project requires additional time and expertise that exceed the scope of this course. I'll offer you a few questions that you can use as a guide for when you're assessing the business value of an ML use case. First, what is the current gap? How is the gap impacting the organization? The answer to these questions should give you a sense of what the current company status is. Second, what would happen if we did nothing? This should help you prioritize the problem. Third, how would solving this problem improve or benefit business, customers or people in general? This is the benefit analysis. Remember, ML usage should be done responsibly and ethically. The goal is to ensure your ML use case doesn't inadvertently harm one or more groups. How would you classify the project? Is it a quick win, long-term development or full transformation? This refers to the value added and estimated impact. If it's a full transformation, for example, it will have the highest value, and it might require the most resources. Next, what is the estimated cost of this project? It may take additional experts to help you identify an estimate. Finally, do we have any existing budget, expertise, and our leadership support? By answering this question, you're establishing the resources you'll need and the strategy for obtaining buying. Having leadership support is crucial for any ML project. The next consideration is formulating the data strategy for your ML use case. Before we talk about formulating a data strategy, let's define what a data strategy is. A data strategy refers to data you have and data you'll need in your ML project, and the method through which you'll collect and prepare the required data. So, how do we begin to formulate a data strategy? As I mentioned in an earlier module, start by examining your available data. Machine learning only works when you have a lot of data. Remember this graph that shows the decrease in error for a machine learning model as the size of the data set increases. The error decrease is linear, but the required increase in data sizes exponential. The first principle for success with machine learning is to look at your data with this insight. It's not enough to have 10% more data, you want twice as much more data. Let's say you run a boat company. You make some good boats, but now you want to build better boats. How would you learn how to build a better boat? If you're going to learn with ML, you have to collect a lot of data. For example, you'll need data about the type of water that your boats are used on. Is it lakes or oceans, or some other body of water? Ways that your customers transport their boats? Do they use trucks? Who your customers are, and their purchasing power, would you classify them as middle-class or on the wealthier side? What type of boat are you considering? Powerboats or kayaks? Who wants to buy your boats but doesn't or can't? Who are your suppliers of marine equipment, and how does all of this data change overtime? You get the idea. The point is, you wouldn't start by building a better boat, instead, you start by collecting data about your business as it pertains to boats. If your company does OKRs, meaning objectives and key results, one of your OKRs must be around building toward a future, where every part of your business is measured and measured obsessively. To help you take this example one step further, I'll talk about the seven pillars of a data strategy in the next video. Think of these as the guiding principles as you formulate your own data strategy.
