Matt Taddy is the chief economist for Amazon’s North America Consumer organization. Taddy’s organization develops solutions that automate and accelerate how Amazon makes decisions that improve the customer experience across the entire breadth of Amazon’s offerings.
Taddy’s book Business Data Science (McGraw Hill; August 2019) brings together concepts from statistics, machine learning, and social science to help businesses use their data effectively. In the lead-up to January's AEA Conference, Taddy talked to us about his book, what economists and scientists do at Amazon, and why economists should consider pursuing a career at the company.
What was the reason you wrote Business Data Science?
I started writing Business Data Science ten years ago. At the time, I was teaching a class of MBA students at the University of Chicago. I realized that there was an appetite for the material covered in the book from people who weren’t specialists in statistics or machine learning. And this idea that we could teach this material to non-specialists really motivated me not to just write this book, but also to push for changing the curriculum at the University of Chicago.
When I started working in the industry, I realized that there was an even bigger market of non-specialists. However, unlike the MBA students I taught at the University of Chicago, these were more technical people like software development engineers, who wanted to add to their toolset, and get into data science. This realization that there was a bigger universe of non-specialists spurred me on, and served as the second life for the book.
The timing for the release of the book couldn’t have been more perfect. Big data and machine learning tools have come of age over the last decade to become more scalable, robust and user friendly. This presents a unique opportunity for data scientists to have an impact. It also makes it easier for non-specialists to get started in learning data science. In the past, when you couldn't simulate something on your computer, you couldn't actually see how uncertainty works. You had to derive it mathematically. And today, when you have all this data, and you have more computation, there’s an opportunity to completely rework how we teach statistics so that it is oriented around these computational tools, helping make it easier to understand what's going on.
What are the primary messages you wanted to get across in your book?
First, I wanted to convey that non-specialists can do very good data science, and that data science can be very useful to them.
I also wanted to get another important point across – making decisions based on data is not just about making predictions. It’s also about understanding why things happen. There’s something to be said for machine learning which uncovers patterns in past data, and is able to predict a future that looks like the past. However, the real value of data science is unlocked when you’re able to explain what would happen if you did something entirely different from what you’ve been doing.
Lastly, I wanted to emphasize that as the tools have matured, and machine learning becomes commoditized, the real value is not building out a faster ML algorithm or developing a slightly better classification algorithm. The biggest value-add from scientists and economists is that they can use their domain knowledge to break complicated business problems into a bunch of tasks that can be solved through algorithms.
So you’re suggesting that if a scientist or economist were to apply for a job at Amazon, she wouldn’t necessarily have to be machine learning expert?
That’s right. Science at Amazon is big tent. I run a team called economic technology. We have software development engineers, scientists, economists, and product and program managers working on my team. And not all of them have a background in machine learning. To give just one example, I don’t have a PhD in economics. I'm a mathematician, an applied mathematician and statistician by training. But my trade has become applying these tools to study economic problems, and more recently to solve business problems.
Economists at Amazon don’t have to come in knowing how to build cutting-edge machine learning algorithms. Some do, but others operate like science product managers. They are able to translate a messy business problem into a set of clear requirements for applied scientists, who in turn can work with the software engineers to build the actual product that can drive impact at scale.
In the book, you write about the difference between artificial intelligence and machine learning. Could you elaborate on the difference?
The terms ‘machine learning’ and ‘artificial intelligence’ are often used interchangeably. But there’s a distinction, and it’s an important one. Machine learning is largely restricted to predicting a future that looks like the past. In contrast, an artificial intelligence system is able to solve complex problems that have previously been reserved for humans.
AI does this by breaking the problem into a series of tasks, each of which can be attacked by a “dumb” machine learning algorithm. This essentially involves three components: a) a well-defined task structure to engineer against, b) a strategy to continue generating data so that the system can continue to learn, and c) general purpose machine learning algorithms that can make predictions against unstructured data. Developing AI systems requires you to consider each of these components and take a holistic view of the problem.
Across the board, you’ll find that economists – and scientists more generally – are making an impact in a number of ways at Amazon.
You are the chief economist for Amazon’s consumer business in North America. How is Amazon combining machine learning and economics to optimize its business and accelerate decision making?
Economists at Amazon don’t just work on problems in economics in the conventional sense. We look at problems, understand the judgments we are making, and develop products that allow us to scale our offerings across the full breadth of Amazon.
The awesome thing about working at Amazon is that there are just so many different ways that we provide value to customers. For example, we help customers find what they are looking for, we help get products to their homes faster, or we factor in brand preferences and seasonality to ensure that the right products are stocked during the holiday season. These are just a few examples. Across the board, you’ll find that economists – and scientists more generally -- are making an impact in a number of ways at Amazon.
The American Economic Association. Conference is occurring in January. If you were a part of Amazon’s recruiting team there, what’s the pitch you would make for why these economists should consider joining Amazon?
At Amazon, we treat economics seriously as a discipline. That means that you will be able to work as an economist, and use the tools that you’ve learned in your PhD. But you will use these tools to do work that has a massive impact from day one.
When you come to Amazon, you’re going to learn what it is to be customer-obsessed and learn how to run a business. Because at Amazon, we are all owners. What this means is that you’ll own what you’re working on end-to-end, as opposed to consulting on a particular piece of the business.
Lastly, Amazon will give you the opportunity to think big like few other places. If you look at our track record, we’re going into places where we’re disrupting things massively. And when you're being asked to be a catalyst for change, or disruptor, the obvious path forward is not going to be laid out in front of you. So you need a big enough vision so you can actually invent your way out of the problems you’re facing. That makes coming to work every day incredibly fulfilling, and it’s why I would advise any economist to come and work at Amazon.
