Finance & FP&A

How to prepare audit and tax workpapers with AI

Finance & FP&A3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Audit and tax workpapers are the documented evidence trail that supports every number on a financial statement or tax return — trial balances, depreciation schedules, accrual support, reconciliation tie-outs, and the narrative memos that explain why each position is defensible. For most small and mid-sized companies, assembling this package is a recurring, time-consuming exercise that happens at year-end, at audit time, or whenever a lender or investor asks to see the books. The work is methodical, cross-referential, and unforgiving of gaps.

AI feels like a natural fit here because so much of the effort is cognitive grunt work: organizing data into consistent formats, drafting explanatory memos, cross-checking figures across schedules, and flagging discrepancies for review. The underlying logic is rule-based enough that a language model can apply it reliably, and the written portions — narrative support, audit inquiry responses, footnote language — are exactly what LLMs are good at generating from structured inputs.

ChatGPT, Claude, and Gemini can meaningfully help with this workflow today. They can draft reconciliation memos, suggest workpaper structure by account type, convert a messy trial balance export into a clean formatted schedule, explain the accounting treatment for unusual items, and write responses to auditor inquiries. The limitation isn't the AI's competence at the task — it's the setup cost of getting your actual data in front of it each time you need it.

Skip the manual work → See the Starch version

Finance & FP&A3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ClaudeChatGPTGemini

Step-by-step

1 Export your trial balance from QuickBooks, NetSuite, or your accounting system as a CSV. Open Claude or ChatGPT and paste the raw data directly into the chat with a prompt asking it to organize the accounts into a standard workpaper index — grouped by financial statement section with a column for the expected supporting schedule type.

2 For each major account balance, paste in the relevant transaction detail and ask the LLM to draft a reconciliation memo: beginning balance, additions, disposals or reductions, ending balance, and a one-paragraph narrative explaining what drives the balance and whether it's consistent with prior periods.

3 Paste in any auditor PBC (Provided By Client) request list and ask Claude or ChatGPT to map each request to the corresponding workpaper you've already prepared, flagging any items where you don't yet have documentation. This produces a gap analysis you can work off of.

4 For depreciation schedules or amortization tables, paste in your fixed asset register and ask the LLM to calculate the current-year depreciation by method (straight-line, MACRS, etc.) and produce a formatted schedule with cost basis, accumulated depreciation, and net book value columns.

5 Draft audit response memos by describing the transaction or account in question and asking ChatGPT or Claude to write a formal narrative explanation suitable for an auditor's workpaper file — including the accounting standard cited, the treatment applied, and why it's appropriate.

6 For tax workpapers, paste in book income figures and ask the LLM to generate a book-to-tax reconciliation template populated with common M-1 adjustment categories, then fill in the amounts you provide and flag any items that need CPA review.

7 Review every output manually before it goes into the actual workpaper file. LLMs don't have access to your historical workpapers, so consistency with prior-year format and prior-year numbers is your responsibility to verify.

Prompts you can copy

Here is my trial balance as of 12/31/2024 [paste CSV]. Organize this into a workpaper index grouped by balance sheet and income statement sections. Add a column for the recommended supporting schedule type for each account.

Draft a reconciliation memo for our accounts receivable balance. Beginning balance: $412,000. New invoices: $1,840,000. Collections: $1,760,000. Write-offs: $22,000. Ending balance: $470,000. Explain whether this balance is reasonable given our revenue trend.

Here is our fixed asset register [paste data]. Calculate straight-line depreciation for each asset using the useful lives listed. Format the output as an audit-ready depreciation schedule with cost, accumulated depreciation, current-year depreciation, and net book value columns.

Write a formal memo explaining our revenue recognition policy for SaaS subscription contracts under ASC 606. We recognize revenue ratably over the contract term. The memo should be suitable for inclusion in our audit workpaper file.

Here is our auditor's PBC request list [paste list]. Map each item to one of these workpapers I've already prepared [paste workpaper index]. Flag any PBC items that don't have a corresponding workpaper yet.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your accounting system — every session starts with a manual export and paste, which means you're always working from a snapshot that's already slightly stale.

Large trial balances or full transaction detail sets frequently exceed what you can paste in a single context window, forcing you to chunk the data and lose the cross-account consistency the LLM would otherwise catch.

Workpaper formatting drifts between sessions — the structure you carefully prompted in October isn't what you get in March, so you spend time reformatting rather than reviewing the actual numbers.

Nothing persists between runs. Next quarter you're rebuilding the same prompt chain from scratch, with no memory of how you handled the same account last time or what the auditors flagged.

LLMs can't pull prior-year workpapers automatically to check for consistency or flag accounts where the balance moved significantly year-over-year — you have to paste that context in yourself.

Audit-ready documentation requires a reliable audit trail. When a memo is generated in a chat session, there's no built-in version control, approval workflow, or record of what inputs produced what output.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — for this workflow, that means an agent builds and runs the persistent app that keeps your workpaper data organized, reconciled, and connected to your live financial records year-round, not just when you're mid-audit.

Starch syncs your QuickBooks data on a schedule — invoices, bills, journal entries, vendors, and payments — so when you ask for a reconciliation or a trial balance pull, it's reading your actual current numbers, not a CSV you exported this morning.

The Transaction Insights starter app connects directly to your Plaid bank accounts and surfaces month-over-month spend by category, flags anomalies, and tracks every vendor charge — which is exactly the transaction-level support auditors ask for, already organized.

Describe the workpaper structure you want in plain English — 'build me an accounts receivable reconciliation that ties beginning balance, new invoices, collections, and write-offs month by month from QuickBooks data' — and an agent builds it as a persistent, auto-refreshing app.

Contract Lifecycle Management — coming soon — will give you a searchable repository with full audit trail for every contract, clause, and amendment, so supporting documentation for revenue recognition or lease obligations is a search away rather than a folder hunt.

Build an automation that runs before each quarter close: pull transaction detail from Plaid and QuickBooks, flag accounts with balances that moved more than 20% versus prior period, and Slack you a summary so you're not discovering surprises when the auditors arrive.

Because the apps Starch builds are persistent and connected to live data, your workpaper framework carries forward each period — prior-year balances are already there, formatting is consistent, and you're reviewing changes rather than rebuilding the whole package from scratch.

Get closed-beta access →

Toolkit