The clearest picture yet of how AI actually gets used inside professional-services firms arrived on November 1, 2026, when the Journal of Accountancy published interviews with leaders at four CPA firms ranging from 11 employees to more than 1,800. The firms described AI running in everyday audit work, not pilots: assessing risk, preparing data, selecting samples, reviewing working papers and drafting reports, in ways that save time and money.
What makes the reporting useful is its specificity. This is not a survey or a vendor case study. It is four named firms, five concrete use cases, and an honest account of what changes and what has to stay human. If your business sells professional judgment, their experience is a preview of yours.
The four firms
The Journal spoke with Donald Wells, CPA, president and owner of Wells CPA, an 11-person Atlanta-area firm that primarily serves not-for-profits and performs more than 80 audits a year; J. Sean Spitzer, CPA, president and COO of Smith + Howard, a 750-employee Atlanta firm; Ronald Martinez, CPA, audit partner at PKF O'Connor Davies, an 1,800-employee firm; and Jennifer Wood, CPA, assurance service line leader at The Bonadio Group, who also helped develop the AICPA's Dynamic Audit Solution framework for using AI in audits.
Their five use cases form a map of where AI is earning its keep in audit work today.
1. Risk assessment and planning
Wells CPA uses AI to review client financial statements, board minutes, policy manuals and other documents to identify possible risks and plan the audit. "AI can go through and do all of those initial document review processes in a matter of two or three hours," Donald Wells said. "If we had a midlevel staff person doing it, we might take a week." The firm runs a closed AI system, so client documents stay inside its own IT environment rather than going to an outside service.
At Smith + Howard, generative AI has cut research time during planning significantly. When researching GAAP rules, Spitzer said, "you can just ask some of these models, 'Where do I find the information about this specific problem.' Instead of spending two hours flipping through guidance, you're analyzing data."
2. Data extraction, formatting and verification
The least glamorous use case may be the most revealing. At PKF O'Connor Davies, a recurring annoyance drove a build: general ledgers exported from client accounting software often arrive in formats unsuitable for audit work and require lengthy manual preparation. The firm built a custom agent that does the preparation. "Automating that helps our associates focus more of their time and attention on planning analytics and risk assessment," Martinez said.
At The Bonadio Group, AI handles the footing, crossfooting and tie-out of client financial statements, the necessary but rote arithmetic verification that could take staff two to three days manually. The staff accountant's role on that task, in Wood's phrase, shifts "from a doer to a reviewer."
3. Sample selection and agentic automation
Some of the most tangible wins come from analyzing large datasets and surfacing exceptions humans would not naturally find. One Bonadio engagement team running a data query with AI identified employees who had authorization to post journal entries but had done so fewer than three times during the year. The finding prompted the client to tighten controls and cut software-license costs for barely used seats.
Both Bonadio and PKF O'Connor Davies described building or deploying AI agents that go a step further than analysis, such as selecting audit samples from client transaction populations. Wood pointed to Microsoft Copilot's agentic tools as one example, "where you can kind of program and have the tools do things like selecting your sample for you," but she added that some of Bonadio's biggest wins came from independently built, in-house agentic tools rather than off-the-shelf platforms. At PKF O'Connor Davies, custom GPTs and agents are vetted by a Digital Transformation Committee before rolling out firmwide, and internal AI champions help colleagues build comfort with new tools.
4. Working paper review
Wells CPA's most instructive practice is a policy, not a product: AI acts as the "first reviewer" of the audit team's own working papers, checking documentation for completeness and clarity before any human sees it. The AI review is mandatory, and no client-facing document reaches a partner or manager until the auditor has responded to every comment the AI made. Wells said the process shifted his own review time away from catching errors and toward judgment calls specific to the client.
5. Report drafting and disclosure checking
The firm also uses AI to check draft audit reports for internal consistency, flagging inconsistent naming or places where a report doesn't "tie to itself," and a vendor-supplied checklist tool that reviews financial statements against disclosure requirements and reports what it found and what it did not, leaving the judgment to the auditor.
The caution the firms volunteer themselves
Every firm in the piece stressed that human judgment and professional skepticism remain central, and the companion piece in the same issue, "AI raises the stakes for auditor judgment," argues that AI can improve audit quality while producing incorrect results, which makes the human review harder, not easier.
Wells flagged the risk most precisely. Junior staff reviewing the AI's risk assessments "tend to say, 'Hey, AI said this is a risk, therefore it's a risk.' I'm like, 'No, we have to think it through.'" The temptation to defer to the machine is highest exactly where the experience to challenge it is lowest, which is why the firms that report the best results are the ones that redesigned the workflow around review, rather than simply subscribing to a tool.
Spitzer's summary is the one worth keeping: "I don't think AI is replacing accountants. It's going to make us more helpful."
What any professional-services firm should take from this
The pattern across all four firms, from 11 people to 1,800, is consistent enough to plan around:
- Start with rote, high-volume work. Document review, data formatting, arithmetic tie-outs. These have clear outputs and low judgment risk, and the time savings are immediate and measurable.
- Keep client data in closed systems where the sensitivity demands it. Wells's approach, a closed AI system with documents staying in the firm's IT environment, is the standard clients will increasingly ask about.
- Prefer build over buy for recurring annoyances. The ledger-formatting agent at PKF O'Connor Davies is the model: a small custom tool aimed at one recurring, well-defined bottleneck.
- Make AI the first reviewer, not the final one. The mandatory-first-review policy at Wells is the most transferable idea in the piece: it puts the machine where it is strong, and reserves scarce human attention for the judgment calls.
- Manage the deference risk explicitly. Pair juniors with the requirement to justify, not accept, AI conclusions.
Accounting is further along this road than most professions, but the road is the same for all of them. If your firm wants help mapping which of its workflows belong in this pattern, that is exactly the kind of work we do. Start at The Vivid Company.
Read next
- Google AI Mode ads just went global: Canada is on the list, and here's the checklist
- DeepSeek's Super Agent removes the human bottleneck from coding agents
- Google says AI search queries now run 3x longer. Here's what advertisers should actually do
- Samaya Sterling beat Claude on a finance benchmark, at about one-eleventh the cost
