Algorithms to Outcomes: AI's Impact in Finance | Global Conference 2025

Panel Composition & Firm Context

• Marsh McLennan (Michal Zoutkevich): AI adoption accounts for approximately one-third of the firm's business; the platform rolled out generative AI to all 85,000 employees two years ago.
• Citadel (Umesh Subramanian): A multi-strategy hedge fund splitting AI use into quantitative (machine learning for 10+ years) and discretionary (accelerating information access for human decision-makers).
• Pantera Capital (Katrina Paglia): A crypto blockchain venture firm using AI for discretionary investment thesis summarization and legal contract interpretation; focuses on confidentiality and regulatory compliance.
• Freestone Growth Partners (Daniel Morello): A fundamental equity long-short firm with co-founders previously at Citadel; integrated AI into discretionary processes from inception rather than as a bolt-on.
• Roquan (Andy Cruz): A quantitative firm founded in 2007 viewing AI as a natural progression of the discipline, emphasizing the shift from linear regression to deep learning and NLP.

Current AI Adoption & Practical Applications

• Efficiency & Automation:
  • Firms are using AI to automate commercial due diligence, potentially reducing execution time by 20% in specific workflows.
  • Marsh McLennan utilized AI to generate 15% more customers and 40% larger cross-sell figures for a banking client by analyzing customer journeys.
  • Legal teams at Pantera use AI to summarize lengthy regulations and contracts, flagging specific terms and definitions to replace manual page-flipping.
  • Quant researchers at WorldQuant use AI for code completion, debugging, and extracting structure from unstructured data (images, video, text) to find new alpha signals.
• Decision Support:
  • AI is used to process alternative data (e.g., scraping crypto Twitter for real-time project launches) faster than human analysts.
  • Discretionary teams leverage AI to provide context-rich information summaries, allowing portfolio managers to process news, filings, and supply chain data across 30–50 stocks rapidly.
  • Digital twins and predictive modeling allow for business simulations that previously took months to complete, now reduced to overnight results.
• Generative Capabilities:
  • Chatbots are enabling autonomous interactions, such as a chatbot talking to another chatbot to execute complex customer service or marketing journeys.
  • Tools are being used to draft employment agreements, emails, and blog posts, serving as a baseline for human editing rather than final output.

Strategic Shifts & Philosophical Approaches

• Quant vs. Discretionary Divergence:
  • Quantitative strategies have relied on machine learning since the 1980s (Jim Simons) and 2010s (TensorFlow), focusing on statistical prediction and backtesting.
  • Discretionary strategies face a risk of "consensus capture" because LLMs predict the most likely next token; firms must intentionally use AI for summarization without losing the nuance required for contrarian investing.
• Human Judgment & Alpha:
  • Panelists argue AI will not replace human judgment but will scale it; the future lies in "man-plus-machine" rather than "all-machine" systems.
  • True alpha is expected to derive from designing the overall firm culture and system (the "factory") to leverage AI, rather than just the tools themselves.
  • The "long-term" investing approach (e.g., Warren Buffett style) remains viable but must be distinguished from short-term automated reacting, which offers no sustainable competitive advantage.
• The "Base of the Mountain" Metaphor:
  • Multiple speakers (Cruz, Morello) described the current state of AI as the "base of the mountain," suggesting the technology is still in early adoption phases compared to its potential future impact.

Risks, Hype, and Limitations

• Overfitting & Explainability:
  • Significant risk exists of overfitting signals, especially with nonlinear models ("black boxes") that generate noise rather than actionable alpha.
  • Regulatory bodies and LPs demand explainability, creating tension with the opaque nature of deep learning models.
• Data Privacy & Confidentiality:
  • Lawyers and compliance officers express concern over proprietary data entering public AI models, necessitating "scrubbers" or private instances to create audit trails.
  • The "black box" of AI inputs makes it difficult to police what data traders or analysts are feeding into external tools.
• Intellectual Complacency:
  • A danger exists that junior employees or lawyers will rely entirely on AI for drafting and summarization, losing the foundational learning that occurs through manual review.
  • Firms warn against "intellectual laziness" where judgment is transferred to the machine without sufficient human oversight.
• Hype vs. Reality:
  • While AI is transformative, panelists disagree on whether it is as revolutionary as the invention of electricity; most view it as a high-impact tool requiring significant engineering and management effort to realize value.
  • The "dot-com boom" analogy is used to suggest that while the internet era had a bubble, the underlying productivity gains (like AI's) eventually create lasting value.

Regulatory, Legal, and LP Considerations

• Disclosure Standards:
  • Pantera Capital (Katrina Paglia) notes that while LPs do not need to know every model detail, firms must demonstrate sufficient oversight and fair disclosure to maintain trust.
  • Regulatory frameworks are lagging behind AI adoption, with new legislation expected in the U.S. (e.g., the new AI czar) to address asset management practices.
• Systemic Controls:
  • Firms are moving from managing "artisans" to "system operators," requiring rigorous engineering controls (e.g., running 10,000 tests) to ensure AI agents behave predictably.
  • Operational functions requiring precise correctness (risk numbers, trade matching) must be treated differently from probabilistic AI outputs.

Talent, Hiring, and Workforce Impact

• Hiring Criteria Evolution:
  • Hiring is shifting from testing specific tool knowledge to assessing adaptability, curiosity, and the ability to learn new tools rapidly.
  • Firms are actively embedding data scientists and AI professionals within investment groups to optimize workflows.
• Job Security vs. Task Evolution:
  • Panelists agree that while specific tasks (code writing, summarization, drafting) will be automated, jobs will evolve rather than disappear, similar to the impact of the internet and cloud computing.
  • There is a concern regarding the development of junior professionals who may miss out on the "grunt work" that traditionally builds expertise.
• Cultural Integration:
  • Older employees face a higher bar to adopt AI but can be convinced once they see tangible efficiency gains in their specific use cases.
  • Firms are experimenting with "low-hanging fruit" like note-taking and meeting summaries but must cultivate a culture that still values active human engagement and learning.

Forward-Looking Statements

• Unstructured Data Monetization:
  • The next phase of alpha generation involves structuring unstructured data (images, video, text) to create new predictive signals currently unavailable to quants.
• Agentic Workflows:
  • Future systems will move beyond passive tools to "agents" that can autonomously research, code, backtest, and report findings to researchers without constant human intervention.
• Systemic Transformation:
  • The finance industry will need to adopt "manufacturing logic," treating AI integration as a system engineering problem involving the design of processes, culture, and controls.
• Barriers to Entry:
  • The economics of finance are shifting; AI will lower barriers to entry for certain analytical tasks (like disease mapping in pharma or deal structures in banking), potentially changing microeconomic dynamics.