New Books Help Me to (Re)focus on Decision Intelligence

In 2018, soon after the end of my last sabbatical, I publicly announced that I had begun writing a book that would “focus on the many human interactions that take place while trying to make mathematical models work in innovative ways.” In 2022, just before my daughter’s departure for college, I publicly announced that I would soon be launching a podcast that would focus “on presenting illuminating stories about data-intensive optimization solutions, with a special emphasis on decisions made in dynamic business environments ...”

Alas, I have trouble with “focus.” For the past several years, these two ambitious projects have been gathering dust on the very large shelf of abandoned initiatives that clutters up both my office and my brain. But as I approach the end of my current sabbatical, I am excited to report that both of these ideas have recently been brought back to life. 

In May, I launched “The Decision Intelligence Laboratory” with co-host Michael Watson from Northwestern University. My original concept for the podcast had been optimization-centric, but we have broadened our focus, largely because of the emergence of generative artificial intelligence (AI) and its vast disruptive potential for decision support, augmentation and automation. Indeed, the podcast examines and explores the very broad range of technical, organizational and human challenges associated with turning raw data into tangible business value through improved decision-making. The podcast’s title was inspired by a new umbrella term, “Decision Intelligence,” that many industry analysts have begun using to encompass a broad class of technologies and managerial practices (Gartner’s official definition is here).

One of our first guests on the podcast was Doug Gray, co-author (with Evan Shellshear) of a terrific new book entitled “Why Data Science Projects Fail: The Harsh Realities of Implementing AI and Analytics, without the Hype.” Gray and Shellshear have both spent decades working in analytics, data science and artificial intelligence (collectively referred to as “ADSAI” throughout their book) roles in a variety of industries. Their partnership came about when Gray’s 10-part series of articles on project failures in Analytics magazine caught the attention of Shellshear, who had been the lead author on a Melbourne Business School white paper examining similar topics.

“Why Data Science Projects Fail” is a terrific book that provides readers with a well-written road map to a wealth of valuable observations and lessons. For starters, the book delivers a sobering, clear-eyed assessment of why so many AI and data science initiatives fall short, often spectacularly. From here, drawing on their extensive personal experience, the authors unpack the organizational, strategic, technical and human issues that doom so many ADSAI efforts.

The book is organized into four main sections (Strategy, Process, People and Technology), with each representing a broad area in which critical errors can occur. Each of these chapters offers diagnostic insight and practical guidance.

Another key theme in the book is the importance of organizational maturity. Gray and Shellshear argue convincingly that many companies jump into ADSAI projects long before they have the data infrastructure, governance or cross-functional culture needed to succeed in producing tangible business results. They also provide some rough data and analysis that suggests that the project failure rates at less analytically mature companies may exceed 90% – more than twice the level of more analytically mature companies.

The book’s real strength is in the vivid (frequently anonymized) stories from the authors’ extensive experience in the field. One memorable misadventure involves a machine learning model designed to predict defaults on home loans that failed to get buy-in either from the IT organization providing the key data inputs or from the would-be users who were accountable for the business results. Another notable failure involves a promising personalization model for a major retailer whose pathway to production usage was derailed by IT scope creep and political games. The book is filled with other such sad but illustrative examples.

Gray and Shellshear avoid hype and easy answers. Their tone is pragmatic but never cynical.

In a similar vein, the book concludes with a straightforward (but by no means simple) assertion: “project failure and success is largely a function of how effectively and how closely data science strategy, people, processes, and projects are integrated and aligned with the business.” Indeed, this is the book’s meta-lesson, a North Star that all of us would do well to heed.

Gray has also recently published a second book, “The Art of Data Science: A Practitioner’s Guide.” Part memoir, part history lesson, part field guide and tutorial, this book reads like an insider’s guide for those seeking career success as industry professionals, as well as a historical chronicle for those of us of a certain age who have spent our careers navigating the messy, imperfect, but often exhilarating world of decision intelligence.

This is not a typical data science book. Gray’s writing is both unabashedly practical and refreshingly unpretentious. In contrast to “Why Data Science Projects Fail,” this book’s subject matter is broader and more reflective. Throughout the book, he draws on his experiences from a long career of building analytical solutions and information systems across different industries, offering hard-earned wisdom to practitioners and leaders alike while presenting a number of valuable lessons for aspiring young data scientists, operations research analysts and AI professionals.

For example, in an early chapter entitled “The Dual Challenges of the Analytical Sciences Practitioner,” he discusses the importance of understanding both the analytical models and the business context in which they are applied, explicitly pointing out that effective data-driven solutions must be both technically sound and practically relevant. Another chapter entitled “Right Tool, Right Place, Right Time” provides insights into the connection between different decision problems on different time frames (planning, scheduling and operational control) and their respective requirements for accurate data, computing time and user interactivity. And the chapters on nontechnical and leadership skills for data scientists present many very important concepts that are rarely presented (and almost never prioritized) in technical graduate programs. 

Along with our recent podcast conversation with Doug, these two very compelling books have inspired me to revive – and reshape – my own long-stalled book project (more about this next time). Until then, I invite you to subscribe to “The Decision Intelligence Laboratory” podcast by sending an email to [email protected] with “Subscribe” in the subject line. You can also use this email address to provide us with your unvarnished feedback and/or ideas for future episodes of the podcast. We would love to hear from you.