As data scientists, we are constantly focused on learning new ML techniques and algorithms. However, in any company, value is created primarily by making decisions. In this talk I present a systematic process where ML is an input to improve our ability to make better decisions, thereby taking us closer to the prescriptive ideal. In a nutshell, this process starts by clearly identifying the KPI or metric that we want to improve (eg. revenue). This metric itself may not be actionable, so we may have to decompose it into actionable metrics. Importantly, we then move backward and identify the levers that we can pull to improve these intermediate metrics. Unfortunately, most decisions are made under uncertainty, so we generally don’t know ex-ante the effect that pulling a lever has on our metrics. This is where ML enters the picture: it allows us to embrace uncertainty in a data-driven way. The process is all about understanding the relationships between metrics, levers, and underlying uncertainty in a systematic way.
Daniel Vaughan is the author of “Analytical Skills for AI and Data Science”, published by O’Reilly in 2020. He has more than 15 years of experience creating predictive and prescriptive solutions in academia and the industry. He is currently Head of Data at Payclip, one of the leading fintech companies in Mexico, where he leads data science and analytics, data engineering, and BI. Before that, he was Head of Data Science in Latin America at Airbnb, Head of Big Data at Telefonica Mexico, senior data scientist at Banorte, and senior research economist at Banco de Mexico. He holds a Ph.D. in economics from NYU (2011).