Introduction

What Am I Doing Here?

If you're reading this book, it's because on some level you understand the importance of both data and data science in your business and career.

The original Data Smart was written more than a decade ago. John Foreman, the first book's author, exposed a new generation of readers to the supposed magic behind the curtain of data science. John proved that data science didn't have to be so mysterious. You could both understand and do data science in something as humble as the spreadsheet.

John's words severed as a prescient warning for what would come. He noted the "buzz about data science," and the pressure it created on businesses to take on data science projects and hire data scientists without even fully understanding why.

The truth is most people are going about data science all wrong. They're starting with buying the tools and hiring the consultants. They're spending all their money before they even know what they want, because a purchase order seems to pass for actual progress in many companies these days.

John's words still ring true today. Ten years after the first wave of interest in data science, the data science machine is still working in full force, churning out ideas faster than we can articulate the opportunities and challenges they present to business and society. In my last book, Becoming a Data Head: How to Think, Speak and Understand Data Science, Statistics and Machine Learning (Wiley, New York, NY, 2021), my coauthor and I called this the data science industrial complex.

To put it bluntly, despite the extensive interest in data and data science, projects still fail sometimes at alarming rates, even as data is supposed to be fact driven. In truth, as much as 87 percent data science projects won't make it into production.1

What is and isn't a "data disaster" is perhaps up from some considerable debate. But it's fair to say the recent past is filled with examples in which technology, data, and the like were hailed as something magical before they ultimately came up short. Here are just a few examples worth considering:

• An attorney used a generative AI chatbot for legal research, submitting a brief to the court with cases that did not exist, but perhaps sounded plausible.2
• The COVID-19 pandemic exposed major issues in forecasting across the board, from supply chain issues to understanding the spread of the virus.3
• When the original Data Smart came out, accurately predicting the outcome of the US presidential election seemed like an easy feat. In 2016, however, model after model inaccurately predicted a win for Hillary Clinton, despite increased money, time, and effort into the subject.4

Most data science projects and outcomes don't fail so spectacularly. Instead, data science projects die slow deaths, while management pours money and resources into chasing elusive numbers they don't entirely understand.

Yet, some of the greatest data achievements did not come from any particular technology. Rather, they came from human ingenuity. For instance, I used to lead projects for a nonprofit called DataKind, which leverages "data science and AI in service of Humanity."

DataKind uses teams of volunteer data scientists to help mission-driven organizations design solutions to tough social problems in an ethical and socially responsible way. When I was there, we worked with major organizations like the United Nations and Habitat for Humanity.

Volunteers built all sorts of models and tools, from forecasting water demand in California to using satellite imagery to identify villages in need with machine learning. The work we did had impact, so it's not all doom and gloom. When you're done with this book, you might consider giving back in your own way.5 Remember: Humans solve problems not machines.

What Is Data Science?

In my last book, Becoming a Data Head, Alex Gutman (my coauthor) and I actually don't define data science. One reason is that the space is too hard to pin down. And we didn't want folks to get caught up in trying to justify whether or not they were data scientists. In the original Data Smart, John Foreman offers this working definition:

Data science is the transformation of data using mathematics and statistics into valuable insights, decisions, and products.

John takes a broad, business-centric view. He's quick to note it's a "catchall buzzword for [everything] analytics today." Ten years later, I and the rest of the industry are still struggling to define exactly what data science is and isn't. So rather than proffer a definition as if that will get us closer to the truth, I'd rather describe what a data scientist does.

• Data scientists identify relevant questions that can be solved with data. This may sound obvious, but many questions can't be solved with data and technology. A good data scientist can tease out the problems in which algorithms and analyses make the most sense.
• Data scientists extract meaningful patterns and insights from data. Anyone can eyeball a set of numbers and draw their own conclusions. On the other hand, data scientists focus on what can be said statistically and verifiably. They separate speculation from science, focusing instead on what the data says.
• Finally, data scientists convey results using data visualization and clear communication. In many cases, a data scientist will have to explain how an algorithm works and what it does. Historically, this has been a challenge for many in the field. But a recent crop of books (like this one) aims at giving data scientists a way to explain how they came to their results without being too stuck into the weeds.

Incredibly, some of the techniques mentioned in the following pages are as old as World War II. They were invented at the dawn of the modern computer, long before you could easily spin up a new instance of R. The hype machine won't tell you these "new" algorithms were first developed on punch cards.

And some of the techniques in this book were invented recently, taking advantage of the wealth of data, self-service analytics, cloud computers, and new graphical processing units developed in the last 10 years.

Again, we're reminded that human ingenuity is what drives this field forward.

Age has no bearing on difficulty or usefulness. All these techniques whether or not they're currently the rage are equally useful in the right business context. It's up to you to use them correctly. That's why you need to understand how they work, how to choose the right technique for the right problem, and how to prototype with them.

Do Data Scientists Actually Use Excel?

Many (but not all) veteran data scientists will tell you they loathe spreadsheets and Excel in particular. They will say that Excel isn't the best place to create a data science model. To some extent, they're right.

But before you throw this book away, let's understand why they say this. You see, there was a time before R and before Python. It was a time when MATLAB and SPSS reigned supreme. The latter tools were expensive and often required a computer with some major horsepower to run a model. Moreover, the files that these tools generated were not easily distributable. And, in a secure corporate or institutional environment, sending files with code in them over email would trip the unsafe-email alarms.

As a result, many in the industry began building their work in Excel. This was particularly true of models that helped support executive decision-making. Excel was the secret way around these email systems. It was a way to build a mini data application without having to get approval from the security team.

Many executive teams relied on Excel. Unfortunately, this also created a myopic view among executives who didn't really understand data science. For them, Excel was the only place to do this type of work. It was where they were most comfortable.

They knew the product. They could see what the analyst created. And the analyst could walk them through each step. In fact, that's why we're using Excel in this book.

But Excel (at the time) was limited. Limited by how much it could process at any moment. Limited by the amount of data it could store. The macro language behind Excel, Visual Basic for Applications (VBA), is still hailed by many executives as an advanced feature. But VBA is based on Visual Basic 6.0, which was deprecated in 1999. The Excel version of this language has received only the barest of updates. When today's data scientists point out that VBA can't do what R or Python can, it's hard to disagree.

On the flipside, however, Microsoft has paid attention over the last few years. The Excel product team has come to understand how data scientists use their tool. They've poured more research into some very specific use cases. For instance, we'll talk about an entirely new data wrangling tool in Excel called Power Query. Power Query can do the same data wrangling tasks as in Python and R, often more quickly. And we'll talk about new Excel functions that make data science in Excel a whole lot easier. Today, there is renewed interest in using Excel for data science problems beyond what was possible only a few years ago.

But if there's a place where...