Data Science in Finance: Trends and Practical Applications Event Recap – New York
Digital transformation is a top agenda item for most banks, asset managers and insurance companies. As these businesses look to create more personalized experiences, reduce cost/optimize revenue, introduce new products, and mitigate risk, Data Science is a powerful tool to have in your arsenal to help solve business challenges.
This March, Synechron’s Round Table – “Data Science in Finance: Trends and Practical Applications” brought together experts Robert Huntsman, Synechron’s Global Head of Data Science to guide a conversation on Data Science best practices within banking including everything from data governance and model management to graph databases.
The speakers covered a range of topics including:
- Data Science in Finance Trends: What financial institutions are doing with Data Science, covering business problem statements, industry use cases, and Synechron’s experience working with global banks and insurance companies on Data Science initiatives
- Model Lifecycle Management: Building a Sustainable Data Science Practice within a Financial Institution that covers data governance and data lineage, model taxonomy, as well as regulatory auditability
- The Data Science Tech Stack: Including best practices for Spark, Hadoop, Tensorflow, Microservices, APIs, Cloud, and Graph Databases
- Enterprise Scalability: Making Data Science a repeatable program in an enterprise
Some of the key takeaways that participants learned from the discussion include:
- Firms are looking at a diverse set of use cases for data science which are moving away from Software-as-a-Service (SaaS) and involve things like building an end-to-end calculation engine for applications like underwriting and risk management.
- Banks are seeing a large ROI in applying data to reduce business operating costs by 20-25%.
- Publicly available data from sources such as SEC filings, weather data, news, and other sources can be read and understood using Natural Language Processing (NLP) and synthesized to help create new, alternative data sources.
- From a trade flow perspective there is a lot of emphasis on generating revenue and being able to match buyers to sellers and get down to a more granular level such as with thematic investing
- When asked, “Are you seeing deep learning, statistical analysis or clustering in your organizations?” attendees responded that predominantly clustering is the popular approach
- While some firms’ projects and programs are more advanced than others, lack of infrastructure, bottlenecks in productionizing models, proof of concepts having dissonance with enterprise scalability / lack of a repeatable process, and latency optimization all are top challenges
- Organizational culture also must be considered for an affective data science program where data is fuel and culture move that fuel into production. Regulations such as the General Data Privacy Regulation (GDPR) have made it more challenging to manage data in production systems, and that regulatory pressure is expected only to increase and needs to be factored into data lineage and model lifecycle management for auditable, compliant AI.
News, tweets and good stuff.