Demystifying AI for the Enterprise

A Playbook for Business Value and Digital Transformation

Prashant Natarajan, Bob Rogers, Edward Dixon, Jonas Christensen, Kirk Borne, Leland Wilkinson, Shantha Mohan

• Verlag: Productivity Press
• Jahr: 2021
• ISBN:
  9781351032926
• Auflage: 1
• Thema: Commerce
• Thema: Business Intelligence

Artificial intelligence (AI) in its various forms –– machine learning, chatbots, robots, agents, etc. –– is increasingly being seen as a core component of enterprise business workflow and information management systems. The current promise and hype around AI are being driven by software vendors, academic research projects, and startups. However, we posit that the greatest promise and potential for AI lies in the enterprise with its applications touching all organizational facets.

With increasing business process and workflow maturity, coupled with recent trends in cloud computing, datafication, IoT, cybersecurity, and advanced analytics, there is an understanding that the challenges of tomorrow cannot be solely addressed by today's people, processes, and products.

There is still considerable mystery, hype, and fear about AI in today's world. A considerable amount of current discourse focuses on a dystopian future that could adversely affect humanity. Such opinions, with understandable fear of the unknown, don't consider the history of human innovation, the current state of business and technology, or the primarily augmentative nature of tomorrow's AI.

This book demystifies AI for the enterprise. It takes readers from the basics (definitions, state-of-the-art, etc.) to a multi-industry journey, and concludes with expert advice on everything an organization must do to succeed. Along the way, we debunk myths, provide practical pointers, and include best practices with applicable vignettes.

AI brings to enterprise the capabilities that promise new ways by which professionals can address both mundane and interesting challenges more efficiently, effectively, and collaboratively (with humans). The opportunity for tomorrow's enterprise is to augment existing teams and resources with the power of AI in order to gain competitive advantage, discover new business models, establish or optimize new revenues, and achieve better customer and user satisfaction.

Chapter 1 AI Strategy for the Executive

Prashant Natarajan

H2O.ai

DOI: 10.4324/9781351032940-1

Contents

1. Introduction
2. Applications of AI
   • Determining Practical Realization: Considerations
3. Definitions
4. IMPACT Framework for Enterprise AI
   • Imagination
   • Maturity
     1. Dimensions of AI Maturity
     2. Assessing and Increasing AI Maturity in Your Organization
   • People
     • Considerations on the Data Scientist Role
   • Automation, Amplification, and Augmentation
   • Culture
   • Transformation
5. Best Practices for the Use of Data in AI
   • Volume
   • Variety
   • Velocity
   • Value and Veracity
   • Value
   • Veracity
     • Data Fidelity over Data Quality
6. Conclusions
7. Notes

Introduction

Human and organizational needs, business trends, evolving customer behaviors, and rapid data and technology innovations are some of the key drivers that are making Artificial Intelligence (AI) an essential foundation of the modern enterprise. In a business landscape that is increasingly informed by both global and local trends, the need to put data to work is more important and relevant than ever before. In the past, leaders and managers have relied on a combination of instincts, intuition, and Business Intelligence (BI) (“what happened in the past?”) to make decisions around services/products, processes, and people.

While instincts, intuition, and BI can be right, at times, the demands on today’s leaders, managers, and as importantly, the demands on and needs of staff/customers, require both decision-making and question framing to be done using a robust framework that privileges quantitative and qualitative data (that represents experience and behaviors). More importantly, digital transformation requires the modern enterprise to put these data to work by taking advantage of the latest developments in AI in all its forms, and thinking beyond automation in favor of augmentation and amplification.

An enterprise still has an option of ignoring AI, or keeping AI boxed to limited value generation; however, the consequences of doing so are severe for the following reasons:

• Competitive disadvantage—an enterprise that puts all its data to work via AI is better placed to create new advantages and positive differentiation w.r.t. the competition—existing and upcoming. In today’s globalized landscape, the competition is not only known entities within a country but also foreign/transnational enterprises and startups/unicorns that are leveraging AI extensively
• Customer and employee experiences—using AI to understand a customer’s or employee’s personalization needs, and address it via micro-campaigns/products/experiences creates a better human experience and drives competitive advantage
• Suboptimal leverage of human resources—by focusing on mundane and robotic tasks in lieu of enabling employees/associates/professionals focus on what matters more
• Misalignment between business strategy and tactical solutions—technology is no longer a nice to have. The best business strategy can’t be realized unless the enterprise can identify/design/develop appropriate data and technology solutions
• Top line and bottom line—by not using AI, an enterprise won’t be in the best position to manage opportunity or risk, and maximize top- and bottom-line revenues, margins, and profitability
• Prepare for new conditions—measuring what’s working or not and predicting known and unknown “unknowns.” Managing appropriate/corrective measures are increasingly being recognized as being necessary to thrive, or at the very least, survive in turbulent times and when faced by “black swan” events such as pandemics, war, major recessions, and market conditions (the Kodak Effect being a well-studied example of not responding in a timely fashion to market and user preferences)

Applications of AI

The applications of AI span multiple industries/verticals, business units/departments, and users—as you will discover in the chapters in this book. While there is value in targeted applications of learning algorithms and visualization, the greatest impact of AI can be realized by treating it as a fundamental and foundational capability—and recognizing that AI can transform any part of your business or organization.

Many enterprises that are currently in this journey of deploying AI are doing it in B2B, B2C, and peer-to-peer customer contexts. Verticals as diverse as e-commerce, retail, banking, insurance, government, healthcare, manufacturing, smart cities, defense, and construction/real estate among many others are deploying AI on structured and unstructured data to optimize and create new value in

• Customer engagement and happiness
• Marketing and Sales
• Supply chain and operations
• Finance and accounting
• Regulations, compliance, and risk management
• Product development and pricing
• Talent acquisition and human resource management
• Operations
• More granular/specialized subdomains and processes

We are confident that the applications of AI will only continue to grow in commercial enterprises, governments, and nonprofits as expectations and talent availability grow, and AI becomes a commodity and democratized as we have seen with other scaled technology innovations in the past 150+ years.

Determining Practical Realization: Considerations

Technology doesn’t exist in a vacuum and this fact must be recognized to ensure the successful applications of AI, or at the very least, reduce the challenges and frustration that will happen when AI is not deployed in the right context.

Success with AI requires company boards, executives, and change leaders to recognize that the readiness and maturity to incorporate the insights and new AI-driven workflows will vary—across cultural, technical, and process dimensions—in your organization. Getting buy-in across the department or an organization including among the workforce is essential as is the need to determine and prioritize use cases, cost–benefit analyses, and value generation.

If there is no precedent in your organization or if you’re taking baby steps, it is critical to define your AI strategy before setting up teams, hiring costly talent, or investing in a machine learning (ML) or analytics product. We also recommend that you validate your strategy with market analyses and customer interviews (when possible) and be prepared for that strategy to change as you learn from tactical successes and operationalization of AI in business processes and workflow solutions.

The current state of AI technologies relies on and leverages data that are created or entered by humans or generated (for example, in the case of IoT or automation) by human-engineered systems. Data may be insufficient in some cases, may not have the appropriate data fidelity, or may require business process changes to get the required data in place for the AI application to generate the intended results. While techniques such as imputations, extrapolations, and trying your best with what’s available are options—the relationship between data and AI follows the bedrock principle of GIGO (Garbage in, Garbage Out). We will discuss best practices around data and AI later in this chapter.

Separating AI from the rest of your legacy technologies is necessary at times—a new AI application can add value by itself (for example, in Natural Language Processing, Computer Vision, and Predictive Analytics) or in conjunction with a new application or digital transformation strategy. That being said, you will find that your ability to design/deploy the latest in AI and applying these to the lowest common technology denominator—spreadsheets, documents, and legacy systems—allows you to successfully meet users where they are and bring value to them—without necessarily requiring widespread revamp, large investments, or needless disruptions.

There may also be cases, especially with “black swans” or “never events,” where an event of large magnitude has no/little precedent; and the available data are sparse, changing rapidly or is not a reliable indicator of future performance. As the business world is discovering with COVID-19, underlying challenges cannot be solely papered over with AI—though in many cases it can help. It is important for decision makers to understand when AI can’t always predict the present or the future with a high degree of confidence or accuracy, understanding what’s normal and what’s an anomaly, and knowing when to supplement AI with human experience.

Similar to starting first with an enterprise strategy, creating the right business case and documenting business/functional requirements are important for executives and mid-level managers before they go down the path of staffing, solutioning, and spending. What works for another vertical, or even a competitor, may not be the best fit for you given your unique organizational needs, culture, and situation. Be enthusiastic about saying yes to high-impact AI solutions and keep an open mind to opportunities. Equally, be open to saying no to a specific use case or business solution where AI is not required—in favor of other more impactful ones. Don’t give in to the temptation of having a vendor or a siloed internal technology team develop and deploy a solution for its own sake—as leaders, it is our responsibility to ensure long-term successes, which are typically not Big Bang but an accumulated journey of individual success stories.

Finally, the intended uses of AI must follow tried and tested best practices (discussed later in this chapter) and ethical considerations. Responsible AI and Explainable AI are increasingly becoming de-rigueur across enterprises of all sizes and types—not just for regulatory/compliance reasons—but also because it’s the right thing to do for the various human stakeholders involved in the strategy, design, and use of AI. Some principles to consider on this topic are

• First, do no harm to your customers, employees, or business
• Build continual feedback loops to ensure and maintain a high degree of trust
• Historical data can be biased; pay attention to resulting algorithmic results and how these results are integrated into transaction systems or operational/business processes
• Set aside time and money to address ethical/responsible dimensions on each individual project
• Encourage collaboration—not just between data scientists and other technologists; but more importantly, between data scientists, data and analytics leaders, business stakeholders, and customers
• Interpretability (enabling a data scientist to understand how the algorithm came about its results from mathematical and data perspectives) and explainability (the ability for a nontechnical user of the solution to understand the process and results in plain language) are related but also serve different uses and audiences
• Understand that we will have to address tradeoffs between interpretability, explainability, time, and accuracy—and there is no single formula. Keep in mind that not every AI solution needs to be interpretable to the nth degree, depending on the use case it supports (some use cases will require more; others will need less)

Definitions

Our goal is to distill data down to refined and actionable Intelligence. The raw materials are different kinds of data, the tools are analytics and learning algorithms.

Let’s define our terms more precisely.

Intelligence

• Human intelligence, is the “mental quality that consists of the abilities to learn from experience, adapt to new situations, understand and handle abstract concepts, and use knowledge to manipulate one’s environment.”¹
• Artificial Intelligence, is “a branch of computer science dealing with the simulation of intelligent behavior in computers, or the capability of a machine to imitate intelligent human behavior.”² In this context, AI as defined for the purposes of this book is the ability to decide on an action or make an inference given some data or information.

Data. Data are any kind of fact or piece of information that we can record, store, and subsequently retrieve for further processing. Data do not drive action in its own right but can be processed or interpreted within a specific context to make an inference or to drive an action. There are three classes of data that we deal with in analytics, data science, and AI:

• Structured data: These are data whose meaning is agreed upon ahead of time and is defined at an atomic level. For example, a table containing names and contact information for customers would typically be structured data. It can be arranged in rows and columns where each row pertains to a different customer, and each column i...