So far you've learned about standard algorithms and data quality and quantity for successfully using ML. The next core component I'll cover has to do with predictive insights. Think about your business or your day to day work. How do you use data today? Do use a dashboard that was set up by analyst or engineers. Maybe you review automated static reports on a weekly or monthly basis. Both the dashboard and the report are examples of backward looking data. Most analytics in your business are probably backward looking where you look at historical data to calculate metrics or identify trends. To create value in a business you need to use the data to make decisions. It's not enough to identify a trend, you need to be able to make decisions based on that trend. Think about this example. A business analyst reviews a report an sees that demand is increasing for a specific product in a specific region. The analyst then suggest a new price for that product in that region that would increase profits. Now the business analyst is making a predictive insight. But is that scalable? Can that business analyst make a similar decision for every product in every region? Can they dynamically adjust the price every second based on how many people want that item at every instant. In order to make repeatable decisions using predictive insights, you need machine learning. You need a computer to help you derive those insights. So machine learning is a tool that enables your business to make many predicted decisions from data. It's about scaling business intelligence and decision making. What do I mean by predictive exactly? Traditional business intelligence dashboards and data analytics describe historical data. Predictive analytics is focused on the unknown. Whether it's the future, a new product or a new consumer experience with predictive analytics, you gain an estimate of the likelihood of something you didn't know before. Now let's look at the last part of our amount definition, which is focused on making repeat decisions. As you think about your own ML use cases, I want you to remember this part of the definition because it's often where you will get the most business value. ML works well when making repeated decisions. In fact, many of the benefits of ML, including scalability and efficiency, are only realized when use it repeatedly. Let's look for the repeat decisions in our manufacturing example. Does it apply? Yes, you're making decisions about which parts to keep and which ones to remove from your assembly line. Aah, and ML wouldn't be very useful if you're only classifying one part. The greater value in this scenario comes from classifying multiple parts every minute. Let's look at a few other use cases and determine whether they are right for ML. We've been looking at quality control of individual parts of an assembly line. But what about recommending the right product to your customers throughout their exploration of your catalog of parts? Or continuously predicting the moment to moment price of commodities. Yes, you guessed it. All of these examples involve repeat decisions at scale, which make them good choices for ML. Let's try another set. What about something like quality control of a space shuttle? The answer is no. There has been less than 200 shuttle launches since 1977, and only a handful per year between 1977 and 2011. So we probably won't be making that many decisions or have that many examples to feed a potential amount model. What about an annual sales forecast? Most businesses do that. This one is tricky because of the word forecast. Because sales forecasting happens only once a year, this is not an ideal use case for ML. There are many statistical analysis tools that we can use instead. What about trying to model the likelihood that a truck might breakdown? At first you might think trucks don't breakdown that often, so this isn't a good use case for ML. But what if I told you that everyday mechanics decide whether they should inspect vehicles for issues to find whether the truck is likely to breakdown? Then it becomes a frequent decision an if we have enough examples of breakdowns, we can solve the problem with ML. And remember 2 important points. The magic of a ML doesn't come from looking for cost savings. It comes from using data to make repeat decisions at scale. Two, look for use cases where these key ingredients come together, then use ML to transform the way you work, interact with your customers and ultimately your business. I hope going over these core ingredients for ML has already triggered some ideas for you and you're starting to re imagine what's possible. I'll cover a few more examples in the next video to give you some more ideas.
