Dr. Sai Zeng, Head of AI CoE & Distinguished Engineer at UBS, at AI for Financial Services NYC – AI Successes in Investment Management

About Sai Zeng: At UBS she leads the implementation of AI-driven strategies to enhance data management, risk reduction, and operational efficiency across the bank’s group functions.

About UBS: UBS is a global financial services company offering wealth management, asset management, and investment banking services. It operates in major financial centers worldwide, providing tailored financial solutions to a diverse clientele​.

Replicating senior data scientists and integrating diverse talents in machine learning, deep learning, and NLP are crucial for scaling AI technology. Co-developing solutions with development teams and business stakeholders promotes a “full stack data scientist” approach.

AI and machine learning are viewed as extensions of existing products and solutions, not standalone products. This encourages close collaboration with development teams and product owners to integrate AI models within existing frameworks.

Teams dedicate about 20% of their capacity to innovation and experimentation with new large language models (LLMs) and technologies. This structured approach ensures problem-driven experimentation aligns with organizational goals, balancing the need for innovation with delivering measurable progress and value.

AI applications include a legal document search tool that improves search functionality for 20 million documents and an auto-commentary tool for financial controllers that automates data aggregation and commentary generation. These tools enhance operational efficiency and address specific business needs.

Hiring people who understand AI algorithms and can integrate capabilities with product solutions is key. This involves a “full stack data scientist” concept, where professionals know AI, software development, deployment, and scalability in production, ensuring a holistic approach from the beginning.

Dedicated time for innovation is driven by real problems. Following the acquisition of Credit Suisse, a specific use case involved HR queries about comparing 401(k) plans. This exploration extended beyond simple questions to complex multi-agent frameworks, inspiring further exploration and innovation.

AI in financial services focuses on operational efficiency and cost savings, with benefits like improved customer experience and sales/marketing. AI applications in internal FAQs chatbots, automated reporting, and automation in areas like anti-money laundering (AML) and know-your-customer (KYC) processes are rapidly moving into production.

A legal document search tool faced challenges like handling 20 million documents and implementing access control. The team developed a Google-like search for legal documents, focusing on scalability and access control. The journey from proof of concept to production took over a year.

An auto-commentary tool for financial controllers automates data aggregation and commentary generation. This tool streamlines the process of explaining significant profit and loss (P&L) changes, replacing manual data sorting with automated solutions. It integrates large language models to discover language patterns and generate daily comments.

AI models follow established approval processes similar to risk models. Providing evidence of input, output, and metrics ensures compliance with governance standards. Large language models need new validation approaches, reflecting the evolving nature of AI technologies.

Ensuring accuracy without hallucination is a challenge, particularly in structured data environments. Solutions involve using AI models to interpret data and compose responses, minimizing errors. Interaction with large language models focuses on composing answers based on accurate data extracted via SQL statements.