What Happened
Prompt engineer and AI educator Jai Sharma (@aiwithjainam) has published a thread of 13 detailed prompts designed to make Anthropic's Claude 3.5 Sonnet perform structured financial analysis. The prompts are presented as tools to make the AI "think like a Goldman Sachs analyst," guiding it through tasks such as building Discounted Cash Flow (DCF) models and generating earnings reports.
The prompts are not a new feature or model from Anthropic, but rather a demonstration of advanced prompt engineering applied to the existing Claude 3.5 Sonnet model. They leverage the model's 200K context window and its stated improvements in coding, task handling, and nuanced instruction-following.
The Prompts and Their Structure
Based on the shared thread, the prompts are designed to enforce a specific, analytical reasoning process. Examples include:
- "Build a 3-Statement Financial Model": A prompt that instructs Claude to start with historical data, project revenue drivers, build out income statements, balance sheets, and cash flow statements, and link them together.
- "Perform a DCF Valuation": This prompt would guide Claude through calculating free cash flow, estimating a terminal value, selecting a discount rate (WACC), and arriving at a present value.
- "Write an Earnings Report Analysis": A prompt to synthesize financial results, analyze key metrics versus expectations, and provide a summary of management commentary and forward guidance.
The common thread is the use of step-by-step instructions, clear output formatting requests (like tables), and defined roles ("Act as a senior investment banking analyst"). This structure aims to reduce hallucinations and produce consistent, usable financial outputs.
Context and Limitations
This development sits at the intersection of two trends: the increasing capability of large language models (LLMs) in structured reasoning and the growing field of prompt engineering as a specialized skill. Claude 3.5 Sonnet, released in June 2024, was specifically benchmarked by Anthropic as outperforming its predecessor on graduate-level reasoning, coding, and nuanced instruction.
Important Caveats:
- The prompts require the user to input accurate, historical financial data. The model's role is analysis, calculation, and projection based on that data, not sourcing the data itself.
- The outputs are suggestions and models built on the provided data and the model's training. They are not certified financial advice and carry the standard risks associated with LLM outputs, including potential calculation errors or misinterpretations.
- This is a community-driven prompt engineering showcase, not an official financial product from Anthropic. The model's core capabilities have not changed.
For financial professionals and analysts, these prompts represent a template for leveraging AI as a productivity tool for drafting, calculation, and initial analysis, potentially speeding up the early stages of model-building and report drafting.
