Industry Guide: Accounting
How to Use Claude in Accounting: A Guide for Bookkeeping, Close, Tax, and Audit Teams
Faster drafts, the same review line.
This guide is for bookkeepers, staff and senior accountants, tax preparers, and audit teams at accounting firms and finance departments who want to put Anthropic's Claude to work on the close, the return, and the audit file. It focuses on what named companies have said about using Claude in their own accounting work, in Anthropic's own customer stories and announcements, since Anthropic has not published a dedicated accounting product the way it has for some other industries.
Knowing how to use Claude in accounting means pointing it at well-defined, document-heavy tasks: reconciling accounts, drafting month-end close narratives, preparing plain-language explanations of a tax calculation, assembling audit document requests, and answering client questions from your own records. Claude drafts and organizes against the numbers you give it. Accountants and auditors still review every number, judgment, and filing before it goes out the door.
Published
Free lesson · no account neededThe Iteration LoopMaster the Essentials6 minStart the free lessonKey Takeaways
- The strongest accounting use cases are bank reconciliation, month-end close narratives, tax-calculation explanations, audit document requests, and client intake, where Claude drafts against your own numbers.
- There is no dedicated Anthropic accounting product. Every example here comes from named companies describing how they use Claude in their own accounting, tax, or audit workflows.
- Campfire's accounting platform reports cutting close time and bank reconciliation time after building Claude into its product, and says its whole engineering team uses Claude to build it.
- Armanino's auditors use Claude on Amazon Bedrock to draft client document requests and explain rejected submissions, cutting manual writing time and follow-up questions.
- KPMG is embedding Claude into tax and legal tools for its Digital Gateway clients, and Intuit uses it inside TurboTax to explain tax calculations in plain language.
Copy and Paste
Prompts You Can Use Today
Bank reconciliation discrepancy triage
You are helping reconcile this month's bank statement against the general ledger. Compare the transactions pasted below line by line, list every discrepancy with the date, amount, and account involved, flag any transaction that appears on only one side, and suggest the most likely explanation for each gap without assuming it is correct. Statement and ledger: [paste].
Flagging unmatched items instead of guessing is what keeps this draft safe to rely on.
Month-end close variance narrative
Using the trial balance and prior-period comparison pasted below, draft a variance narrative for month-end close: identify every account that moved more than the threshold I give you, state the direction and rough size of the change for each, propose a plausible business reason based only on the notes I provide, and flag any account where no note explains the movement. Data: [paste].
The flagged, unexplained accounts are exactly what a reviewer should check first.
Audit document request list
You are supporting an audit engagement. From the attached trial balance and prior-year workpapers, draft a client document request list organized by financial statement area, write one clear sentence explaining what each requested item is for, note which items are typically the most error-prone based on the notes I provide, and leave a blank field for the engagement manager to adjust scope. Workpapers: [paste].
The engagement manager still sets final scope; this draft just saves the first pass.
Plain-language explanation of a tax calculation
A client is asking why their refund changed from the estimate. Using only the tax calculation worksheet pasted below, explain in plain language which lines changed, why each change affected the refund, and what documentation would be needed to verify the result, without giving any new tax advice beyond what the worksheet already shows. Worksheet: [paste].
Restricting the explanation to the worksheet keeps it from drifting into new advice.
Client intake email for missing tax documents
Draft a friendly but specific email to a client who has not yet sent their tax documents. List exactly which documents from our intake checklist below are still missing, explain briefly why each one is needed, give a clear deadline based on the filing date I provide, and keep the tone warm rather than demanding. Checklist and filing date: [paste].
A specific, checklist-based ask gets a faster response than a generic reminder.
In Practice
How Accounting Teams Use Claude
According to Anthropic's published customer story
Campfire
Campfire, an accounting platform built for high-growth tech companies covering invoicing, billing, revenue accounting, and financial statements, told Anthropic that building Claude into the product cut three days off its monthly close, cut bank reconciliation time by 90%, and cut reporting time roughly in half. Campfire also said its entire engineering team, not just a data team, uses Claude to build the product, including its conversational reporting feature, Ember AI.
According to Anthropic's published customer story
Armanino
Armanino, one of the largest accounting firms in the United States, told Anthropic that its auditors use Claude on Amazon Bedrock to draft client document requests and to explain why a submission was rejected. The firm reported a 65% reduction in time spent on manual writing tasks, a 60% decrease in follow-up clarifications auditors needed to send, and said the feature went from idea to production in two weeks, with 90% of users reporting more confidence using it on engagements where auditors typically reject 50 to 75 items and spend 10 to 15 minutes writing up each one.
According to Anthropic's published customer story
KPMG
KPMG, a global professional services network with more than 276,000 employees across 138 countries and territories, told Anthropic it formed an alliance to embed Claude into KPMG's Digital Gateway, the software KPMG's clients use directly, beginning with tax and legal tools. KPMG described building an agent to help clients adjust to changing tax regulations as work that used to take weeks and now takes minutes.
According to Anthropic's published customer story
Intuit
Intuit told Anthropic it integrated Claude, through Amazon Bedrock, into Intuit Assist inside TurboTax, where it generates natural-language explanations of how a federal tax calculation produced its result while a filer reviews their return. The underlying tax engine content the assistant draws on runs to more than 100,000 pages, covering more than 15,000 federal and state tax forms.
According to Anthropic's published customer story
PwC
PwC told Anthropic that its Office of the CFO business group uses Claude internally for journal entries, variance analysis, and request-for-proposal drafts, and that its finance teams use Claude Code for annual planning work. PwC did not break out figures specific to these accounting tasks, so no number is attached to this use here.
Illustrative walkthrough
A month-end close variance narrative
- A staff accountant loads the trial balance, the prior period's numbers, and the close checklist into a project.
- Claude drafts a first-pass variance narrative, flagging any account where it cannot find a note explaining the movement.
- The accountant checks every flagged account against source documents and fills in the missing explanations.
- A senior reviewer reads the narrative against the general ledger before it goes into the close package.
- The finalized narrative moves into the month-end file alongside the supporting workpapers.
Illustrative walkthrough
A client tax-document intake workflow
- A preparer loads the firm's document checklist and the client's prior-year return into a project.
- Claude compares the checklist against what the client has sent and drafts a note listing exactly what is missing.
- The preparer edits the note for tone and confirms the deadline before it goes out to the client.
- Once new documents arrive, the preparer asks Claude to re-check them against the same checklist.
- The preparer signs off on a complete file before any return work begins.
Core Workflows
Claude in Accounting: Core Use Cases From Bookkeeping to Audit
These are the areas where ai for accounting firms saves the most reviewer time while keeping every number traceable to the source record.
Bookkeeping and reconciliation
AI for bookkeeping flags mismatched transactions, drafts reconciliation notes, and summarizes a ledger for review, leaving every adjusting entry to a human.
Month-end close
AI for month-end close turns a trial balance and prior-period numbers into a first-draft variance narrative, so the close team edits instead of starting from a blank page.
Tax preparation
AI for tax preparation drafts plain-language explanations of a calculation and organizes client documents against an intake checklist, leaving every filing position to a preparer.
Audit support
AI for audit drafts client document requests and explains why a submission was rejected, the pattern Armanino reports using with Claude on Amazon Bedrock.
Financial reporting and analysis
Journal entry drafts, variance analysis, and RFP responses for the finance function, the kind of work PwC describes using Claude for inside its own Office of the CFO group.
Client and team communication
Intake emails, status updates, and internal explainer notes drafted from your own records, so staff spend review time on judgment instead of formatting.
What Accounting Companies Report
How Accounting Firms and Finance Teams Use AI Today
How is AI used in accounting firms today? Campfire, an accounting platform built for high-growth tech companies that covers invoicing, billing, revenue accounting, and financial statements, told Anthropic that building Claude into the product cut three days off its monthly close, cut bank reconciliation time by 90%, and cut reporting time roughly in half. Campfire also said its whole engineering team, not just a data team, uses Claude to build the product, including its conversational reporting feature, Ember AI. This is what ai in accounting looks like once a pilot moves into the actual product.
Armanino, one of the largest accounting firms in the United States, told Anthropic that its auditors use Claude on Amazon Bedrock to draft client document requests and to explain why a submission was rejected. The firm reported a 65% reduction in time spent on manual writing tasks, a 60% decrease in the follow-up clarifications auditors needed to send, and said the feature went from idea to production in two weeks. Ninety percent of users said they felt more confident using the AI-assisted feature, on engagements where auditors typically reject 50 to 75 items and spend 10 to 15 minutes writing up each rejected item.
Across generative ai in accounting, the pattern in these two examples is consistent: document-heavy, repetitive writing work goes first, and the judgment behind a journal entry, a tax position, or an audit opinion stays with a qualified professional.
Tax, Audit, and Large Firms
What KPMG, Intuit, and PwC Say About AI in Accounting
KPMG, a global professional services network, has formed an alliance with Anthropic to put Claude inside KPMG's Digital Gateway, the software KPMG's own clients use directly, starting with tax and legal tools. KPMG described building an agent to help clients adjust to changing tax regulations as work that used to take weeks and now takes minutes, across a firm with more than 276,000 employees in 138 countries and territories.
Intuit told Anthropic it integrated Claude, through Amazon Bedrock, into Intuit Assist inside TurboTax, where it generates natural-language explanations of how a federal tax calculation produced its result while a filer reviews their return. The underlying tax engine content the assistant draws on runs to more than 100,000 pages, covering more than 15,000 federal and state tax forms.
PwC told Anthropic that its Office of the CFO business group uses Claude internally for journal entries, variance analysis, and request-for-proposal drafts, and that its finance teams use Claude Code for annual planning work. None of these three accounts describe a named Anthropic accounting product: they describe Claude embedded into each company's own software, or used directly by its own finance, tax, and audit staff.
Bookkeeping and Close in Practice
AI for Bookkeeping and Month-End Close Done Well
The pattern that holds up for ai for bookkeeping is a narrow scope and a hard reference back to the ledger. Give Claude the transactions, the chart of accounts, and your reconciliation template, and ask it to flag anything it cannot match instead of guessing at an adjusting entry. A draft built that way is fast for a bookkeeper to check line by line, because every flagged gap points straight at the source document.
AI for month end close works the same way. A trial balance and the prior period's numbers are enough for a first-draft variance narrative: which accounts moved, in which direction, and what a plausible reason might be. The accountant still confirms the reason against source documents before anything is final, and still owns every adjusting entry that goes into the ledger.
How accounting firms use ai for repetitive writing is the clearest lesson from Campfire and Armanino: keep Claude's draft tied to a specific source, whether that is a bank statement, a ledger, or a rejected audit item, and the review step becomes a check rather than a rewrite. Loosen that tie and the time saved on drafting gets eaten up by a reviewer tracing an unsupported claim back to nothing.
Choosing Tools
Claude vs ChatGPT for Accounting
The claude vs chatgpt for accounting comparison is best run on your own work: a reconciliation you already finished, a close narrative you already approved, a client document request list you already sent. Compare accuracy against your source numbers, how each tool flags gaps instead of filling them in, and how each handles a messy, inconsistent ledger with missing entries.
In accounting the decision usually comes down to governance rather than a single output: which deployment your IT and risk teams approve, how client and financial data are handled, whether the tool connects to your general ledger or practice management system, and how your firm documents AI-assisted work for its own quality control review.
Is ai accurate enough for tax preparation? Accuracy depends entirely on the quality and completeness of what you give it. A tool working from a complete, correct worksheet can explain a calculation faithfully, the pattern Intuit describes inside TurboTax. A tool working from incomplete source documents can state a confident-sounding wrong answer. Preparers should treat every AI-drafted explanation as a draft to verify, never as a filing position on its own.
Getting Started
A Safe First Month for an Accounting Team
Start with finished work
Test prompts on a reconciliation, a close narrative, or a document request list you already finished, and compare the AI draft with what you actually sent.
Load your own templates
Put your chart of accounts, close checklist, and document request template into a project so every draft follows your firm's own format.
Pick one recurring task
Choose one high-volume task, such as bank reconciliation or client intake emails, and measure review time before and after.
Agree on the review line
Decide with your quality control partner which drafts are exploratory and which steps every AI-assisted filing or audit workpaper still needs.
Limits
Where Accountants and Auditors Must Still Decide
Claude can misread a number, miss a transaction, or state a plausible-sounding reason that is not the real one. Every balance, every filing position, and every audit conclusion needs to trace back to a source document that a person checked, not to a draft taken on faith.
Can ai replace accountants? Nothing in these examples supports that. Campfire, Armanino, KPMG, Intuit, and PwC all describe Claude drafting and organizing work that a professional still reviews, approves, and signs. The judgment behind a filing position, an audit opinion, or a client recommendation stays with a licensed professional, not with the model that helped draft it.
Client financial data, tax records, and unpublished audit findings are sensitive. Use only an approved, firm-sanctioned deployment, keep client data out of consumer-grade tools, and keep your workpapers and filings in your firm's own validated systems of record.
Build Reliable AI Habits for Accounting Work
Getting trustworthy drafts from AI is a practiced skill: grounding every claim in your own numbers, referencing sources, and reviewing before anything goes out. The free Nightschool AI curriculum is hands-on from the first lesson.
Frequently Asked Questions
How is AI used in accounting firms?
Mostly for document-heavy drafting and reconciliation work: comparing a ledger against a bank statement, writing a first-draft close narrative, preparing audit document requests, and explaining a tax calculation in plain language. Campfire, Armanino, KPMG, Intuit, and PwC each describe this pattern in what they have told Anthropic about using Claude. The accounting judgment, the filing position, and the audit opinion still belong to a qualified professional.
Can AI help with bank reconciliation?
Yes, for comparing a bank statement against the general ledger, flagging transactions that only appear on one side, and drafting a plausible explanation for each gap. Campfire reports a large cut in bank reconciliation time after building Claude into its accounting platform. A bookkeeper still confirms every flagged item against the source document before closing the books.
Can AI draft audit document requests?
Yes. Armanino's auditors use Claude on Amazon Bedrock to draft client document requests and to explain why a submitted item was rejected, cutting manual writing time and the number of follow-up clarifications auditors need to send. The engagement manager still decides what is actually requested and still signs off on the final list.
Is AI accurate enough for tax preparation?
It depends entirely on what you give it. Working from a complete, correct worksheet, Claude can explain a calculation faithfully, the pattern Intuit describes inside TurboTax's Intuit Assist. Working from incomplete documents, it can produce a confident-sounding wrong explanation. Every AI-drafted explanation needs a preparer's review before it reaches a client or a filing.
How much time can AI save in accounting?
It varies by task and by how complete your source data is, so measure it on your own work. Campfire reports cutting days off its monthly close and a large share of its bank reconciliation and reporting time, and Armanino reports cutting manual writing time on audit document requests. Review time does not disappear, and it should not.
What are the biggest AI use cases in accounting?
Based on what accounting firms and software companies have told Anthropic, the biggest ai use cases in accounting are bank reconciliation, month-end close narratives, audit document requests, tax-calculation explanations, and journal-entry or variance-analysis drafting for finance teams. ai in accounting works best when every draft stays tied to a specific source record.
Can AI replace accountants?
No. Every example in this guide describes Claude drafting and organizing work that a professional still reviews, edits, and approves. The judgment behind a filing position, an audit opinion, or a client recommendation stays with a licensed accountant or auditor, not with the model.
Is there a dedicated Claude product for accounting firms?
Not as of this guide. Anthropic has not published a named accounting offering, so everything here comes from named companies describing how they use Claude in their own accounting, tax, and audit work, not from a page about one specific accounting product.
Keep going
Related Guides and Courses
Claude in Accounting, by role
Draft Faster, Keep the Review Intact
Hands-on lessons for accounting and finance teams.