Primary Financial Company manages an investment program for institutional investors to invest substantial funds in federally insured CDs. This break down to managing nearly 40,000 CDs and over $7 billion in assets while supporting relationships with 5,000 financial institutions and institutional investors.
They wanted to improve sales targeting to predict both CD issuing companies' funding needs and institutions' desire to invest. As a massive company with a large portion of the industry's market share, PFC had a vast quantity of valuable data at their fingertips, but they didn't know how to leverage it to look into the future to predict consumer behavior. Having partnered with Fusion on several projects in the past, they turned to us once again to explore how advanced analytics and machine learning in financial services could provide data-driven, predictive outcomes.
With a key objective in place, PFC and Fusion worked together to explore the following machine learning models:
With a plan in place, our machine learning experts outlined the process, including data acquisition, transformation, model development, and predictive analytics. First, all private and public data sources were identified and acquired to gain insight into current and prospective customers. PFC and Fusion then worked together to determine meaningful and available factors.
The next step was to transform the data so these factors would be consistent and accurate. With a solid foundation, Fusion developed machine learning models that would learn and identify patterns, then recognize those patterns when seen again to apply lessons to predict outcomes. We identified over 100 candidate "features" of data from both public and private data sources, then applied "practical analytics" which focuses on data that is applicable to the described use case.
Having a clear understanding of the target accuracy of the predictive models was essential to the success of this project, but more important was client participation. By collaborating with our client, PFC learned to quickly identify and understand errant data and numbers to drive the value of our models even higher. Also, by defining the required utility of the models, PFC could realize business value without having to endlessly tune them.