1
Introduction

1.1 Why Managers Need to Know About Data Science

There are many "data science for managers" books on the market today. They are filled with business success stories, pretty visualizations, and pointers about what some of the hot trends are. That material will get you rightfully excited about data science's potential, and maybe even get you started off on the right foot with some promising problems, but it isn't enough to see projects over the finish line or bring the full benefits of data to your organization. Depending on your role you may also need to decide how much to trust a piece of analytical work, make final calls about what tools your company will invest in, and hire/manage a team of data scientists. These tasks don't require writing your own code or performing mathematical derivations, but they do require a solid grounding in data science concepts and the ability to think critically about them.

In the past, mathematical disciplines like statistics and accounting solved precisely defined problems with a clear business meaning. You don't need a STEM degree to understand the idea of testing whether a drug works or balancing a checkbook! But as businesses tackle more open-ended questions, and do so with datasets that are increasingly complex, the analytics problems become more ambiguous. A data science problem almost never lines up perfectly with something in a textbook; there is always a business consideration or data issue that requires some improvisation. Flexibility like this can become recklessness without fluency in the underlying technical concepts. Combine this with the fact that data science is fast becoming ubiquitous in the business world, and managers and executives face a higher technical bar than they ever did in the past.

Business education has not caught up to this new reality. Most careers follow a "business track" that teaches few technical concepts, or a "technical track" that focuses on hands-on skills that are useless for businesspeople. This book charts a middle path, teaching non-technical professionals the core concepts of modern data science. I won't teach you the brass tacks of how to do the work yourself (that truly is for specialists), but I will give you the conceptual background you need to recognize good analytics, frame business needs as solvable problems, manage data science projects, and understand the ways data science is likely to transform your industry.

In my work as a consultant I have seen PMs struggle to mediate technical disagreements, ultimately making decisions based on peoples' personalities rather than the merits of their ideas. I've seen large-scale proof-of-concept projects that proved the wrong concept, because organizers set out inappropriate success metrics. And I've seen executives scratching their heads after contractors deliver a result, unable to see for themselves whether they got what they paid for.

Conversely, I have seen managers who can talk shop with their analysts, asking solid questions that move the needle on the business. I've seen executives who understand what is and isn't feasible, instinctively moving resources toward projects that are likely to succeed. And I've seen non-technical employees who can identify key oversights on the part of analysts and communicate results throughout an organization.

Most books on data science come in one of two types. Some are written for aspiring data scientists, with a focus on example code and the gory details of how to tune different models. Others assume that their readers are unable or unwilling to think critically, and dumb the technical material down to the point of uselessness. This book rejects both those approaches. I am convinced that it is not just possible for people throughout the modern business workforce to learn the language of data: it is essential.

1.2 The New Age of Data Literacy

Analytics used to play a minor role in business. For the most part it was used to solve a few well-known problems that were industry-specific. When more general analytics was needed, it was for well-defined problems, like conducting an experiment to see what product customers preferred.

Two trends have changed that situation. The first is the intrusion of computers into every aspect of life and business. Every phone app, every new feature in a computer program, every device that monitors a factory is a place where computers are making decisions based on algorithmic rules, rather than human judgment. Determining those rules, measuring their effectiveness, and monitoring them over time are inherently analytical. The second trend is the profusion of data and machines that can process it. In the past data was rare, gathered with a specific purpose in mind, and carefully structured so as to support the intended analysis. These days every device is generating a constant stream of data, which is passively gathered and stored in whatever format is most convenient. Eventually it gets used by high-powered computer clusters to answer a staggering range of questions, many of which it wasn't really designed for.

I don't mean to make it sound like computers are able to take care of everything themselves - quite the opposite. They have no real-world insights, no creativity, and no common sense. It is the job of humans to make sure that computers' brute computational muscle is channeled toward the right questions, and to know their limitations when interpreting the answers. Humans are not being replaced - they are taking on the job of shepherding machines.

I am constantly concerned when I see smart, ethical business people failing to keep up with these changes. Good managers are at risk of botching major decisions for dumb reasons, or even falling prey to unscrupulous snake oil vendors. Some of these people are my friends and colleagues. It's not a question of intelligence or earnestness - many simply don't have the required conceptual background, which is understandable. I wrote this book for my friends and people like them, so that they can be empowered by the age of data rather than left behind.

1.3 Data-Driven Development

So where is all of this leading? Cutting out hyperbole and speculation, what does it look like for an organization to make full use of modern data technologies and what are the benefits? The goal that we are pushing toward is what I call "data-driven development" (DDD). In an organization that uses DDD, all stages in a business process have their data gathered, modeled, and deployed to enable better decision making. Overall business goals and workflows are crafted by human experts, but after that every part of the system can be monitored and optimized, hypotheses can be tested rigorously and retroactively, and large-scale trends can be identified and capitalized on. Data greases the wheels of all parts of the operation and provides a constant pulse on what's happening on the ground.

I break the benefits of DDD into three major categories:

1. 1. Human decisions are better-informed: Business is filled with decisions about what to prioritize, how to allocate resources, and which direction to take a project. Often the people making these calls have no true confidence in one direction or the other, and the numbers that could help them out are either unavailable or dubious. In DDD the data they need will be available at a moment's notice. More than that though, there will be an understanding of how to access it, pre-existing models that give interpretations and predictions, and a tribal understanding of how reliable these analyses are.
2. 2. Some decisions are made autonomously: If there is a single class of "killer apps" for data science, it is machine learning algorithms that can make decisions without human intervention. In a DDD system large portions of a workflow can be automated, with assurances about performance based on historical data.
3. 3. Everything can be measured and monitored: Understanding a large, complex, real-time operation requires the ability to monitor all aspects of it over time. This ranges from concrete stats - like visitors to a website or yield at a stage of a manufacturing pipeline - to fuzzier concepts like user satisfaction. This makes it possible to constantly optimize a system, diagnose problems quickly, and react more quickly to a changing environment.

It might seem at first blush like these benefit categories apply to unrelated aspects of a business. But in fact they have much in common: they rely on the same datasets and data processing systems, they leverage the same models to make predictions, and they inform each other. If an autonomous decision algorithm suddenly starts performing poorly, it will prompt an investigation and possibly lead to high-level business choices. Monitoring systems use autonomous decision algorithms to prioritize incidents for human investigation. And any major business decision will be accompanied by a plan to keep track of how well it turns out, so that adjustments can be made as needed.

Data science today is treated as a collection of stand-alone projects, each with its own models, team, and datasets. But in DDD all of these projects are really just applications of a single unified system. DDD goes so far beyond just giving people access to a common database; it keeps a pulse on all parts of a business...