John Hawkins - Getting Data Science Done

Getting Data Science Done

Getting Data Science Done

Managing Projects From Ideas to Products

John Hawkins

Getting Data Science Done: Managing Projects From Ideas to Products

Copyright Business Expert Press, LLC, 2023.

Cover design by John Hawkins

Interior design by Exeter Premedia Services Private Ltd., Chennai, India

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any meanselectronic, mechanical, photocopy, recording, or any other except for brief quotations, not to exceed 400 words, without the prior permission of the publisher.

First published in 2022 by

Business Expert Press, LLC

222 East 46th Street, New York, NY 10017

www.businessexpertpress.com

ISBN-13: 978-1-63742-277-9 (paperback)

ISBN-13: 978-1-63742-278-6 (e-book)

Business Expert Press Big Data, Business Analytics, and Smart Technology Collection

First edition: 2022

10 9 8 7 6 5 4 3 2 1

Description

Data science is a field that synthesizes statistics, computer science and business analytics to deliver results that can impact almost any type of process or organization. Data science is also an evolving technical discipline, whose practice is full of pitfalls and potential problems for managers, stakeholders and practitioners. Many organizations struggle to consistently deliver results with data science due to a wide range of issues, including knowledge barriers, problem framing, organizational change and integration with IT and engineering.

Getting Data Science Done outlines the essential stages in running successful data science projects. The book provides comprehensive guidelines to help you identify potential issues and then a range of strategies for mitigating them. The book is organized as a sequential process allowing the reader to work their way through a project from an initial idea all the way to a deployed and integrated product.

Keywords

learn data science; data science process; managing data science projects; problem framing; project management; predictive analytics; machine learning delivery; data science solutions

Contents

This book began in 2014 as a collection of notes I was taking on the general process of running a data science project. It was largely driven by a desire to improve my own process and bring more discipline to how I delivered results. It was also inspired by my personal curiosity about the extent to which data science can be distilled into a procedure that can be automated.

If you spend any time with data scientists, you will soon discover that one of their favorite topics for lunchtime conversations is whether their job can be automated. This is the data scientists existential dilemma. We are, to a large extent, involved in the process of reshaping jobs and automating many tasks. It is therefore natural to ponder whether our jobs will suffer the same fate.

As I worked on this book and read more about data science project delivery, it became clearer to me that the predominant reasons that data science projects fail are not technical. Projects either never get to the stage where a technical solution can be built or the solution is never taken through to production because it solved the wrong problem. These failures are less about the technical decisions involved, and more about how data scientists nurture the stakeholders of the project.

Consequently, this book is different from most data science books. It is about how you effectively deliver data science projects from the first meeting to the last. To this end, my focus is on the aspects of data science that are not covered by other books. Some of these topics are semitechnical, including the choice of metrics and interventions, but many would be more accurately classified as soft skills. This includes framing a problem and recognizing the real problem that needs to be solved (as opposed to what your client or manager is asking of you), and many other skills related to managing and delivering an analytics project. Some of these skills are common to running any kind of technical project, but because of the unusual nature of data science they take on their own flavor. This can include project scoping, expectation setting, communication, prioritization, documentation, and project management.

All these aspects of a data science project are necessary, but generally receive little to no attention in data science courses. The lack of development of these topics means that as a community we tend to do these things poorly. Everything in this book I have either learned the hard way or learned from watching and working with people who are much better at these things than I am.

It has been my experience that far too many projects fail for preventable reasons that have little to do with the technical capability of the data scientists involved. I hope that you find the ideas collected herein useful in driving your own projects, and I hope you will not hesitate to reach out and let me know how they work for you.

This book would not have been possible without the numerous fantastic colleagues I have worked with over the years. People from whom I have learned through collaboration and argument, as well as observation of success and failure. In particular, my former academic supervisors William Herfel and Mikael Bodn, both of whom exerted a great deal of influence on the development of my thought patterns. My former colleagues David Rowe, Jesse Wu, Sabrina Rodrigues, and James Petterson with whom I have discussed many of the specific ideas and problems in this book.

I am immensely grateful to David Dufty, Sabrina Rodrigues, Jesse Wu, Trung Nguyen, Will Hanninger, and Gourab De for reviewing and providing invaluable feedback on this manuscript. I would like to thank Scott Isenberg from BEP for his guidance on preparing and publishing this book. I am grateful to my wife Diamond Hawkins for supporting me through the process, and finally, to my mother Margaret Hawkins for allowing me the intellectual freedom to discover what I loved doing.

Data science projects regularly fail to deliver results. Industry analysts report a spread of statistics, with conservative estimates suggesting that projects completely fail, or result in insufficient returns, almost as often as they succeed.

Investigations into specific failures reveal a variety of contributing factors. It can be a combination of organizational change challenges, poor problem framing, issues with data quality, and an inability to productionize results. More importantly, for practitioners of data science, it has been identified that many machine learning and analytics experts are not spending enough time and energy on defining the right problem, and then driving toward a specific business outcome.

In the following chapters, we will be exploring the ways these problems manifest themselves throughout the project life cycle. I will discuss strategies for mitigating the problems where possible, and approaches for identifying and managing risks in other situations. By discussing these issues in an approximately sequential fashion, my hope is that it will help you get an understanding of how unresolved issues early in a project can manifest themselves later as much larger problems.

A problem well stated is a problem half solved.

Charles Kettering

At the start of any project, there is a period when everyone involved is trying to determine what the project means for them. The resulting project scope will include a definition of what is being delivered, who is delivering it, and the arrangement of milestones across the timeline. In data science work, there is usually an additional stage at the beginning. Before any of the specific deliverables can be defined, you will need to analyze and understand the problem you are solving.