How to Talk to Data Scientists

Jeremy Elser

Description

Every major company has or will soon have a data science program. Most fail, expensively, imperiling their executive sponsors. Unfortunately, executives have been misled to focus on the latest buzzwords. Although buzzwords change— big data, data science, machine learning, deep learning, and artificial intelligence –the distraction from fundamentals manifests as a predictable trajectory from exuberant program launch, to stagnation, to awkward decommissioning.

After architecting data science programs at over a dozen companies, across sectors and scales, Dr. Elser has formulated a reliable framework for successful data science programs. Surprisingly, software and algorithms are secondary. Rather, the key is understanding how the available data aligns to the problem to be solved. The business executive understands the problem sufficiently to enforce this alignment, while data scientists act on it. But executives tend to underestimate their role and thereby fail to construct the necessary connective tissue with their data scientists.

This book provides business executives with a concrete exercise, populating a “Master Table,” accessible to nontechnical managers and data scientists, which serves as the connective tissue between them. Rather than teach a diluted version of data science, this book describes how to start projects and how to detect and fix problems—the moments when leadership is critical. Insights are provided through real world examples, including a Playbook featuring common projects. The intended audience is executives (C-suite through VP). However, ambitious mid-level managers and data scientists will also benefit.

Sujets

• Gestion et management
• Economics
• Business
• Management

Informations

• Publié par: Business Expert Press
• Date de parution: 05 août 2021
• Nombre de lectures: 0
• EAN13: 9781637420980
• Langue: English
• Poids de l'ouvrage: 1 Mo

Extrait

How to Talk to Data Scientists
How to Talk to Data Scientists
A Guide for Executives
Jeremy Elser, PhD
How to Talk to Data Scientists: A Guide for Executives
Copyright © Business Expert Press, LLC, 2022.
Cover design by Charlene Kronstedt
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 means—electronic, 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 2021 by
Business Expert Press, LLC
222 East 46th Street, New York, NY 10017
www.businessexpertpress.com
ISBN-13: 978-1-63742-097-3 (paperback)
ISBN-13: 978-1-63742-098-0 (e-book)
Business Expert Press Collaborative Intelligence Collection
Collection ISSN: 2691-1779 (print)
Collection ISSN: 2691-1795 (electronic)
First edition: 2021
10 9 8 7 6 5 4 3 2 1
Description
Every major company has or will soon have a Data Science program. Most fail, expensively, imperiling their executive sponsors. Unfortunately, executives have been misled by technologists to focus on the latest buzzwords. Although buzzwords change—“Big Data,” “Data Science,” “Machine Learning,” “Deep Learning,” and “Artificial Intelligence,” the distraction from fundamentals manifests as a predictable trajectory from exuberant program launch, to stagnation, to awkward decommissioning.
After architecting Data Science programs at over a dozen companies, across sectors, from single-application startups to Fortune500 enterprisewide transformations, Dr. Elser has formulated a reliable framework for successful Data Science programs. Surprisingly, software and algorithms are inconsequential. Rather, the key is understanding how the data you have align to the problem you intend to solve. The business executive understands the problem sufficiently to enforce this alignment, while data scientists act on it. But executives tend to underestimate their role and thereby fail to construct the necessary connective tissue with their data scientists.
This book provides business executives with a concrete exercise, populating a “Master Table,” accessible to nontechnical managers and data scientists, which serves as the connective tissue between them. Rather than teach a diluted version of Data Science, this book is action-oriented, describing how to start projects and how to detect and fix problems—the moments when leadership is critical. Insights are provided through real-world examples and diagrams, including a Playbook featuring common projects. The intended audience is commercial executives (C-suite through VP). However, ambitious mid-level managers and even data scientists will also benefit.
Keywords
data science; machine learning; artificial intelligence; deep learning; big data; leadership; management; executive
Contents
Chapter 1 Introduction
Chapter 2 What Is Data Science
Chapter 3 The Master Table
Chapter 4 Mistakes Machines Make
Chapter 5 Mistakes Business Analysts Make
Chapter 6 Mistakes Data Scientists Make
Chapter 7 How to Properly Deploy Data Science
Chapter 8 Playbook
References
About the Author
Index
CHAPTER 1
Introduction
“I for one welcome our new computer overlords” was the response of Ken Jennings in 2011 as he ceded his Jeopardy crown to IBM’s Watson, a Data Science-driven machine. Given the sophistication of Jeopardy’s prompts, both phrasing and content, it seemed that humans were on the verge of being automated away. And yet, almost a decade later, there is not one robo-CEO, or even a robotic Head of regional marketing. In fact, the list of successful corporate Data Science programs is short; there are only a few categories of operational tasks that have benefitted. Worse, though most companies are reluctant to admit it, most corporate Data Science projects fail entirely.
Having served as a long-term consultant for various Fortune 500 companies across a variety of sectors, overseeing data-driven projects of all types spanning from mergers to marketing, I’ve detected a pattern that differentiates successful Data Science programs from failures. I’ve also served as Chief Data Scientist for a successful startup in the real estate data analytics space, and the patterns hold here as well, suggesting a very broad generality across scale. I have been the hands-on-keyboard data scientist, been the leader of Data Science teams, and been the business-focused project leader interacting with executives, middle-management, and end business operators. And my conclusion is that the primary gap isn’t insufficiently fancy algorithms or raw computing power, but rather a gap of knowledge between the data scientists (who understand math and not business) and the business executives (who understand their businesses but none of the Data Science). A successful Data Science program must not merely contain both of these personas on the team, but actually exchange some understanding so an overlap exists within each team member’s brain.
Data Science was sold as a panacea, a set-it-and-forget-it commodity, a “click here to increase sales” button. The key word being “sold.” In conferences and pitch decks, executives were given access to only the highlight reels of Data Science, usually embellished a bit as well. Even I’ve contributed, as an entrepreneur, I recognized that part of my job as Chief Data Scientist was to answer the question “Yes, Potential Investor, we are indeed using all of the Machine Learning,” and then rattle off the latest buzzwords in a confident tone. We, in fact, were successfully employing Machine Learning in our scoring and suggestion algorithms, and I did try to be educational, but I still had to play into the preexisting narrative of the omnipotence of algorithms. And these algorithms can be powerful; Amazon’s product recommendation algorithm and Gmail’s spam filtering are competitive advantages. But the reasons for the success weren’t just in deploying the latest sci-fi sounding technology (Machine Learning, Deep Learning, Artificial Intelligence), or even in hiring the world’s leading data scientists, but rather they were at the interface of those technical elements with the business itself—its data ecosystem and its business-specific people.

Takeaways for Chapter 1: Introduction
• Real-world outcomes of Data Science projects rarely match anticipated success.
• Successful Data Science projects require active understanding, participation, and leadership from business executives in addition to technical/analytical talent.
CHAPTER 2
What Is Data Science
Data Science encompasses a technological journey from statistics (dating to the 1700s) to the modern (circa 2020) embodiment of Artificial Intelligence.
Statistics : First, some terminology. For the last century, the closest technology to Artificial Intelligence (AI) was “statistics.” If you wanted to understand whether a higher price increase was associated with higher customer churn, or if the shipments to the Midwest were going on back-order more than in the East Coast, you would plot averages on a chart and observe the differences. Because natural variation always occurs, and may not always be meaningful (e.g., two regional branches might perform generally equivalently, but rarely will they have the exact same monthly sales), a rigorous analyst might look at how much fluctuation there is within the averages to determine whether the differences are significant—that is, unlikely to have happened by random chance fluctuations. However, most businesses rarely take such statistical caution.

Figure 2.1 Price elasticity example
In this fictional example, natural variation in the pricing of two products (premium and value), possibly due to periodic promotional discounts, is plotted with the resulting sales at those price points. The premium product sales seem to be independent of the unit price, suggesting price inelasticity, whereas the value product sales are dramatically lower with high price point. Although statistical tests can determine whether these results are likely to be due to random fluctuations or truly represent elasticity, the human eye can see that the lines are compelling and the conclusion matches our intuition.
Simple plots like Figure 2.1 have one variable of interest that you believe might be a driver (such as price) and one outcome variable (sales quantity). Human eyes are great at analyzing these kinds of charts.
The trickiness comes from the fact that outcome variables like sales are never just the result of a single variable like price, rather there are dozens or hundreds of little factors driving sales up and down. One of the more common drivers include seasonality—toy sellers sell more toys at Christmas than in spring, so if you are trying to compare the effectiveness of a 25 percent discount promotion that ran in spring versus a 50 percent promo that ran at Christmas, you had better account for the seasonal difference, or you’ll mistakenly inflate the usefulness of 50 percent discounts. In fact, the charts in Figure 2.2 assume that the reasons we have data comparing different price points are merely because different prices were arbitrarily chosen, like experiments, whereas in reality, the company may have deliberately changed pricing strategies as a result of changing business circumstances that themselves are “hidden” potential key variables causing the observed results. Business analysts generally try to account for the key variables by segmenting, which unfortunately multiplies the number of charts beyond what a human can simultaneously comprehend, so this approach works when only a few key drivers are needed to understand the outcome.

Figure 2.2 Discount promotions example
In this fictional example, promotional discounts of various severity were given on certain months and r