[MUSIC] The 4th consideration for managing a successful ML project is of course your team and expertise. By a successful ML team here, I'm referring to the collective set of expertise that you'll need to carry out a project end to end. It's important to consider all possibilities for obtaining the right expertise, which could be done either in-house or through an extended partnership. So think about your team and the expertise you have and the ones you'll need to get. What skills will you need to build, borrow, or buy? Let me start by going over the skills you'll need and then I'll cover how you can go about acquiring them. First, focus on data science expertise. The three most important roles are professional data engineers to build pipelines that routinely ingest and transform data. ML engineers who build predictive models using curated data. And data analysts to collect, curate, and explore data opportunities further. They'd build dashboards and create reports to support decision making throughout the org. This is not something you'd want the ML engineers to do. Ultimately, your team as a whole should know how to build end-to-end machine learning solutions to real world problems. Let's look at a couple of examples of what your team structure could look like. I'll show you a team structure for a large organization and another first smaller organization. Keep in mind these are just to give you ideas, but they may not necessarily reflect your specific needs. I'll start with a large organization. You will have lots of data analysts. These are the domain experts who know your business. You will have fewer data engineers and you will have only a handful of machine learning engineers. This is assuming you were putting together a data science team of about 60 people. In this figure, colored shapes represent tech leads. Gray circles are individual contributors. Transparent boxes represent managers whose primary function no longer involves coding. In this case, the vice president of the whole team and the manager for the data analysts are purely managers. It is expected that you'll have three times as many data engineers as engineers focused on machine learning, and twice as many data analysts as data engineers. This is because you need a lot of domain experts. And they're exactly that. You also need the ability to build and maintain data pipelines, hence the data engineers. On the other hand, machine learning itself is mostly just the application of standard algorithms to data. So you need far fewer of this skill. Now, if you're a startup, for example, your team will inevitably be smaller. There will be more overlap between roles, but the approximate ratios will stay the same. All your people will also be technical with no pure managers. The director of the team could have as their core strength data engineering or domain expertise. Here's a comparison of the two teams side by side. If one of these teams is the team structure you want to build, how do you get there? Your first instinct might be to go out and hire everyone. Let me offer you some alternatives to help you bring together the expertise you need. Take a look at the expertise within your organization now. Determine which of these three skills we just talked about you'll need to buy, which ones you can borrow, and which ones you can build. I'll explain what I mean by each one in turn. When I say by skills, what I'm referring to is hiring the talent for your team. Before hiring a machine learning team, there are two factors to consider here. The first one is that you want your data scientists to have the domain knowledge relative to your business. This is the only way they will recognize what kinds of labels exist. What kind of proxy features they can use, and what objectives will have problematic side affects. Second is that abstraction levels in machine learning are increasing. What this means is that due to the availability of common ML models and their advancements. You no longer need to have a PhD in machine learning to train, for instance, and image classification model. What you want are people with a strong understanding of your domain and a good sense of statistics. They should know how to evaluate an ML model and all the different levers that they can use that affect model performance. Think about the problems you're trying to solve with ML and then align the expectations with potential job descriptions. This brings us to the second point about borrowing expertise. Most common use cases don't require ML experts, so I always recommend exhausting alternatives. For example, your existing information technology and business analyst teams can be up skilled. The IT folks can learn how to use AutoML, the business analyst teams can learn how to use BigQuery ML. Together, these will solve the vast majority of your ML problems. That said, they will have to learn the basics of machine learning in order to use the powerful technology effectively. So unless you're in a research area such as self-driving cars, you don't need a dedicated machine learning team. When hiring machine learning engineers, make sure to differentiate between people who wish to do machine learning research versus those who wish to apply machine learning to solve business problems. For your ML teams, hire people who are strong programmers who will mostly reuse existing frameworks and libraries but are comfortable with their ambiguity inherent in data science. This last bit is very important. A lot of programmers like things that are deterministic, but machine learning models are never perfect. It is experimental and you won't know if something is going to work unless you try it out. And finally, by build, I'm referring to new skills you can acquire. Right now you're taking this course from a learning platform. Take advantage of it for the rest of your team. At Google Cloud, we offer training for all three of the roles we discussed, so you can upscale your existing team members. Check out cloud.google.com/training/data/ml for several courses to deepen your knowledge beyond this course. In fact, for specific roles such as data engineering, Google Cloud also offers professional certifications. I highly recommend checking out our website to learn more about the learning path that best suit your needs. When it comes to deep machine learning expertise, you can again have the option to borrow. This means that instead of hiring new people, you can use machine learning consultants instead. I've seen this approach used successfully in many organizations. Google Cloud's professional services organization can help you build ML models. We also have a number of Google Cloud partners specialized in machine learning as well. They can help you build ML models and integrate them with the rest of your systems. And that's it. In this video, we covered a lot of great tips for how you can build successful ML teams for your projects. In the next video, I'll talk about the importance of creating a culture of innovation to ensure your ML projects are successful.