Which finance jobs can you hand to an AI agent today?

AI agents for finance can do real work today, but only the repetitive middle of most jobs. According to product documentation, they check expenses against your written policy and ask for and use missing context on bank transactions. We recommend that a person approves anything that moves money, and law requires named people to sign certain filings. If you run a finance team or do the books yourself, the practical question is not whether to use an agent but which job to hand over first, how you will check its output, and who keeps the final say.

This page is not financial, tax or accounting advice. It describes what agents do according to the cited sources, as of November 2026.

What AI agents do in finance today

Two products show what this looks like, according to their official documentation:

• Intuit's help center describes an Accounting AI in QuickBooks Online that, for bank transactions needing more context, asks follow-up questions, handles the back-and-forth between you and your accountant, and updates the transactions. The same page describes a Payroll AI that runs payroll and involves payroll managers "for confirmations or approvals," and a Sales Tax AI that reviews a return for errors before you file.
• Ramp's help center describes Policy Agent, which applies a company's written expense policy to every card transaction and reimbursement and returns one of three recommendations: approve, review or reject. When a workflow is set up that way, it can approve clearly in-policy expenses automatically. Ramp states that "reviewers always retain final authority" and that the agent marks an expense for review when it cannot decide. Ramp reports that early customers cut manual reviews by about 85% while keeping "99%+ accuracy." That is Ramp's own claim about its own product, not a market figure.

In both, the agent does the volume work and a person handles exceptions and approvals. Ramp also documents an activity feed for each expense that shows the agent's recommendation, its rationale with the cited policy text, and any reviewer overrides.

Among these surveyed large organizations, adoption is broad, and so is the checking. In Deloitte's own CFO Signals survey of 200 CFOs at North American companies with at least $1 billion in revenue (second quarter of 2026), 93% said their organizations use AI extensively or modestly across multiple functions, 44% use it for financial planning and budgeting, and 41% use it to analyze financial data. Datarails, a finance software vendor, surveyed 270 CFOs and finance leaders at U.S. organizations with over 1,000 employees and revenue of $100 million or more in July 2026. In that survey, finance teams spend 26% of their working week verifying or correcting AI output. Only 5% of respondents trust AI to produce board-ready reports without human review, and only 4% trust it with the month-end close. Neither survey tells you how small companies use AI.

Read those numbers together. Agents take on volume, and you budget time to check their work. A finance agent that saves hours on data entry but has no audit trail just moves the work into review.

Jobs you can hire an agent for

Each job below has its own page with a task breakdown, a brief template you can copy, risks and how to run a trial.

• JobSupervised
  AI agent for accounts payable: what it does and costs

  An AI agent can read, code and match supplier invoices and draft them in your accounting system. A person still approves new vendors, exceptions and every payment.
• JobSupervised
  AI agent for accounts receivable and collections

  An AI agent can send reminders, log customer replies and propose payment matches. A person still decides disputes, payment plans, credits, write-offs and escalations.
• JobSupervised
  AI agent for month-end close: what it does and needs

  An AI agent can move close tasks along, prepare reconciliations, flag mismatches and draft variance commentary. A person still approves adjusting entries, closes the period and signs off the numbers.
• JobSupervised
  AI agent for bookkeeping and bank reconciliation

  Accounting software AI can suggest categories for bank transactions and match or reconcile the ones it is confident about. A person still reviews its work, decides unusual treatments and signs off each reconciliation.
• JobSupervised
  AI agent for tax preparation: what it can do today

  An AI agent can organize client tax documents, extract the data and populate a draft return for a preparer to review. Tax positions, the preparer's signature and filing stay with people.
• 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.
• JobSupervised
  AI agent for expense audit: what it checks, who approves

  An AI agent can check every expense report and card transaction against your written policy, flag duplicates and suspicious receipts, and approve clearly in-policy spend under limits you set. People still decide rejections, repayments and suspected fraud.
• 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.

Ready

0of 8

Supervised

6of 8

• AI agent for accounts payable: what it does and costs
• AI agent for accounts receivable and collections
• AI agent for month-end close: what it does and needs
• AI agent for bookkeeping and bank reconciliation
• AI agent for tax preparation: what it can do today
• AI agent for expense audit: what it checks, who approves

Assist only

2of 8

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

Not yet

0of 8

They are not equally ready. Here is a rough way to sort them:

Job	Where agents are today	What a person keeps
Accounts payable	Capture, coding and matching of invoices	Approving payments and new vendors
Expense review and audit	Checking every expense against written policy	Exceptions, rejections and policy changes
Bookkeeping and reconciliation	Categorizing and matching bank transactions	Posting unclear items and closing the period
Accounts receivable	Drafting reminders and matching payments	Customer disputes, credits and write-offs
Month-end close	Preparing reconciliations and variance notes	Review and sign-off of the statements
FP&A and forecasting	Drafting forecasts and variance commentary	Assumptions and what goes to the board
Financial research	Gathering and summarizing filings and data	Conclusions and investment decisions
Tax preparation	Organizing documents and flagging errors	Preparing, signing and filing the return

Two limits come from law, not from the technology. For U.S. public companies, SEC Rule 13a-14 requires each principal executive and principal financial officer, or persons performing similar functions, to sign a certification included with each covered quarterly and annual report, and the certification may not be signed on their behalf under a power of attorney (paragraph (c) of the rule). On the tax side, the IRS requires anyone who prepares or helps prepare federal tax returns for pay to hold a valid preparer tax identification number (PTIN). Plan the human roles in any finance agent setup around those obligations.

Roles they affect

Agents change the work of three finance roles most visibly. Each page goes task by task through what an agent can take and what stays with the person.

• Role
  Can AI replace bookkeepers? A task-by-task answer

  QuickBooks and Xero document AI features that suggest transaction categories, auto-reconcile confident bank matches, draft invoice reminders and run payroll with approval. People review that work, and most other bookkeeping tasks have no cited agent evidence.
• Role
  Can AI replace financial analysts? A task-by-task answer

  Anthropic and Microsoft document AI that builds and edits financial models, comps tables, charts and decks, and Morgan Stanley announced AskResearchGPT in October 2024 as an assistant for searching and summarizing its research. A person reviews that work, and recommendations, client advice and certified research views stay with a person.
• Role
  Can AI replace accounts payable clerks? Task by task

  AI agents documented by Microsoft and Ramp draft supplier bills, suggest coding, propose purchase-order matches and flag duplicates, with a person reviewing the output. Supplier questions, purchase orders, reporting and tax forms have no cited agent evidence and stay with a person.

The surveys above do not show agents replacing these roles. In the Datarails survey, 60% of CFOs said they redeploy staff to other work as AI takes on routine tasks, and 3% said they are cutting finance headcount because of AI. Treat that as one vendor-commissioned survey of large U.S. companies, not a rule.

Where to start

Start with the job that has the most volume, the clearest rules and the cheapest mistakes. For most small and mid-size companies that is accounts payable or expense review. Both have a written rulebook (your chart of accounts, your expense policy), a natural approval step you already run, and outputs you can check by sampling.

For example: a supplier emails a $1,240 invoice. The agent reads it, matches the vendor and purchase order, suggests a general ledger account and queues it for approval with a note on how it decided. Your approver checks the coding and clicks approve, or corrects it. That correction is the feedback you track. The payment itself still goes through your normal approval.

Before you hand over any finance job:

1. Write the rules down. An agent applies the policy you give it. Ramp's documentation, for example, notes that its agent cannot read rules inside images or hyperlinks and works best with clear, explicit limits.
2. Decide what it may do alone. Reading, coding and drafting are low risk. Paying, posting to a closed period, changing vendor bank details and filing are not. Keep a person on those.
3. Measure it on a sample. Run the agent beside your current process for a few weeks and compare its output with what your team would have done. Track how often reviewers override it.
4. Keep the audit trail. Every action should show what the agent saw, what it decided and who approved it. Lack of auditability was the top reason CFOs in the Datarails survey gave for not fully trusting AI (75%).
5. Know who your data goes to. Finance agents touch bank, vendor and employee data. Check what the provider stores, where, and who at your company can see it.

If you would rather describe the job and have someone help you scope it, use the form below. The guide to hiring an AI agent covers scoping, trials and handover in more detail.

Sources

1. 1
   Finance Teams Now Spend a Quarter of Their Week Fact-Checking AI, New Datarails Survey Finds

   DatarailsVendorAccessed Nov 14, 2026

2. 2
   North American CFOs express concerns about AI governance and risk management (CFO Signals, Q2 2026)

   Deloitte InsightsVendorAccessed Nov 14, 2026

3. 3
   Overview of Intuit AI in QuickBooks Online

   IntuitPrimaryAccessed Nov 14, 2026

4. 4
   Policy Agent Overview

   RampPrimaryAccessed Nov 14, 2026

5. 5
   17 CFR 240.13a-14: Certification of disclosure in annual and quarterly reports

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

6. 6
   PTIN requirements for tax return preparers

   Internal Revenue ServicePrimaryAccessed Nov 14, 2026