Abstract: The lean startup methodology is one of the most popular product development paradigms. It ¯emphasizes the concept of product-market fit and quick build-measure-learn iterations to reduce the uncertainty about how the market will react to the product. In most traditional software products, there is no uncertainty around the ability to build the product as specified: software engineering is deterministic. But nowadays, more and more products are heavily dependent on machine learning and AI components. This adds a second source of uncertainty to the product development process: we can’t guarantee model performance in advance.
This added uncertainty raises new product development challenges. For example, what are the acceptance criteria for a machine learning model? What is the Definition of Done? If a modeling effort is unsuccessful, should it lead to a product pivot? The goal of this talk is to provide a unifying framework to tackle these challenges. We introduce the concept of product-data fit and describe how modeling iterations should interact with the product development cycle. We emphasize the flow of validated learning from the product side to the data side and vice versa, and how product development drives the choice of modeling metrics. We discuss several use cases, highlighting guiding questions and principles to help achieve product-data fit and product-market fit while avoiding common pitfalls. We also illustrate how this approach can shorten time to market and help achieve financial business goals of AI driven products.
Bio: Daniel is a data science executive with over 8 years of experience leading product-oriented data science organizations in building products from inception through scaling. Daniel has deep expertise in the healthcare technology and biotech spaces, having led data science and R&D teams in companies from early stage ventures to large organizations. As VP of data science at eviCore healthcare, he currently oversees multiple strategic data initiatives resulting in multimillion-dollar savings. Daniel also consults to companies across various verticals on data and product strategy. He holds a PhD in mathematics from Princeton University, and his computational biology contributions have been published in Cell, Nature Biotechnology, and other top journals.