Anthropic Just Invaded Excel—and Wall Street Is the Real Target

Anthropic isn't building a chatbot for finance. It's building a replacement for junior analysts. The company just unveiled Claude for Excel, embedding its AI directly into spreadsheets with the ability to read, analyze, modify, and create workbooks while showing its work at the cell level. It also announced integrations with Microsoft Copilot Studio, plus six new data partnerships—LSEG (London Stock Exchange Group), Moody's, MT Newswires, Aiera, Chronograph, and Egnyte—giving Claude access to live market data, credit ratings, earnings transcripts, and proprietary research. And it launched six pre-configured "Agent Skills" designed to automate the daily grind of Wall Street: DCF models, comp tables, data room processing, earnings analysis.

According to VentureBeat's reporting, Anthropic's Sonnet 4.5 now tops the Finance Agent benchmark at 55.3% accuracy on tasks expected of entry-level financial analysts. That number sounds low—until you realize it's state-of-the-art, and it represents the compression of an entire job category into software.

This isn't about productivity gains. It's about headcount reduction. And Anthropic just told every financial institution in the world that the math works.

Excel Is the Battlefield Because Excel Is Where the Money Lives

Anthropic's decision to build directly into Excel is surgical. Excel isn't just software—it's the operating system of finance. Every valuation model, every LBO, every risk analysis lives in a spreadsheet. Analysts spend their entire careers in Excel. Some never leave. So Anthropic went there.

Claude for Excel works in a sidebar where it can manipulate workbooks, preserve formula dependencies, debug cells, populate templates, and build new models from scratch. It doesn't just answer questions about your data—it modifies the models that drive trillion-dollar investment decisions. And critically, it shows its work. According to the announcement, Claude tracks and explains changes at the cell level, letting users "navigate directly to referenced cells."

This addresses the "black box" problem that's kept AI out of high-stakes finance. When billions of dollars ride on a model's output, you can't just trust the AI—you need to audit its logic. By making every change transparent and traceable, Anthropic is building the trust necessary for institutions to replace humans with software. Which is exactly what they're doing.

Norway's $1.6 trillion sovereign wealth fund (Norges Bank Investment Management) reported 20% productivity gains, equivalent to 213,000 hours. AIG compressed review timelines by 5x while improving data accuracy from 75% to over 90%. These aren't pilot programs. These are production deployments at institutions managing trillions in assets. And they're publicly endorsing the platform.

The Data Moat: Why Access Beats Algorithms

Here's what matters more than the Excel integration: Anthropic just locked down privileged access to the financial information infrastructure. The six new data partnerships—plus the seven announced in July (S&P Capital IQ, Daloopa, Morningstar, FactSet, PitchBook, Snowflake, Databricks)—give Claude access to virtually every category of financial data an analyst needs: fundamental company data, market prices, credit assessments, private company intelligence, alternative data, breaking news.

This is a land grab. Generic LLMs trained on public internet data can't compete with systems that have direct pipelines to Bloomberg-quality financial information. Anthropic is building data moats that competitors will struggle to replicate. Because the quality of AI outputs depends entirely on the quality of inputs, and Anthropic just secured the inputs that matter.

The strategic bet is clear: domain-specific AI systems with privileged data access will outcompete general-purpose assistants. It's a direct challenge to the "one AI to rule them all" approach. And it's working. When LSEG gives you live market data, Moody's gives you credit ratings, and Aiera gives you real-time earnings transcripts, you're not just another chatbot. You're infrastructure.

Agent Skills: Productizing the Entry-Level Analyst

The six new Agent Skills are Anthropic's attempt to automate the career ladder. Each skill targets a specific task that junior and mid-level analysts spend their days doing:

• DCF models with full free cash flow projections, WACC calculations, scenario toggles, sensitivity tables
• Comparable company analysis with valuation multiples and operating metrics
• Data room processing into Excel spreadsheets with financial info, customer lists, contract terms
• Company teasers for pitch books and buyer lists
• Earnings analysis using quarterly transcripts and financials
• Initiating coverage reports with industry analysis, company deep dives, valuation frameworks

These aren't generic "AI assistance." These are solutions to specific, well-defined analyst tasks. And they package AI capabilities in terms financial institutions already understand. You don't buy "AI." You buy "the thing that builds DCF models so I don't need to hire two analysts."

The 55.3% accuracy on the Finance Agent benchmark is both promising and limiting. It's state-of-the-art, but it's not reliable enough to operate autonomously. Which is actually perfect for adoption. CFOs don't want to replace humans entirely—they want to make each human 5x more productive so they can reduce headcount through attrition. A 55% accurate AI that still requires human oversight is politically safer than a 95% accurate AI that doesn't.

The Uncomfortable Economics: Productivity Gains Mean Fewer Jobs

Let's do the math. If Norway's sovereign wealth fund achieved 20% productivity gains (213,000 hours), that's the equivalent of eliminating roughly 100 full-time employees. They didn't fire anyone—yet. But when attrition happens, those roles don't get backfilled. The AI absorbs the work.

AIG compressed timelines by 5x. That's not "everyone works faster." That's "we need one-fifth the people." Again, they're not announcing layoffs. But the economic logic is undeniable: if one analyst with Claude can do the work of five without it, you don't hire five.

This is the productivity paradox of AI: it makes individuals more efficient, but it makes teams smaller. Anthropic's pitch is "your analysts will be 20% more productive." The CFO hears "I can reduce my analyst pool by 20% and maintain output." That's not a bug. That's the business model.

And financial institutions are incentivized to move fast. According to the article, the industry is projected to spend $97 billion on AI by 2027, up from $35 billion in 2023. The companies that adopt early and reduce costs will outcompete those that don't. This creates a race to automate, regardless of whether individual firms want to.

The Regulatory Vacuum and the Liability Trap

Here's the twist: Anthropic is deploying this into a regulatory vacuum. The Consumer Financial Protection Bureau issued AI guidance in 2023, then had it revoked under the current administration. According to the Brookings Institution analysis cited in the article, we've swung from the Biden administration's cautious AI development framework to the Trump administration's "cement U.S. dominance through deregulation" approach.

This creates both opportunity and risk. Less federal oversight means faster adoption. But it also means no guardrails. And when things go wrong—when the AI hallucinates a decimal point, when a model produces discriminatory lending outcomes—there's liability. Massachusetts AG just settled with Earnest Operations for $2.5 million over AI models that disadvantaged Black and Hispanic loan applicants. That's the preview of what's coming.

Anthropic knows this. Jonathan Pelosi, their global head of industry for financial services, emphasized that Claude requires "a human in the loop" and isn't intended for autonomous decision-making. But that's a legal hedge, not a technical constraint. The tools can operate autonomously. The question is whether institutions will take that risk—and whether regulators will stop them before the lawsuits start piling up.

What This Means for Everyone Not Managing a Trillion-Dollar Fund

If you're a financial analyst, this is your compression event. The skills you spent years building—financial modeling, comp analysis, data room processing—are being automated. Not because the AI is better than you. Because it's good enough, faster, and cheaper. Your competitive advantage can't be execution anymore. It has to be judgment, client relationships, and strategic insight the AI can't replicate. Yet.

If you're a CFO evaluating AI adoption, Anthropic just handed you the business case: 20% productivity gains, 5x faster timelines, 90% data accuracy. The ROI is undeniable. The question is whether you move first and gain competitive advantage, or wait and get disrupted by firms that did.

And if you're in any other knowledge work industry watching this unfold, here's your reminder: finance is the testing ground. If AI can handle the precision, complexity, and regulatory scrutiny of Wall Street, it can handle your industry too. The tools are coming. The question is whether you adapt early or get adapted around.

Want to build an AI strategy that accounts for what's actually happening, not just what vendors promise? Let's talk. Because the companies that win won't just adopt AI tools. They'll understand the economics, the risks, and the inevitable restructuring that comes with them.