Neha Gupta, Sr. Manager ML/DS, Linkedin

Lifetime Value (LTV) of a customer is one of the key industry metrics used to measure the value of a customer and is commonly used to acquire, grow, and retain the “right” or profitable customers in an organization. In this talk, we share how Linkedin’s Go-to-Market data science team uses machine learning to understand and predict customer value by building a single, scalable platform to support all of Linkedin’s LTVs across different businesses. Three key platform components of this platform include i) data platform that unifies data into a few standardized tables that capture both modeling features as well as transactional data, ii) modeling library that provides the ability to plug and play different machine learning models to do an extensive back testing to get an estimate of model accuracy, and iii) production pipeline that can be efficiently run to update LTVs on demand (daily as needed). This new platform has led to a 50-60% increase in model accuracy and has substantially reduced the time to run/build new models/data pipelines.