Minding the Machines. Jeremy Adamson

Чтение книги онлайн.

СКАЧАТЬ a pharmacist make you healthier? Would hiring an actuary increase your longevity?

In the last 20 years, the once sleepy academic fields of statistics, operations research, and decision support have exploded and been rebranded using terms such as data science, artificial intelligence, big data, and advanced analytics, among others. These practices have matured, and they have entered the mainstream. Advanced analytics and AI have moved from being an investment in the future to a core component of corporate strategy and a key enabler for all areas of the business. Regardless of their size or the industry or sector in which they are used, these technologies are becoming an organizational necessity throughout the world. Organizations use advanced analytics and AI to develop new products, understand their customers, control their costs, and make better-informed decisions.

      This new corporate function has been integrated in several support and core functions and has quickly become indispensable. Analytics is expected to add $16 trillion US to the global economy by 2030 and companies are eager to realize some of that value (PwC, 2017). As a result there has been a surge in demand for practitioners. There are approximately 3.3 million people employed in analytics in North America, and this is projected to grow by 15 percent a year in the United States over the next decade according to the US Bureau of Labor Statistics (2020). Educational institutions are eager to meet this demand.

      Essentially every major college and university in the world offers some sort of analytics program or specialization, within multiple faculties such as mathematics, engineering, or business. There are several hundred books published in this space, and it enjoys a highly active online community. These resources are strong, edifying, and comprehensive and cover every new technology, framework, algorithm, and approach. With such an active community, new algorithms and methodologies are packaged and made publicly accessible for tools such as R, Python, and Julia, almost immediately after being developed. The best and brightest are choosing to enter the field, often called the “Sexiest Job of the 21st Century” (Davenport & Patil, 2012).

      So, with overwhelming demand and a staggeringly capable pool of talent, why are there so many failures? Why are most organizations struggling to unlock the value in data science and advanced analytics? With so much executive support, so much talent, so much academic focus, why are so few organizations successfully deploying and leveraging analytics? In the 1980s, economist Robert Solow remarked that “you can see the computer age everywhere except in the productivity statistics.” Why now can we see data science transforming organizations without a commensurate improvement in productivity?

The fundamental reason is in the opening questions. Clearly, having five engineers on your payroll will not improve buildings in the vicinity. Even if directly requested, ordered, mandated, or incented to “improve a building,” they will struggle to do so. Replaced with any other profession, it is clear why this cannot work, but it persists as the typical furtive first steps into analytics. This is the essence of what most organizations have done with their data analytics team. They have hired talented, passionate, ambitious data scientists and asked them to simply “do data science.” Advanced analytics and AI as a practice has been poorly defined in its scope, has been subjected to great overspending, and has been lacking in relevant performance measures. Because of this, it is failing to live up to its potential.

      Effectively all organizations realize the benefits of analytics. In a survey by Deloitte in 2020, 43 percent believed their organization would be transformed by analytics within the next 1 to 3 years, and 23 percent within the next year (Ammanath, Jarvis, & Hupfer, 2020). Though most organizations are on board with analytics being a key strategic advantage, they are unaware of how exactly to extract value from the new function.

      Short-tenured data scientists, employed in a frothy and competitive market, share stories of unfocused and baffled companies where they have been engaged in operational reporting, confirming executive assumptions, and adding visualizations to legacy reports. Uncertain what to do with the team, and in a final act of surrender, the companies no longer expect the function to “do data science” and transform the team into a disbanded group of de facto technical resources automating onerous spreadsheets in a quasi-IT role.

      Contrasted with those organizations who have truly got it right, the differences are stunning. For several companies, well-supported analytical Centers of Excellence are a key team, perpetually hiring and growing, and are solicited for their input and perspectives on all major projects. In others, internal Communities of Practice encourage cross-pollination of ideas and development opportunities for junior data scientists. New products are formed and informed after a thorough analysis by data scientists, who are also supporting human resources with success indicators and spending Fridays pursuing their transformative passion projects. Theories and hypotheses are quickly tested in a cross-functional analytical sandbox. Individuals are sharing their work at conferences and symposia, building eminence, and gaining acclaim for the organization.

The irony is that while most analytics teams exercise great care in deconstructing a problem, modeling each of the elements, and developing robust simulations and marvelously elegant solutions, the business processes that they employ in their project delivery are often at best immature and at worst destructive. The challenge for any leader is to encourage and stimulate logical thinking from a lens of interconnectedness and value creation, and to direct that philosophy to execution as well. As the logician Bertrand Russell said, “What is best in mathematics deserves not merely to be learnt as a task, but to be assimilated as a part of daily thought, and brought again and again before the mind with ever-renewed encouragement” (Russell, 1902).

      The title of this book, Minding the Machines, is meant to be an affectionate recursion. Those talented and creative practitioners, craftspeople, data scientists, and machine learning engineers, who create the algorithms that are transforming the way business is done, mind and care for those machines like a shepherd. Those machines need to be trained, informed, given established processes, encouraged to be broadly interoperable, and developed to be applicable to many different situations and problems. Similarly, the practitioners themselves need to be minded, cared for, cultivated, and encouraged for both the team and the individuals to be successful. This concept of recursion occurs throughout the book.

      The second key theme is one of parsimony. Parsimony is a philosophy of intentionally expending the minimum amount of energy required in an activity so as to maintain overall efficiency. This is a key part of modeling; it is about keeping things as simple as possible, but no simpler. Similarly, and in the vein of recursion, teams themselves must be parsimonious. For analytics to mature as a practice while still delivering an accelerated time-to-value, teams need to be scrappy and lean.

      At a foundational level, the objective of this book is to provide clear insights into how to structure and lead a successful analytics team. This is a deceptively challenging objective since there are no generalized templates from which to work. Establishing a project management office, information services, or human resources department is an understood process and does not vary materially between organizations. Establishing an analytics team, by contrast, requires a significant up-front investment in understanding and contextualizing the initiative. Many organizations have attempted to use operating models and templates from other functions—often IT and operations research. This fundamental misunderstanding of where analytics fits within an organization has led to visible failures and has set back the analytical maturity of many organizations. Business leaders need to hire or develop value-centric talent who can step back from analysis and project management to view their work as existing within a network of individuals and teams with competing priorities and motivations.

Corporations, without a template or a default methodology to benchmark against, have made expensive missteps in building these teams by installing the wrong leaders, copying other functions, positioning the function under the wrong executives, incentivizing destructive behaviors, hiring the wrong people, and committing to the wrong projects. They have contracted expensive consultants to provide roadmaps that do not consider the unique culture or competencies of their organization. They СКАЧАТЬ