How Claude is transforming financial services

Key Concepts

• Claude for Financial Services: A specialized application of Anthropic's Claude AI model designed to address the unique needs and workflows of the financial services industry.
• Artifact Feature: A feature within Claude that allows for live dashboards and direct connections to data sets like S&P and FactSet, enabling dynamic metric comparison and updates.
• Model Context Protocol (MCP): A protocol that enables Large Language Models (LLMs) to interact with external systems and data sources, enhancing their capabilities beyond simple chat.
• Agentic Capabilities: The ability of an AI model to perform tasks autonomously, interact with tools, and make decisions based on context and available information.
• Retrieval, Analysis, Creation: The three core verbs that define the functionality of Claude for Finance, focusing on data extraction, processing, and output generation.
• Evals (Evaluations): A process of defining specific tasks and desired outcomes to measure and improve the performance of AI models in real-world enterprise scenarios.
• AI Safety: A foundational principle for Anthropic, encompassing secure deployment, accurate responses, and user trust through verification and auditability.
• Pre-training and Post-training: Specialized training of AI models on domain-specific data (finance in this case) to enhance their performance and understanding of industry nuances.

Claude for Finance: Transforming Financial Services with AI

This discussion explores the evolution and application of Anthropic's Claude AI model within the financial services sector, highlighting its capabilities, adoption, and future trajectory. Alexander Bricken, leading Applied AI Engineering for Financial Services, and Nick Lin, Product Lead for Claude for Financial Services, share insights into how Claude is moving beyond theoretical curiosity to practical, production-level deployment in finance.

The Shifting AI Landscape in Financial Services

Nick Lin observes a significant shift in the enterprise AI landscape, particularly in the past few months, moving from passive observation to active building and deployment. While coding was an early domain with strong product-market fit for AI, this is now extending to other verticals, including finance.

• Example: The Norwegian Sovereign Wealth Fund (NBIM) has built integrations using Model Context Protocol (MCP) for their portfolio managers to query daily insights into their portfolio companies (PortCos). This allows analysts to spend less time on mundane tasks and more on relationship building and business model understanding.

Enhanced Analyst Productivity and Work Transformation

The core benefit of Claude for Finance lies in its ability to automate tedious tasks, freeing up financial analysts to focus on higher-value activities.

• Traditional Approach: Analysts often perform static analysis in Excel sheets, requiring manual refreshes weekly or quarterly.
• Claude's Approach:
  • Artifact Feature: Enables direct connection to data sets like S&P and FactSet, creating live dashboards for metric comparison.
  • Simplified Updates: A single prompt to Claude can easily update these artifacts.
  • Collaboration: These artifacts are shared with managing directors, facilitating direct interaction with platforms.
  • Outcome: This leads to not just acceleration of work but a fundamental transformation of how work is done.

Claude's Core Capabilities: Retrieval, Analysis, and Creation

Claude for Finance is built around three key verbs that define its functionality:

1. Retrieval:

   • LLMs are adept at digging into large data pools and gathering insights rapidly (estimated 5,000 times faster than humans).
   • Claude for Finance focuses on connecting to core data sources used by financial analysts.
   • Key Advantage: The ability to uncover insights faster than competitors is a critical advantage in finance.

2. Analysis:

   • The ability to perform analysis at scale, whether through code or spreadsheets, is foundational.
   • Financial models are not just spreadsheets but a means for analysts to inject their judgment about future outlooks and valuations.
   • Claude is designed to understand core finance concepts and manipulate systems like Excel for calculations.

3. Creation:

   • Work in enterprises is often shared, making client-ready and boardroom-ready outputs crucial.
   • Claude aims to generate outputs in formats like spreadsheets, PowerPoint, and Word.
   • Goal: To be an end-to-end agentic autonomous system.

Technical Foundations and Differentiators

Claude's strength in finance stems from its underlying capabilities and Anthropic's approach to AI development.

• Code Proficiency: Claude's strong foundation in code provides a flexible skill and a shortcut for complex tasks.
  • File Creation Feature: Enables Claude to create Excel and PowerPoint documents by running Python code at scale within a virtual machine, facilitating tasks like creating DCF models.
• Model Primitives: Claude has built-in primitives, including security, designed to govern its interaction with the world.
  • Training: Models are trained to be helpful, harmless, and honest, reflecting the data they interpret and the outputs they generate.
• Safety and Trust: Safety is paramount, encompassing:
  • Secure Deployment: Ensuring solutions are deployed securely in enterprise environments.
  • Accuracy and Fidelity: Models accurately answer questions with the right level of understanding.
  • Trust and Auditability: Providing users with verification and auditability of results.
• MCP and System Interaction: The development of MCP allows Claude to interact with systems analysts care about, moving beyond simple chat interfaces. This enables a snowball effect where data from one system (e.g., Snowflake) can be used to interact with another (e.g., Salesforce).

The Three Layers of Claude for Finance

The Claude for Finance solution is structured around three key layers:

1. Models:

   • Anthropic's research focus is on making Claude the best model for financial services.
   • Customer Collaboration: Working with early customers like BCI, Perella Weinberg, and NBIM is crucial to understand finance-specific use cases, define "good" performance, and identify research gaps.
   • Domain Expertise: While Anthropic has strong coding expertise, they acknowledge the need to build finance domain knowledge through customer partnerships.

2. Agentic Capabilities (Product):

   • This refers to the code that enables user interaction with the models.
   • Deep Research: A key capability being developed.
   • Ubiquitous Embedding: Investing in embedding Claude across core work surfaces like Claude.ai, browser extensions, Excel, and Chrome.

3. Platform:

   • Building a flexible platform that can be easily tailored and deployed.
   • Industry Partnerships: Collaborating with S&P, FactSet, and PitchBook to build integrations that enhance agent power.

Adoption and Customer Success Stories

Adoption of AI in finance is less about specific sub-verticals and more about a customer's culture, requiring a blend of top-down encouragement and bottom-up experimentation.

• BCI (Example):
  • Transformation: Fundamentally transformed their work processes.
  • Comps Analysis: Traditionally done statically in Excel, requiring manual updates.
  • Claude Implementation: Used the Artifact feature to connect to S&P and FactSet for live metric comparison dashboards.
  • Impact: Accelerated work and transformed workflows.

Memory and Context in AI Models

Memory is crucial for AI models to maintain context across various tools and surfaces, similar to human memory.

• Importance for Finance: Allows Claude to understand and maintain context when interacting with different tools (Claude.ai, Excel, browser) and data sources (FactSet, S&P).
• Functionality:
  • Understanding patterns and user preferences (e.g., a specific DCF template).
  • Learning from user feedback (e.g., correcting a formula).
  • Remembering specific preferences for calculations (e.g., using S&P for EBITDA).
• Analogy: Functions like a good intern, remembering details and preferences.

Future of Claude for Finance: Research and Product Evolution

Anthropic is committed to an enterprise-first approach, focusing on specific domains like finance.

• Research:
  • Investing in specialized pre-training and post-training for finance.
• Product:
  • Deeper Sub-Vertical Focus: Understanding the nuances of private equity, hedge funds, insurance, and investment banking workflows.
  • Ubiquitous Presence: Expanding Claude's availability within Excel, PowerPoint, and other enterprise applications, with a focus on improving output quality.
  • Partnerships: Continuing to foster industry collaboration, noting the rapid development of MCP servers by leaders like S&P and FactSet.
  • Customer Collaboration: Working closely with enterprise customers to translate their needs into research and product development.

The Role of Evals in AI Development

Evals are critical for understanding how AI models perform in production and for guiding development.

• Definition: Evals are simply tasks that matter to users and problems they want to solve, with clear articulations of what constitutes success.
• Customer Engagement: Enterprise customers are encouraged to be thoughtful about defining these problems rather than broadly aiming to "infuse AI."
• Integration: Evals are directly integrated into the training process and product pipeline to deliver relevant capabilities.

Conclusion

Claude for Finance represents a significant advancement in applying AI to the complex and regulated financial services industry. By focusing on core capabilities like retrieval, analysis, and creation, coupled with a strong emphasis on safety, enterprise integration, and continuous customer collaboration, Anthropic is enabling financial professionals to transform their workflows, accelerate insights, and achieve greater efficiency. The future development will involve deeper domain specialization, broader platform integration, and a continued commitment to co-creation with industry partners and customers.