This past week saw the launch of Anthropic’s Claude for Financial Services and OpenAI's “ChatGPT agent” that can generate Excel spreadsheets and PowerPoint decks, among other capabilities. (XLS and PPT fall under Microsoft’s Open Specification Promise not to sue.) ChatGPT agent follows OpenAI’s quiet launch a couple months ago of a website section focused on industry-specific solutions for financial services, healthcare, and other sectors. The application of AI to financial services isn’t exactly new, especially among fintech startups and financial institutions. However, these forays from major AI players signal a maturing and verticalization akin to what happened with the cloud in the late 2010s and early 2020s.

Claude for Financial Services is a “tailored version of Claude” for the enterprise, and specifically for financial services. This version has been trained on finance domain knowledge, including large-scale data analysis, financial reasoning, and Excel manipulation. According to the offering’s product lead, it can produce pitch decks, investment memos, benchmark analyses, stock price and volume charts, discounted cash flow (DCF) models, and other Excel and PowerPoint documents, even referencing internal templates to produce polished, on-brand deliverables. The solution links directly to original data sources for verification.

According to Anthropic, Claude 4 models (Opus, Sonnet) as a family outperform OpenAI’s frontier models on independent benchmarking platform Vals AI's Finance Agent benchmark. (On an individual model basis, OpenAI’s o3 is currently topping the leaderboard, followed by Claude Sonnet 4.) When tested against Financial Modeling World Cup (FMWC) tasks, Claude Opus 4’s Excel agent was able to pass 5 of the 7 levels, scoring 83% on complex Excel tasks.

Claude for Financial Services brings in real-time data through integrations (pre-built MCP connectors) with partners such as Databricks, Snowflake, S&P Global, PitchBook, Morningstar, FactSet, Palantir, Box, and Daloopa. (Customers need subscriptions to get access though.) Anthropic says it doesn’t train models on customer data by default. The offering is already available on AWS (Amazon Web Services) Marketplace, and will soon be on Google Cloud Marketplace. Anthropic will provide customers with 6 weeks of hands-on training, and has partnered with consulting firms (Deloitte, KPMG, PwC, Turing, TribeAI, Slalom) to streamline implementation within enterprises.

AI is a natural fit with the financial services industry – a high-value, intensely analytical realm with an abundance of structured data. According to Anthropic, the financial-services companies already using its offering in research preview include Ray Dalio-founded Bridgewater Associates (used in its Investment Analyst Assistant); insurance giant AIG (which reported Claude compressing the timeline for underwriting reviews by more than 5x and improving data accuracy from 75% to 90%+); Norway’s sovereign wealth fund NBIM (which automated news monitoring on 9,000 companies and saw 20% productivity gains, equivalent to 213K hours saved); Commonwealth Bank of Australia; and private-equity firm HG Capital.

Anthropic’s pursuit of business-oriented use cases is driving the growth of its recurring revenue. In just the past month alone, Anthropic’s annualized revenue has grown from $3B to $4B, likely on the back of its code-generation capability. Claude’s popularity for coding among software engineers is likely to give it a leg up in the financial services realm, where IT has an expansive footprint and is typically a major cost line for big firms. Anthropic’s new offering integrates Claude Code to make it easier to analyze very large datasets, conduct Monte Carlo simulations, and model risk. (Users of the offering will enjoy expanded context windows and usage limits.)

This week also saw Anthropic rival OpenAI’s launch of a more general-purpose “ChatGPT agent” – a cross between its Operator browser-using agent and its Deep Research capability. Available to paid subscribers, ChatGPT agent is intended for hard tasks that might require 10-15 min using its own “virtual computer.” (Users can even close their laptops and go out for lunch while it’s working.) Examples of use cases include collecting publicly available data and building a downloadable Excel spreadsheet; researching options and government incentives for a new office in Singapore; briefing on upcoming client meetings based on recent news; or creating a slide deck based on a competitor analysis. In one test by industry watcher Ethan Mollick, ChatGPT agent was able to come up with a startup idea, build a multi-tab cost model, and develop a pitch deck for the startup. Mollick noted that the agent still needed a “manager” to catch errors, provide feedback, and help iterate.

In the near term, these offerings are more a threat to Bloomberg than to individual financial analysts, who are likely to use these AI tools to be more efficient. Anthropic says its offering still needs a “human in the loop.” Because of the high stakes in financial services, less-than-perfect accuracy can be an issue for extended tasks. In coding, for instance, if there is a 10% probability of an AI error in every 10-min step, then the overall success rate after 1 hour would be 53% and the success rate after 10 hours would be just 0.002%.

However, similar to what’s been happening in the IT realm, the jobs available for junior financial analysts could be eroded over time. Anthropic’s own CEO believes that AI could “wipe out half of all entry-level white-collar jobs” and spike unemployment to 10-20% within the next 5 years. The bar will be raised in terms of per-employee productivity, with some bifurcation emerging between experienced employees capable of “context engineering” to wield AI effectively vs. junior staff still developing their judgment.

On the other hand, knowledge workers can be highly adaptable and it’s only a matter of time before many of them develop the AI chops to meet the new bar for productivity. Not all of them will go back into large companies, however. Given the tools that will become available in this next era of AI, many of these financial-analyst types will instead pursue the swath of novel opportunities that emerge – and some may even delve into “vibe investing.”

Apr 18 2025 (3 Shifts): The growth of AI big tech and $200-per-month subscriptions

Feb 14 2025 (3 Shifts): “Deep research” tools everywhere