By: Felicia Guo
New York–based product designer Ginny (Yuanfei) Zhao is emerging as a leading innovator at the intersection of artificial intelligence, financial analysis, and human-centered design. Her newly published research, “Generative AI Interface Design Considerations for Private Equity,” presented at the 2025 ACM Conference on Intelligent User Interfaces (IUI) in Cagliari, Italy, offers one of the first empirical examinations of how private equity professionals actually use generative AI—and what design principles are needed to make these tools reliable in high-stakes financial environments.
Zhao serves as a senior product designer at Kensho Technologies, the AI innovation arm of S&P Global. Her work spans AI-powered transcription systems, LLM-driven chat interfaces, and data platforms supporting cancer research. With academic training in Human–Computer Interaction and Statistics & Machine Learning from Carnegie Mellon University, she brings a rigorous, multidisciplinary approach to designing intelligent systems that must operate with transparency, precision, and trust.
Her new study addresses an increasingly urgent question: how can generative AI meaningfully support private equity workflows, where analysts depend on accurate information, time-sensitive data, and verifiable insights? Although generative AI has surged across industries, the finance sector—particularly private equity—still lacks research-based design frameworks suited to its unique operational demands. Zhao’s publication helps fill this gap through a rare mixed-methods investigation grounded in real user behavior rather than theory.
Working with a functional GenAI prototype created for private equity research and due-diligence tasks, Zhao and her coauthors analyzed 825 real chatbot queries, 12 in-depth interviews, in-app user feedback, and system interaction logs. The findings reveal significant disconnects between how analysts attempt to use generative AI and how the technology interprets their intent. The study shows that most users approached the system as if it were a traditional search engine. As a result, 94 percent of all queries resembled keyword searches, leading to vague or inaccurate responses that could not support complex financial reasoning. Interview participants also described recurring challenges with unclear intent interpretation, hallucinated outputs, and confusion surrounding the recency and reliability of financial data.
These results point to a deeper structural issue: private equity analysts bring well-established mental models from search tools and databases, while generative AI relies on a fundamentally different interaction logic. Without proper interface design, these mismatches undermine trust and limit the technology’s usefulness. In response to these challenges, Zhao’s paper introduces a domain-specific design framework for generative AI in private equity. The framework emphasizes clearer prompt guidance, transparent sourcing to support auditability, hybrid workflows that combine traditional machine learning with generative reasoning, and explicit communication of dates, time sensitivity, and data freshness. It also highlights the importance of designing interfaces that mirror the rhythms and responsibilities of analyst workflows.
The significance of this work extends beyond private equity. As financial institutions face mounting pressure to adopt AI responsibly, Zhao’s research offers practical guidelines for building systems that meet compliance, risk-management, and accuracy expectations. The insights also apply to adjacent fields such as venture capital, asset management, hedge funds, and M&A, where analysts rely on trustworthy document synthesis and transparent data interpretation.
Looking ahead, Zhao plans to continue expanding this line of research by validating the framework with broader user testing and refining the prototype based on real-world task performance. Future directions include adapting the framework to other financial sectors and integrating generative AI directly into analysts’ everyday tools—such as PDF readers, spreadsheets, and browser extensions—to create more seamless and context-aware assistance. The research also calls for deeper exploration into the temporal reasoning limitations of large language models and the potential of hybrid architectures designed for greater accuracy and reliability.
With this publication, Zhao demonstrates both technical and creative leadership within the rapidly evolving field of AI-driven finance. Her work marks an important step toward building intelligent tools that financial professionals can trust—tools that are not only powerful but also transparent, responsible, and aligned with the realities of modern analytical work.
Disclaimer: The content provided is for informational purposes only and does not constitute investment, financial, or professional advice. Readers should consult with qualified financial or investment professionals before making any decisions based on the information presented. The outcomes and insights discussed may vary depending on individual circumstances and market conditions.
