AI for accountants: how firms actually use it

AI for accountants is most useful when it is pointed at a specific job, not at "accounting" in general. In a tax and accounting firm, that means eight places: client intake, data extraction, return prep, review, workpapers, tax planning, client communication and firm-wide analytics. General chat assistants handle drafting and research. Work on source documents needs purpose-built tools that cite their evidence, and a CPA still signs off on every return.

Plenty of partners have already tried something. Someone pasted a K-1 into a chatbot, got a confident answer, found a wrong number and closed the tab. That shows exactly where general tools stop. This guide goes stage by stage through the firm: what the AI does, what a person still does, and what to check before you trust it. For a vendor-by-vendor list, see our roundup of AI tools for tax professionals.

Where does AI for accountants actually help?

AI earns its place at the stages where staff spend hours moving information from one place to another and then checking it. Here they are in the order work moves through a firm.

1. Client intake and document chasing

AI compares what a client has sent against what they sent last year and tells you what is missing before anyone opens the tax software. If the prior-year return shows a Schedule E and two brokerage accounts, and this year's upload has one 1099-B, the gap gets flagged in February instead of the week of the deadline. The better tools treat this as non-blocking, so prep starts on what is there while someone chases the missing K-1. A person still decides whether the gap matters (a closed account won't produce a 1099) and writes the note to the client.

2. Data extraction

Extraction is the most mature use of AI in accounting. Models read W-2s, 1099s, K-1s, brokerage statements and prior-year returns, including scans and phone photos, and pull out each value with a pointer back to where it came from. The hard cases are consolidated brokerage statements that run dozens of pages, K-1 footnotes and state supplemental schedules. Test any tool on your ugliest client files, not the vendor's sample W-2.

3. Return prep

AI prep takes the extracted data and produces a first-draft return inside the firm's tax software. The preparer's job shifts from keying to checking: confirming the items the system was unsure about, making elections and handling anything that needs judgment. Good prep tools say which fields they are confident about and which they are not, instead of filling every box with the same apparent certainty. Firms should also be able to set rules that stop a return and wait for a person, for example when a client with wage history has no W-2 on file.

4. Review

Review is where AI saves senior time, which is the scarcest time a firm has in March and April. Tax software diagnostics check math and missing fields. They don't check whether line 2b matches the 1099-INT the client actually sent. An AI reviewer compares the draft return against the source documents line by line and produces a list of discrepancies with a citation for each, so the reviewer verifies instead of re-performing the work. Firms using Filed Reviewer report 64% shorter review cycles and about 42 minutes saved per return. The reviewer still signs off.

5. Workpapers

AI can build workpapers from the same data: trial balance, book-to-tax reconciliation, depreciation, basis schedules and an open-items list for anything it couldn't support. The useful version fills in the firm's existing Excel templates, so the output looks like what your reviewers already expect. Whether a reconciling item is permanent or timing, and whether a shareholder has enough basis to take a loss, stays with the preparer and reviewer.

6. Tax planning and advisory

Planning is where the completed return becomes a starting point instead of the finish line. AI can test a return against a long list of strategies, estimate the effect of each and draft a client-ready summary with the code section or IRS guidance behind every number. The CPA decides which strategies fit the client's goals and appetite for risk, and has the conversation.

7. Client communication

This is where general chat assistants do well. A reminder about an unsigned engagement letter, a plain-English explanation of a balance due, a summary of an IRS notice, a partner's bullet points turned into a client email: all fine for a general assistant, as long as no identifiable client data goes into a consumer account. Keep a person on the send button. An AI draft that misstates a deadline is still the firm's mistake.

8. Firm-wide analytics

Once every client's documents and returns sit in one system, you can ask questions across the whole book. Which clients have consolidated brokerage statements? Who realized a large capital gain and made no estimated payments? Which entity clients should be looked at for a pass-through entity tax election? That turns a week of spreadsheet work into a query.

When is a general chat assistant enough, and when do you need a purpose-built tool?

A general assistant such as ChatGPT, Claude or Gemini is enough when the output is text a professional will read and edit, and no client data is involved. It falls short on anything that has to be right to the line, tied to a source document or entered into tax software.

General assistants work well for:

• First drafts of client emails and internal memos
• Summarizing long guidance or a notice, before you read the source yourself
• Excel formulas, quick scripts and reformatting data you have already anonymized

You need a purpose-built tool for:

• Reading a client's documents and extracting values with a citation for each
• Preparing a return inside CCH Axcess, UltraTax, Lacerte, Drake or ProConnect
• Comparing a draft return against source documents before sign-off
• Anything that needs an audit trail from document to form

Accuracy is the dividing line. Filed clears 94% line-by-line accuracy on TaxCalcBench (here is how we measured it against ChatGPT, Claude and Gemini). Treat a number like that as the bar a tool must clear before it is worth testing. The errors that remain tend to be the ones that need judgment, which is why review stays with a person.

How should an accounting firm handle client data security with AI?

Treat an AI vendor like any other service provider that touches taxpayer data, because legally it is one. Three sources set the floor.

• The FTC Safeguards Rule. The FTC's guidance on the Safeguards Rule names tax preparation firms as covered financial institutions. The rule requires a written information security program, a designated Qualified Individual, multi-factor authentication for anyone accessing customer information, oversight of service providers, and notice to the FTC within 30 days of a breach involving 500 or more consumers' unencrypted information.
• IRS Publication 4557. Safeguarding Taxpayer Data is the IRS's practical guide for preparers, with checklists you can hold any new tool up against.
• Section 7216. The Internal Revenue Code limits how preparers use and disclose tax return information. Confirm how a vendor's data handling fits within it, and whether client consent is needed for anything beyond preparing the return.

Before you sign, ask the vendor five questions. Is client data used to train models? Where is it stored, for how long, and how is it deleted? Can you see a SOC 2 report? Can access be restricted by role, with an audit log? Does every output trace back to a source document? The consumer tier of a chat assistant fails most of these by default. Business and enterprise plans do better, but read the terms yourself.

How should an accounting firm get started with AI?

Start with one bottleneck, test on returns you have already filed, and change workflow outside busy season. A sequence that works for most firms:

1. Pick the bottleneck. For firms doing volume it is usually review, because senior reviewers are the constraint in March and April.
2. Test on last season's returns. You already know the right answers. Include the messy ones: multi-state clients, K-1 heavy returns, the client who sends photos of documents.
3. Measure two things. Time per return, and what the tool caught that your team missed (and the reverse).
4. Write the rules down. Every return still gets human sign-off. No identifiable client data goes into general assistants. One person owns the rollout.
5. Train in the off-season. Summer or after the October extension deadline. Nobody wants a new workflow in the middle of March.
6. Expand one stage at a time. Review is often first because it needs no change to how prep is done. Extraction and prep come next, then planning.

Filed covers most of the list above for mid-sized and high-volume firms. Filed Prep reads the documents and prepares the return. Filed Reviewer checks the draft against the source documents. Filed Tax Planner builds cited plans from the completed return. Prowork sits next to CCH Axcess, UltraTax, Lacerte, Drake or ProConnect and reviews inline as the preparer works; you correct it once in plain language and that correction becomes a permanent skill. Filed handles 1040, 1065 with K-1, 1120, 1120-S, 1041 and 990 returns.

Frequently asked questions

Will AI replace accountants?

No. AI takes on the keying, matching and first-draft work. Elections, professional judgment, client relationships and the signature stay with the CPA. Firms adopting AI mostly use it to handle more returns with the same team and to move hours toward review and advisory work.

Is it safe to put client tax data into ChatGPT or Claude?

Not into a consumer account. Tax firms are covered by the FTC Safeguards Rule and Section 7216, so any tool holding client data gets vetted like any other service provider. Use a business plan with model training turned off, a purpose-built tool with a SOC 2 report, or strip identifiers before you paste anything.

What is the best AI for accounting firms?

It depends on the stage. General assistants are good for drafting and research. For extraction, prep and review, look for a tool that reads source documents, cites every value, works inside your existing tax software and publishes its security posture.

Does Filed work with our tax software and return types?

Filed works with CCH Axcess, UltraTax, Lacerte, Drake and ProConnect, and covers 1040, 1065 with K-1, 1120, 1120-S, 1041 and 990 returns. Filed is SOC 2 compliant, with ISO 27001 certification in progress. Pricing is one flat monthly platform fee for the whole firm, scoped to taxpayer count.

If you want to see what AI for accountants looks like on your own returns, book a demo and bring a messy one.

‍