How can a small accounting firm use AI?

The most practical uses are chasing missing client documents, sorting and renaming uploads, drafting replies to routine client email, suggesting transaction coding, flagging reconciliation exceptions and summarizing IRS notices. Tax positions, advice, signing and filing stay with licensed professionals.

Is it safe to put client tax data into AI tools?

Only with the right setup. Use business-grade tools that do not train on your data, include them in the written information security plan the FTC Safeguards Rule requires, and check whether Section 7216 consent is needed for how the tool handles tax return information. Never use free consumer chatbots for client data.

Can AI do bookkeeping in QuickBooks or Xero?

It can suggest coding with more context than the built-in rules and, more usefully, flag the transactions and reconciliations that look wrong so a bookkeeper reviews a short list instead of everything. A person still reviews and closes the books.

Can AI read W-2s, 1099s and K-1s?

Yes, current models can identify and extract data from tax documents, including scans and photos. Accuracy drops on poor scans and long brokerage statements, so every extracted number should be tied out by a preparer before it is used.

Will AI replace bookkeepers?

Not at small firms in any way I would recommend. It removes data entry and document sorting, which most firms are short-staffed on anyway. The review, client relationships and judgement are still the job.

AI for small accounting and bookkeeping firms: the busywork, not the judgement

Ask anyone at a small firm where tax season goes and the answer is rarely "preparing returns." It goes to chasing documents, answering the same client emails, sorting uploads, coding transactions and reminding people for the fourth time that their K-1 is still missing.

Jake BoylesIndependent AI consultant, Oxford, OhioNovember 6, 20266 min read

The short answer

Small accounting and bookkeeping firms can use AI to chase missing client documents, sort and rename uploads, draft replies to routine client email, suggest transaction coding and flag reconciliation exceptions, and summarize IRS and state notices. It should not take tax positions, give clients advice or touch client data without a written security plan and client consent where the law requires it.

Start with data protection, because the rules are real

Most posts about AI start with the fun part. For a firm that handles tax and financial data, the order matters. Before any client information goes into an AI tool, three things should be settled.

Your written information security plan. The FTC Safeguards Rule requires tax preparers to have one, and IRS Publication 4557 lays out what it should cover. Any AI tool you add belongs in it.
Consent for tax return information. Section 7216 of the Internal Revenue Code restricts how preparers use and disclose tax return information. Sending it to a third-party service can require client consent, depending on how the tool works. Ask your professional liability carrier or counsel how it applies to the tools you are considering.
Business-grade tools only. Use AI on a business plan where your data is not used to train the model, with access controls you manage. A staff member pasting a client's return into a free chatbot is the scenario to prevent.

None of that rules AI out. It means the boring setup comes first. I cover the rollout side in Microsoft Copilot adoption, since many firms already pay for Microsoft 365.

Chasing documents without being the one who chases

The single biggest time sink in most small firms is getting clients to send what you need. You send an organizer or a list in January. Half the clients send half the documents. Then someone spends weeks comparing what came in to what was expected and sending reminders.

AI is good at the comparison. It can read last year's return or file to work out what this client normally sends (two W-2s, a 1099-B, a 1098, a K-1 from the family partnership), check what has arrived in the portal, and send a short, specific reminder: "We have everything except the K-1 from Smith Holdings and your 1098 for the rental." That is far more effective than "please upload your remaining documents."

If you use a practice management system like TaxDome, Canopy, Karbon or SmartVault, this sits on top of the portal you already have.

Sorting, renaming and filing uploads

Clients upload files named "scan0042.pdf" and photos of documents taken on a kitchen table. Someone has to open every one, figure out what it is, rename it and file it in the right folder for the right year.

Current AI models can read a document and tell you that it is a 2025 Form 1099-INT from a specific bank for a specific client, then rename and file it to your convention. They can also pull the key numbers into a worksheet for the preparer to review. It is the same kind of work I describe on document processing.

Client email: the same twenty questions

"Did you get my documents?" "When will my return be done?" "What is my estimated payment for June?" "Can you send me last year's return?" Each takes a minute to answer and a few minutes to look up.

An assistant connected to your practice management system can draft replies to these with the actual answer filled in: the status of the engagement, the payment amount and due date from your workpapers, a secure link to the document. Staff read and send. Anything that is really a question about tax treatment, like "can I deduct my home office," gets flagged for a person and does not get an AI answer.

Done well, this does not make the firm feel less personal. It makes replies faster and frees your staff to have the conversations clients actually value.

Bookkeeping: coding, reconciliation and month-end

QuickBooks Online and Xero already suggest categories from bank feeds, with mixed results. AI can do better when it has more context: the vendor, the memo, the client's history, and your notes about how this client wants things coded. It can also explain why it chose a category, which makes review faster.

The bigger time saver is exceptions. Rather than reviewing every transaction, AI can surface the ones that look wrong: a vendor coded differently than usual, a duplicate payment, an unusually large expense, an uncategorized deposit, a reconciliation that does not tie. Your bookkeeper spends the time on twenty items instead of four hundred.

At month-end it can draft the close checklist status and a short note to the client listing what you need from them. I go into QuickBooks specifically on QuickBooks automation, and on desktop users moving to online on QuickBooks Desktop migration.

IRS and state notices

Clients forward notices in a panic. Someone has to read it, work out which notice it is, what tax year it covers, what the IRS thinks is wrong, the deadline, and what the client needs to do.

AI can read the notice and produce that summary in a consistent format, plus a plain-English explanation you can send the client after review: what it is, whether it is serious, and what happens next. It can also pull up the relevant return so the preparer can check the issue quickly.

The response to the notice is the preparer's work. AI saves the first twenty minutes, not the professional judgement.

Where AI stays out of the firm

Tax positions and advice. Whether something is deductible, how to structure an entity, what to do about an audit. That is your license and your liability.
Signing or filing. The preparer reviews, signs and files. Always.
Final numbers. Anything AI extracts or codes is reviewed before it touches a return or a financial statement.
Client data in unapproved tools. If it is not in your security plan, it does not get client data.

A sensible first project for most firms is document chasing and sorting, because it is high volume, easy to review and does not involve any professional judgement. If you want a broader way to decide, how small businesses can start using AI lays out how I think about it.

Questions people ask me

How can a small accounting firm use AI?: The most practical uses are chasing missing client documents, sorting and renaming uploads, drafting replies to routine client email, suggesting transaction coding, flagging reconciliation exceptions and summarizing IRS notices. Tax positions, advice, signing and filing stay with licensed professionals.
Is it safe to put client tax data into AI tools?: Only with the right setup. Use business-grade tools that do not train on your data, include them in the written information security plan the FTC Safeguards Rule requires, and check whether Section 7216 consent is needed for how the tool handles tax return information. Never use free consumer chatbots for client data.
Can AI do bookkeeping in QuickBooks or Xero?: It can suggest coding with more context than the built-in rules and, more usefully, flag the transactions and reconciliations that look wrong so a bookkeeper reviews a short list instead of everything. A person still reviews and closes the books.
Can AI read W-2s, 1099s and K-1s?: Yes, current models can identify and extract data from tax documents, including scans and photos. Accuracy drops on poor scans and long brokerage statements, so every extracted number should be tied out by a preparer before it is used.
Will AI replace bookkeepers?: Not at small firms in any way I would recommend. It removes data entry and document sorting, which most firms are short-staffed on anyway. The review, client relationships and judgement are still the job.

Written by Jake Boyles

I am an independent AI consultant in Oxford, Ohio, with fifteen years of software engineering behind me. I help small businesses work out where AI actually fits, and I resell nothing, so I can tell you when a tool is not worth buying. More about me.