In an accounting firm AI arrives in three places before anyone decides anything: the tax department that summarises the client's documents and drafts the research memo, the audit team that writes the workpaper narrative and the management letter, and the advisory desk that answers the client's question about the numbers. The managing partner's job is not to pick the tool. It is to decide three things before a tool touches a client's return, a workpaper or an engagement letter: who in the firm owns the question, what may never go into an outside tool, and what the firm will never let a machine do alone. This article is for the managing partner or owner of a CPA or accounting firm with a few dozen to a few hundred people, independent or newly backed by private equity.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs, and accounting firms are the group where a tool's mistake carries a licensed signature, a client's confidence and a peer reviewer's attention at the same time.
Where AI already is in your firm
Ask a senior in tax what the free tools on the staff's phones are doing this week and the honest answer is: reading the client's brokerage statements into a summary, drafting the research memo, writing the answer to the client's question about an estimated payment. Ask the audit team and it is the first draft of the workpaper narrative, the summary of the board minutes, the management letter comment. Ask advisory and it is the explanation of the cash-flow forecast and the draft of the client's banker's letter. Ask the office and it is the engagement letter from last year's with the dates changed, the job posting and the collections note. And the practice-management vendor and the tax-software vendor each switched on a drafting feature with their last release. None of it was approved. Some of it has already told a client something about their taxes with nobody's name on it, and some of it has carried a client's return, general ledger or identification onto a vendor's server under no agreement at all.
The three decisions, in the order a firm makes them
Who owns the question. Not the IT provider, not the tax-software vendor and not a new hire. In most firms it is the quality-control partner, the firm administrator or the tax department head, because they are the people the staff already ask when something new shows up. Give that person a written scope of half a page: what they may approve alone, what comes to the managing partner, when the page is reviewed. The article on AI for a small business with no IT department draws the line between their column and yours.
The data rule. One sentence. For an accounting firm it usually reads: no client's return, financial statements, general ledger, bank or brokerage records, identification or payroll, no workpaper, and nothing about an employee go into a tool the firm has not approved, and no tool is approved to hold client information without the written vendor terms the firm's confidentiality duties require. The managing partner writes it, because the client's information is protected by the profession's rules and by the tax code's own confidentiality provisions, and the vendor's terms are the firm's to sign.
The never-build list. The three things the firm will never let a machine do alone. For an accounting firm the list writes itself: no return, financial statement, opinion or attestation is signed or filed without the licensed person reading it; no advice on a tax position, a transaction or a client's numbers leaves the firm from a tool without a partner's or manager's name on it; no engagement letter, fee quote or scope change is sent by a tool. Everything else is a question of fit. These three are not.
What fits at fifty people, and what does not yet
The first project is the work your people already do by hand, the same way, every busy season: the client's documents read into a summary for the preparer to check against the source, the first draft of the research memo for the reviewer to finish, the workpaper narrative drafted from the senior's notes, the engagement letter drafted from the template for the partner to read, the collections note, the job posting. The article on picking your first AI project has the five tests. What does not fit yet, in most firms this size, is anything that decides or signs: the position, the opinion, the filed return, the advice to the client, the fee. A tool can prepare any of those for a licensed person. It should not finish them.
The peer reviewer, the liability insurer, the regulator and the buyer will ask
The AI question reaches an accounting firm from four directions that have nothing to do with technology. The peer reviewer asks which systems touch workpapers and who signed what a tool drafted, and the answer decides the firm's standing. The professional-liability insurer's renewal form asks whether automated tools give advice or produce documents in the firm's name, because advice a tool gave is advice the firm gave. The state board and the profession's rules ask it about confidentiality and competence in their own words, and the tax code adds its own penalties for client information disclosed without consent. And a buyer, a larger firm or private equity, asks it in diligence, because accounting firms are being bought in platforms now; the article on whether AI makes a business worth more when you sell has that version, and the one on what to tell your bank about AI covers the bank behind the partner buyouts. One dated page answers all of them.
Picture a regional CPA firm of one hundred twenty people with three offices and a new private-equity partner. The tax department at the largest office has been running client documents through a free tool on the seniors' own logins for a busy season, and the returns have gone out faster for it. The managing partner finds out when the peer reviewer asks which system produced a research memo and who checked it before the position was taken, and then finds that a client was told a deduction was allowed when it was not, that the answer went out under a manager's name without her reading it, and that the client's return, brokerage statements and identification went into the same tool. The fix takes two weeks and one call to the firm's liability carrier. The quality-control partner is named as the person who owns the question, with a half-page scope. The managing partner writes the data rule: no client file, no workpaper, no employee file into an outside tool, and no tool without written vendor terms. The never-build list goes on the page, starting with no advice to a client without a licensed name on it. The free tool is replaced by the firm's plan the same month, the client gets a corrected answer in writing from a partner, the files are deleted from the vendor's account with the deletion confirmed in writing, and the peer reviewer gets the page. This is a hypothesis, a composite and not a client.
The one page
What is in use, by department and task, including the free tools and the features the practice-management and tax-software vendors switched on. Who owns the question, with the written scope. The data rule in one sentence, with the vendor terms listed. The never-build list, the three items above. The review date and who signs. The managing partner signs it, the quality-control partner keeps it with the quality-management documents, and the firm administrator has it ready for the peer reviewer and the insurer's renewal. The board-briefing template has the same page in a form a firm can fill in. If you want to see which of the three is missing before anyone writes the page, the AI readiness scorecard asks the questions a managing partner can answer from memory.
Mistakes that repeat
Letting the software vendor switch it on. A drafting feature arrives in a release, defaults to on, and the next memo carries a position nobody read. The managing partner decides, feature by feature.
The client's return in a free tool. The one mistake that puts the firm in front of the state board and the client's lawyer. The data rule exists for this.
Advice nobody licensed read. A tool drafts the answer to the client's question; a manager sends it at the end of a long day. The client relies on it. A licensed person reads every piece of advice that leaves.
An engagement letter the template wrote. A tool rolls last year's letter forward with the dates changed; the scope has changed and nobody noticed. A partner reads every engagement letter before it goes.
No review date. Standards change, insurers change their forms, people change offices. A page without a date is a page nobody opens.
Frequently asked questions
Should an accounting firm use AI?
It already does, through the tax department, the audit team and the advisory desk. The decision worth making is who owns the question, what may never go into an outside tool, and the three things the firm never lets a machine do alone: a return or opinion signed unread, advice to a client without a licensed name, and an engagement letter or fee sent by a tool.
Who should own AI in a CPA firm?
The person the staff already ask when a new tool shows up, usually the quality-control partner, the firm administrator or the tax department head, with a written scope of half a page and a review date. Not the IT provider and not the tax-software vendor.
Can AI draft our research memos and workpapers?
It can draft from documents a person has given it, for the reviewer to check against the source and the standard and for the licensed person to sign. It should not take the position, because the signature is the licensed person's and the peer reviewer will ask who checked it.
What about client information in these tools?
It is first in the data rule. No client's return, financial statements, ledger, bank or brokerage records, identification or payroll goes into a tool the firm has not approved, and no tool is approved to hold client information without the written vendor terms the firm's confidentiality duties require. If client files have already gone into a free tool, delete them from the vendor's account, get the deletion confirmed in writing, and tell the firm's liability carrier the same week.
Why does my peer reviewer or my insurer ask about AI?
Because each carries the risk when a memo, a workpaper or an answer to a client that a tool touched turns out to be wrong, and each wants to know whether a licensed person stands behind it. One dated page with names on it is the answer, and the same page answers the state board, a buyer and the bank.
Not sure where you stand? Take the 3-minute AI readiness scorecard.