Succeeding with AI: How to make AI work for your business

by Veljko Krunic

Overview

Summary

Companies small and large are initiating AI projects, investing vast sums of money on software, developers, and data scientists. Too often, these AI projects focus on technology at the expense of actionable or tangible business results, resulting in scattershot results and wasted investment. Succeeding with AI sets out a blueprint for AI projects to ensure they are predictable, successful, and profitable. It’s filled with practical techniques for running data science programs that ensure they’re cost effective and focused on the right business goals.

Purchase of the print book includes a free eBook in PDF, Kindle, and ePub formats from Manning Publications.

About the technology

Succeeding with AI requires talent, tools, and money. So why do many well-funded, state-of-the-art projects fail to deliver meaningful business value? Because talent, tools, and money aren’t enough: You also need to know how to ask the right questions. In this unique book, AI consultant Veljko Krunic reveals a tested process to start AI projects right, so you’ll get the results you want.

About the book

Succeeding with AI sets out a framework for planning and running cost-effective, reliable AI projects that produce real business results. This practical guide reveals secrets forged during the author’s experience with dozens of startups, established businesses, and Fortune 500 giants that will help you establish meaningful, achievable goals. In it you’ll master a repeatable process to maximize the return on data-scientist hours and learn to implement effectiveness metrics for keeping projects on track and resistant to calcification.

What's inside

    Where to invest for maximum payoff
    How AI projects are different from other software projects
    Catching early warnings in time to correct course
    Exercises and examples based on real-world business dilemmas

About the reader

For project and business leadership, result-focused data scientists, and engineering teams. No AI knowledge required.

About the author

Veljko Krunic is a data science consultant, has a computer science PhD, and is a certified Six Sigma Master Black Belt.

Table of Contents:

1. Introduction

2. How to use AI in your business

3. Choosing your first AI project

4. Linking business and technology

5. What is an ML pipeline, and how does it affect an AI project?

6. Analyzing an ML pipeline

7. Guiding an AI project to success

8. AI trends that may affect you

Product Details

ISBN-13:	9781638356318
Publisher:	Manning
Publication date:	03/15/2020
Sold by:	SIMON & SCHUSTER
Format:	eBook
Pages:	288
File size:	3 MB

About the Author

Veljko Krunic is a data science consultant, has a computer science PhD, and is a certified Six Sigma Master Black Belt.

Table of Contents

Preface xiii

Acknowledgments xv

About this book xvii

About the author xxi

About the cover illustration xxii

1 Introduction 1

1.1 Whom is this book for? 2

1.2 AI and the Age of Implementation 4

1.3 How do you make money with AI? 6

1.4 What matters for your project to succeed? 7

1.5 Machine learning from 10,000 feet 8

1.6 Start by understanding the possible business actions 11

1.7 Don't fish for "something in the data" 13

1.8 AI finds correlations, not causes! 15

1.9 Business results must be measurable! 16

1.10 What is CLUE? 19

1.11 Overview of how to select and run AI projects 21

1.12 Exercises 23

True/False questions 24

Longer exercises: Identify the problem 24

2 How to use AI in your business 26

2.1 What do you need to know about AI? 27

2.2 How is AI used? 29

2.3 What's new with AI? 31

2.4 Making money with AI 33

AI applied to medical diagnosis 34

General principles for monetizing AI 36

2.5 Finding domain actions 38

AI as part of the decision support system 39

AI as a part of a larger product 40

Using AI to automate part of the business process 42

AI as the product 43

2.6 Overview of AI capabilities 45

2.7 Introducing unicorns 47

Data science unicorns 47

What about data engineers? 48

So where are the unicorns? 49

2.8 Exercises 50

Short answer questions 51

Scenario-based questions 51

3 Choosing your first AI project 53

3.1 Choosing the right projects for a young AI team 54

The look of success 54

The look of failure 57

3.2 Prioritizing AI projects 59

React: Finding business questions for AI to answer 60

Sense/Analyze: AI methods and data 63

Measuring AI project success with business metrics 65

Estimating AI Project Difficulty 68

3.3 Your first project and first research question 69

Define the research question 70

If you fail, fail fast 74

3.4 Pitfalls to avoid 74

Failing to build a relationship with the business team 75

Using transplants 75

Trying moonshots without the rockets 76

It's about using advanced tools to look at the sea of data 77

Using your gat feeling instead of CLUE 78

3.5 Exercises 80

4 Linking business and technology 82

4.1 A project can't be stopped midair 83

What constitutes a good recommendation engine? 83

What is gut feeling? 85

4.2 Linking business problems and research questions 85

Introducing the L part of CLUE 86

Do you have the right research question? 87

What questions should a metric be able to answer? 87

Can you make business decisions based on a technical metric? 88

A metric you don't understand is a poor business metric 91

You need the right business metric 93

4.3 Measuring progress on AI projects 94

4.4 Linking technical progress with a business metric 96

Why do we need technical metrics? 97

What is the profit curve? 97

Constructing a profit curve for bike, rentals 99

Why is this not taught in college? 102

Can't businesses define the profit curve themselves? 103

Understanding technical-results in business terms 105

4.5 Organizational considerations 106

Profit curve precision depends on the business problem 106

A profit curve improves over time 107

It's about learning, not about being right 108

Dealing with information hoarding 108

But we can't measure that! 109

4.6 Exercises 110

5 What is an ML pipeline, and how does it affect an AI project? 112

5.1 How is an AI project different? 113

The ML pipeline in AI projects 113

Challenges the AI system shares with a traditional software system 117

Challenges amplified in AI projects 117

Ossification of the ML pipeline 118

Example of ossification of an ML pipeline 121

How to address ossification of the ML pipeline 123

5.2 Why we need to analyze the ML pipeline 126

Algorithm improvement: MNIST example 126

Further examples of improving the ML pipeline 127

You must analyze the ML pipeline! 128

5.3 What's the role of AI methods? 129

5.4 Balancing data, AI methods, and infrastructure 131

5.5 Exercises 133

6 Analyzing an ML Pipeline 135

6.1 Why you should care about analyzing your ML pipeline 136

6.2 Economizing resources: The E part of CLUE 138

6.3 MinMax analysis: Do you have the right ML pipeline? 140

6.4 How to interpret MinMax analysis results 142

Scenario; the ML pipeline for a smart parking meter 142

What if your ML pipeline needs improvement? 146

Rules for interpreting the results of MinMax analysis 147

6.5 How to perform an analysis of the ML pipeline 147

Performing the Min part of MinMax analysis 149

Performing the Max part of MinMax analysis 149

Estimates and safety factors in MinMax analysis 152

Categories of profit curves 154

Dealing with complex profit curves 157

6.6 FAQs about MinMax analysis 159

Should MinMax be the first analysis of the ML pipeline? 160

Which analysis should you perform first? Min or Max? 160

Should a small company or small team skip the MinMax analysis? 161

Why do you use the term MinMax analysis? 161

6.7 Exercises 162

7 Guiding an AI project to success 165

7.1 Improving your ML pipeline with sensitivity analysis 166

Performing local sensitivity analysis 167

Global sensitivity analysis 170

Example of using sensitivity analysis results 171

7.2 We've completed CLUE 172

7.3 Advanced methods for sensitivity analysis 175

Is local sensitivity analysis appropriate for your ML pipeline? 176

How to address the interactions between ML pipeline stages 179

Should I use design of experiments? 180

One common objection you might encounter 181

How to analyze the stage that produces data 184

What types of sensitivity analysis apply to my project? 184

7.4 How your AI project evolves through time 186

Time affects your business results 186

Improving the ML pipeline over time 187

Timing diagrams: How business value changes over time 188

7.5 Concluding your AI project 190

7.6 Exercises 192

8 AI trends that may affect you 195

8.1 What is AI? 196

8.2 AI in physical systems 198

First, do no harm 198

IoT devices and AI systems must play well together 201

The security of AI is an emerging topic 202

8.3 AI doesn't learn causality, only correlations 203

8.4 Not all data is created equal 206

8.5 How are AI errors different from human mistakes? 207

The actuarial view 208

Domesticating AI 210

8.6 AutoML is approaching 211

8.7 What you've learned isn't limited to AI 213

8.8 Guiding AI to business results 214

8.9 Exercises 216

Appendix A Glossary of terms 219

Appendix B Exercise solutions 225

Appendix C Bibliography 244

Index 257