Can an AI agent do a financial analyst's job?

Short answer

Can AI replace financial analysts? Not as a whole job, based on public evidence as of November 2026. The tools cited here document AI that does parts of the analytical work: building and populating financial models, running comparable company tables, answering questions about a workbook, creating charts and drafting decks, and searching and summarizing research. In every case the vendor or the firm using it keeps a person in the loop. The parts of the job that carry a name and a judgment, such as recommending investments, advising clients and signing off on research views, have no cited agent doing them.

For a team lead, the practical answer is that some analyst work can shift from building spreadsheets to checking them. Whether that cuts the analyst hours your team needs is not something the sources establish.

Two pieces of public data frame the role. The U.S. Bureau of Labor Statistics (BLS) puts the median annual wage for financial and investment analysts at $102,740 in May 2025, with about 377,200 jobs in 2025. Across all financial analysts, including risk specialists, BLS projects employment to grow 7% from 2025 to 2035, "much faster than the average for all occupations", with about 29,500 openings a year. BLS expects analysts to be needed "to assess growing volumes of data." That is BLS's projection about the occupation, not a claim about any single team.

Task by task

The table below starts from the 26 task statements O*NET lists for financial and investment analysts (code 13-2051.00), merged into 12 tasks a manager would recognize. Narrow specialties, such as green construction finance, are folded into the general tasks. A task is marked for an agent only where a cited source documents AI doing it. Where O*NET lists the task but no cited source shows an agent doing it, the task is marked human, and the note says so.

Agent

0of 12

Agent with review

5of 12

• Monitor economic, industry and company developments from publications and data services
• Analyze financial and operational data and interpret trends in price, yield and risk
• Build and run financial models, including valuation of businesses and transactions
• Evaluate and compare securities and companies within an industry
• Draw charts and prepare written reports and presentations

Human

7of 12

• Prepare materials for transactions and deal execution
• Determine offering prices for securities
• Recommend investments and timing, and prepare investment plans
• Advise clients on capital structure, debt and financing, and keep client relationships
• Assess companies by visiting facilities and meeting management
• Purchase investments under company policy
• Supervise, train and mentor junior analysts

On that standard, 0 of 12 tasks are ones an agent completes with only spot checks, 5 of 12 are ones a documented AI tool does with a person reviewing the output, and 7 of 12 stay with a person. The five are monitoring developments, analyzing data, building models, comparing securities, and preparing charts, reports and presentations.

The evidence comes from three organizations. Each describes its tool as working with a person, not instead of one.

• Anthropic. According to Anthropic's documentation, Claude for Excel answers questions about open workbooks with cell-level citations, changes assumptions while keeping formula relationships intact, finds the root cause of errors, and builds new models or populates existing templates. It can pull external data through connectors such as S&P Global, LSEG and Daloopa. Anthropic's March 2026 product post lists starter skills for auditing models, building LBO, DCF and three-statement templates, and running comparable company analyses. The same documentation says the add-in is not recommended for final client deliverables without human review, for audit-critical calculations without verification, or for highly sensitive or regulated data without proper controls. It also warns that spreadsheets from outside sources can carry hidden instructions that manipulate the add-in.
• Microsoft. Microsoft's support page says Copilot in Excel can create charts and PivotTables, generate formulas and edit a workbook directly. It also says Copilot "can sometimes make mistakes" and advises users to "avoid using Copilot for decisions in sensitive areas such as finance."
• Morgan Stanley. In October 2024, Morgan Stanley announced the launch of AskResearchGPT. At launch, the firm described it as an assistant for its investment banking, sales and trading and research staff that can look for data and summarize the firm's research, which the firm said runs to more than 70,000 reports a year, with findings moved into an email draft that staff modify before sharing with clients. That announcement is just over two years old, and this research found no later public statement on whether the tool is still in use, so treat it as a dated example of how one bank framed the job at launch, not as evidence of current use. It stays on this page because it is the only source here from a firm describing such a tool for its own analysts rather than a vendor describing its product.

FINRA's 2026 oversight report adds a regulator's view of the same pattern: among FINRA member firms it surveyed, the top generative AI use case is summarization and information extraction. FINRA also lists the risks of AI agents, including acting without human validation and approval, acting beyond their intended authority, and lacking the domain knowledge for complex, industry-specific tasks.

No cited source reports an independent accuracy rate for these tools. That is why every documented task here keeps a review step.

A worked example

This is an illustration, not a reported case. Assume, for this example, that an analyst at a mid-size company is asked whether a $40 million acquisition target is fairly priced. The analyst opens the firm's DCF template and asks an AI add-in to populate it from the target's last three years of financials and to build a comparable company table from a data connector. Assume the add-in fills the template and the table and cites the cells it used. The analyst checks the inputs and finds that one peer in the comps table had a one-off asset sale that inflates its margin. The analyst removes it, reruns the valuation and writes the recommendation. The AI did the assembly. The person caught the comparison that did not belong and owned the conclusion.

Jobs you can hand to an agent today

The tasks above group into separate jobs, each with its own page covering what to give the agent, a brief you can copy, risks and how to run a trial.

• JobAssist only
  AI agent for financial research: what it does and needs

  An AI agent can search filings and research libraries, extract figures and terms, and draft cited primers and summaries. An analyst still checks every number that matters, builds the model and owns the conclusion.
• JobAssist only
  AI agent for FP&A and forecasting: what it can do

  An AI agent can draft budget-versus-actual commentary, recurring management reports and forecast refreshes for an analyst to finish. A person still sets the assumptions, approves the forecast and presents it.
• JobAssist only
  AI agent for spreadsheet analysis: what it can do

  An AI agent inside Excel can summarize a dataset, add formula columns and PivotTables, flag outliers and trace formula errors. Both vendors cited here tell you to verify the output, so a person checks the numbers before anyone relies on them.

Ready

0of 3

Supervised

0of 3

Assist only

3of 3

• AI agent for financial research: what it does and needs
• AI agent for FP&A and forecasting: what it can do
• AI agent for spreadsheet analysis: what it can do

Not yet

0of 3

• Financial research and analysis covers monitoring companies and markets, summarizing filings and research, and building comparisons.
• FP&A and forecasting covers budget models, forecasts and variance analysis.
• Spreadsheet analysis covers cleaning, analyzing and charting data in workbooks.

If you are deciding where to start, look at financial research. Three of the five documented tasks above (monitoring developments, analyzing data and comparing securities) belong to it. In a trial, one way to judge it is to count how many figures in the agent's output your analyst had to correct against the source.

What stays human

Four things stay with a person in the evidence cited here.

Recommendations and certified views. Under the SEC's Regulation AC, a broker or dealer or covered person that publishes, circulates or provides a research report prepared by a research analyst to a U.S. person in the United States must include a certification by the research analyst that the views in it "accurately reflect the research analyst's personal views", along with a statement about the analyst's compensation. Some third-party research is exempt under conditions the rule sets out. The rule applies to those firms and reports, not to every in-house analysis. Ask counsel which obligations apply to your role and jurisdiction.

Review of what the AI produces. Anthropic does not recommend its Excel add-in for final client deliverables or audit-critical calculations without review, and Microsoft advises against using Copilot for decisions in finance. FINRA says its rules, including Rule 3110 on supervision, continue to apply when member firms use generative AI. Someone who knows the business has to check the model before anyone relies on it.

Clients, management and deals. O*NET lists advising clients on capitalization and debt, collaborating with bankers and lawyers, examining company facilities and preparing deal materials. BLS lists meeting company officials. No cited source shows an agent doing any of these.

Spending money. O*NET lists purchasing investments under company policy. No cited source documents an agent doing it, and FINRA names autonomy without human approval as a risk of AI agents. This page's recommendation, not a legal requirement: a named person approves every trade or commitment.

A setup the cited documentation supports is AI doing the first pass on research, models, comps and charts, and an analyst checking it, making the call and owning the relationship. When the job board opens, you will be able to post that kind of job below.

This page is not financial, tax or accounting advice.

Sources

1. 1
   13-2051.00 - Financial and Investment Analysts

   O*NET OnLine (U.S. Department of Labor)PrimaryAccessed Nov 14, 2026

2. 2
   Financial Analysts: Occupational Outlook Handbook

   U.S. Bureau of Labor StatisticsPrimaryAccessed Nov 14, 2026

3. 3
   17 CFR 242.501 - Certifications in connection with research reports

   Electronic Code of Federal Regulations (eCFR)PrimaryAccessed Nov 14, 2026

4. 4
   GenAI: Continuing and Emerging Trends (2026 FINRA Annual Regulatory Oversight Report)

   FINRAPrimaryAccessed Nov 14, 2026

5. 5
   Use Claude for Excel

   AnthropicPrimaryAccessed Nov 14, 2026

6. 6
   Advancing Claude for Excel and PowerPoint

   AnthropicVendorAccessed Nov 14, 2026

7. 7
   Frequently asked questions about Copilot in Excel

   MicrosoftPrimaryAccessed Nov 14, 2026

8. 8
   Morgan Stanley Research Announces AskResearchGPT Powered by OpenAI

   Morgan StanleyVendorAccessed Nov 14, 2026