Data Science Strategy: Memahami pendorong perubahan di Data Science

Recognizing what to avoid when driving change in data science Around the globe, a number of businesses have made significant investments in data science, having realized (correctly) its revolutionary potential. Not having done their homework in terms of a proper situational analysis, however, 56 PART 1 Optimizing Your Data Science Investmentmany of these businesses have suffered huge losses rather than the expected benefits. Failing with the data science investment is particularly common among smaller and medium-size businesses. Why is that? Why are medium- and small-scale businesses unable to derive sufficient value by implementing data science? What obstacles stand in their way? In an attempt to come up with some answers to these questions, Computer Asso- ciates interviewed 1,000 IT managers across companies with more than half a billion dollars in revenue in a range of different industries, from retail to financial services to pharma. Their research findings revealed that the biggest obstacle by far is an insufficient infrastructure. Many times, companies are stuck with their legacy environment due to previous costly investments that “cannot be thrown away.” So, rather than creating a new, modern data architecture which puts the focus on the data, companies tend to add applications and system elements to their old environment, making data science inefficient and even more costly. The second largest obstacle is organizational complexity. This usually becomes a problem when the company management underestimates how transformative data science is. All aspects of the company must change in order to become data- centric, meaning that all managers across the company must understand and use data and new data-driven insights for decision making related to finance, mar- keting, sales, product and service development, and so on. However, in reality many companies tend to treat data science like a side business by adding new roles and functions to work with data rather than transforming existing functions and roles. The third most significant obstacle is security and other compliance concerns. This is not surprising, considering the growing awareness of the importance of handling data in a secure and ethically correct manner. New laws and regulations are becoming more and more strict in order to protect people’s right to privacy, and as long as there is still very little standardization in data science, require- ments will keep on changing. A general finding in the study was that, based on the type of analytics approach that was chosen, the level of resistance varied. That’s worth a closer look, so I walk you through some different types of high-level analytics projects in the fol- lowing sections. Then you can get a better sense of the major factors underlying the success (or failure) of a data science transformation project. CHAPTER 4 Managing Change in Data Science 57Descriptive analytics transformation projects Descriptive analytics projects involve tasks aimed at using data to describe what has happened or how things are right now — why, for example, we have sold x number of products this month of this specific product type. It includes activities such as developing graphs, charts, and dashboards, accompanied by no (or rela- tively simple) data analysis functions. The focus is on identifying the right set of metrics and presenting information in an effective manner. Descriptive analytics solutions generally face lesser resistance challenges in their implementations. The reasons are obvious — the deliverables are easily under- stood by stakeholders. However, it is sometimes difficult to justify the business value of descriptive analytics projects. At the end of the day, with the limited analysis happening in descriptive analytics, what is really the value of investing in understanding what happened yesterday when what you really want is to be prepared for tomorrow? Diagnostic analytics transformation projects The objective of diagnostic analytics projects is to understand the reasons for a particular phenomenon and to conduct a root cause analysis. Diagnostic analytics projects can culminate in the development of statistical models (explanatory models, causal models, and so on) and dashboards. However, the output must include insights and recommendations designed to help stakeholders understand the reasons for what’s happening and initiate appropriate actions. Organizations are usually receptive for analytical findings and insights based on diagnostic analytics outcomes, but there is a slightly higher resistance when it comes to implementing recommendations. This is mainly due to the fact that business users are aware that some recommendations aren’t actionable because they require too many changes or have too many restrictions. Predictive analytics transformation projects Predictive analytics projects involve forecasting a certain metric or predicting a certain phenomenon. Predictive modeling is the process of applying a statistical model or data mining algorithm on data for the purpose of predicting new or future observations. Predictive models can be used for not just predictions but also simulation purposes. Examples include clinical research, sales prediction, pro- duction failure, and weather forecasting. 58 PART 1 Optimizing Your Data Science InvestmentAs you might expect, predictive analytics solutions face the highest degree of resistance. Diagnostic and descriptive solutions largely deal with what has already happened, and predictive solutions relate to something that is yet to happen. Thus, business users have reservations about predictive solutions. This skepticism isn’t groundless, because the cost of making wrong predictions can be astonishing.