AI in Accounting: What’s Actually Working in U.S. Firms Right Now

Ask a managing partner about AI in accounting two years ago and you’d get a shrug and something noncommittal about “watching the space.” Ask today and you’ll usually get a much more specific answer, a tool name, a workflow, a number of hours saved during busy season, and often a complaint about something that didn’t pan out.

That shift matters more than any headline about robots replacing CPAs. The conversation in American firms has moved from whether to adopt AI to where it belongs, who reviews the output, and what happens to the people whose work it changes. Those are harder questions, and they don’t have vendor-friendly answers.

This guide covers what AI in accounting actually looks like inside U.S. practices today: the adoption data, the workflows that pay for themselves, the risks that get glossed over, and a 90-day path for firms still sitting on the fence.

The state of AI in accounting: past the tipping point, short of maturity

The numbers have moved fast, and it’s worth separating signal from noise.

Thomson Reuters Institute’s 2025 Generative AI in Professional Services Report found 68% of tax and accounting professionals feel excited or hopeful about generative AI’s future in the profession. At that point, 21% of tax firms said they were already using GenAI, with another 53% planning or considering it. The share with no plans at all had dropped to 25% down from 49% the year before. By the 2026 edition, roughly a third of tax firms reported using generative AI, with 14% already working with agentic tools.

Tax research has become the clearest proof point for AI in accounting. A June 2026 survey of more than 1,000 U.S. tax professionals from Blue J and CPA.com found 60% now use AI for tax research at least weekly, nearly double the 33% recorded a year earlier.

But here’s the detail that says the most about where firms really are: among tax firm respondents already using a GenAI tool, 52% were using general-purpose technology like ChatGPT, while only 17% used an industry-specific tool. In other words, a lot of AI in accounting started as individual staff quietly using consumer chatbots, not as a firm-level technology decision. That gap between personal experimentation and governed deployment is where most of today’s problems live.

Where AI in accounting is genuinely earning its keep

Strip away the marketing and a consistent set of use cases keeps surfacing across firms of every size.

1. Tax research

This is the beachhead for AI in accounting. Research tools trained on human-edited, tax-specific content can return sourced answers to a complex question in minutes instead of an afternoon in the code and regs. The important distinction is between a general chatbot guessing at a Section reference and a purpose-built tool that cites authority you can verify. Only one of those belongs in a client file.

2. Tax return preparation

Document ingestion is where using AI in accounting produces the least ambiguous return. OCR combined with AI verification now handles a large share of standard source documents, W-2s, 1099s, K-1s, with automated splitting and categorization of client uploads. Newer agentic tools go further, assembling substantially complete 1040 drafts before a preparer opens the file. The work doesn’t disappear; it converts from assembly to review. Firms running outsourced tax preparation and planning support alongside these tools are effectively stacking two kinds of capacity on the same bottleneck.

3. Bookkeeping and the month-end close

AI-powered software categorizes expenses, reconciles accounts, flags anomalies, and drafts financial reports. This is where AI in accounting unlocks capacity at scale for client accounting services, and where the economics of fixed-fee CAS engagements finally start to work. It’s also the layer where automation and outsourced accounting and bookkeeping overlap most, which is worth planning deliberately rather than discovering by accident.

4. Document summarization and anomaly detection

Contracts, invoices, receipts, lease agreements, board minutes. AI can surface key terms and flag items that look wrong, speeding up both audit fieldwork and compliance review. It doesn’t decide what’s material. It decides what’s worth a human’s attention first.

5. Advisory and planning

This is the one partners care about, and the point where AI in accounting stops being a cost story and starts being a revenue one. Predictive tools let a firm model the tax implications of client decisions year-round instead of reporting on them in April. Tax advisory has climbed into the top three GenAI use cases in the Thomson Reuters research, behind only research and return preparation. That ordering tells a story: firms start chasing efficiency and end up with a different service mix — which is why CFO and business advisory capability has become the growth question for mid-sized practices.

6. Audit

Audit is where AI in accounting has the highest ceiling and the slowest clock. The largest firms are furthest along, using AI for initial reviews of audit documentation, sample matching, evidence tracing, and risk identification. Deloitte has built generative and agentic capabilities into its Omnia audit platform. EY has rolled AI into the technology supporting its global assurance engagements. PwC expects an end-to-end AI-driven audit solution to be complete in 2026. KPMG’s Trusted AI framework focuses on helping clients deploy AI responsibly, a service line built out of governance expertise rather than automation.

What the Big 4 do that smaller firms can copy

It’s easy to look at Big 4 spending on AI in accounting and conclude the whole thing is out of reach. That’s the wrong takeaway. Two things the largest firms do are entirely portable to a 12-person practice in Ohio.

• They pick a workflow, not a technology: Deloitte didn’t adopt “AI.” It applied AI to audit documentation review. PwC’s in-house teams built tools for specific engineering tasks and measured productivity gains of 20% to 50% in software development. Scope beats ambition.
• They build a governance framework before they scale: KPMG’s Trusted AI framework exists because the firm concluded that trust, not capability, is the real constraint on adoption. A small firm can write a two-page AI use policy in an afternoon. Almost none have.

What smaller firms have instead is speed. No global rollout committee, no legacy platform migration, no 40,000 people to retrain. A 15-person firm can pilot a research tool on Monday and have a verdict by Friday. That’s a genuine advantage, and the firms using it are quietly pulling ahead of larger competitors still in evaluation. Thomson Reuters covers this size divide well in its breakdown of how different accounting firms use AI.

The evidence: what actually changes when firms adopt AI

Vendor case studies are not evidence. Fortunately, there’s now real research on AI in accounting.

Researchers at MIT Sloan and Stanford’s Graduate School of Business, Jung Ho Choi and Chloe Xie, analyzed hundreds of thousands of transactions across 79 small and medium-sized companies using AI-enabled accounting software, paired with survey responses from 277 accountants. Their findings, widely covered in 2025:

• 7.5 days cut from the monthly close cycle
• 12% increase in the granularity of financial reports
• 8.5% of accountant time shifted from routine back-office processing to higher-value work
• 55% more clients supported per week by AI-using accountants compared with non-users, along with more billable hours logged

One finding deserves more attention than it got. More experienced accountants got larger gains from the tools, they intervened selectively, particularly where the AI’s confidence was low. Judgment didn’t become less valuable. It became the thing that determined whether the technology paid off.

Separately, Thomson Reuters Institute’s Future of Professionals research put expected time savings at roughly five hours per professional per week, about 240 hours a year. Whether that becomes recovered capacity or just an earlier finish depends entirely on what a firm does with the time.

The risks nobody puts in the brochure

If you take one thing from this article, take this section.

Confidential client data and public tools don’t mix.

When more than half of AI-using tax professionals are working in consumer chatbots, some portion of client data is going where it shouldn’t. Anything entered into a public AI service should be treated as having left the firm’s control. This is a Circular 230 problem, a state board problem, an engagement letter problem, and a professional liability problem all at once. Any serious AI in accounting policy starts with a written approved-tool list and enterprise accounts carrying contractual data protections, before staff experimentation becomes a disclosure event.

AI output is a draft, not a work product:

Generative models produce confident, fluent, plausible text regardless of whether it’s correct. A fabricated citation to a revenue ruling that doesn’t exist looks exactly like a real one. Every AI-generated conclusion that touches a return, an opinion, or a set of financials needs to be traced to authority by a human who signs.

Responsibility doesn’t transfer:

No state board, no IRS examiner, and no plaintiff’s attorney will accept “the software produced it” as a defense. The CPA who signs owns the work.

Client transparency is becoming table stakes:

Clients increasingly ask how their data is handled and whether AI touched their engagement. Firms that address this proactively in engagement letters look competent. Firms that get asked and improvise don’t. The same due diligence applies to any external partner, which is why standards like ISO 27001 belong on your checklist when evaluating CPA outsourcing providers, not just software vendors.

Skills are the bottleneck, not tools:

Most AI in accounting projects stall for one reason: nobody in the firm knows how to prompt, evaluate, or supervise the output well enough to trust it.

The talent equation behind AI in accounting

Any honest discussion of AI in accounting and finance has to sit alongside the profession’s staffing problem. The U.S. pipeline has been contracting for years — accounting degree completions well off their peak, CPA exam candidates down sharply since 2017, and a retirement wave that pulled experienced people out faster than new ones came in.

This reframes the automation anxiety considerably. Most American firms are not looking for ways to need fewer people. They are looking for ways to serve existing clients with the people they already have, and to make the first three years of a young accountant’s career something other than data entry.

That’s the pitch the profession’s own institutions are making too. In April 2026, AICPA and CIMA launched the AI Accelerator Skills Program, a three-tier training program built on the premise that mindset, skills, leadership, and governance drive transformation more than tools do. It grew out of a 2025 convening of CFOs and finance leaders from more than 50 Fortune 1000 companies and was refined through pilots with U.S. public accounting firms. Participants can earn CPE credit.

The AICPA & CIMA AI resource hub is worth bookmarking regardless: it collects guidance, glossaries, risk toolkits, and research aimed at building competence and credibility rather than just enthusiasm.

AI or outsourcing? That’s the wrong question

Firms tend to treat automation and external support as competing answers to the same capacity problem. They aren’t. They solve different halves of it.

AI in accounting is strongest where work is high-volume, rule-shaped, and repetitive, extraction, categorization, reconciliation, first-pass research. It is weakest where work requires accountability, contextual judgment, or cleanup of something messy. That second category is where trained human capacity still earns its cost, whether in-house or through a white-labeled offshore team working under your brand.

The practical model most growing firms land on has three layers: automation handles extraction and routine processing, a support team handles preparation and production volume, and licensed U.S. staff handle review, sign-off, and the client conversation. Firms that skip the middle layer usually discover their savings get eaten by the exception cases nobody has time to clear.

The decision isn’t tools versus people. It’s which layer each piece of work belongs in — and whether you’ve priced the engagement to reflect that.

A realistic 90-day AI in accounting plan for U.S. firms

• Days 1–15: Find out what’s already happening. Ask staff, without penalty, what AI tools they’re already using and for what. You will be surprised. This inventory is the real starting point.
• Days 16–30: Write the policy. Approved tools, prohibited data types, review requirements, client disclosure language. Two pages. Have it reviewed by counsel and your insurance carrier.
• Days 31–60: Pick one workflow and pilot it. Choose something high-volume, low-judgment, and easy to measure, document intake, expense categorization, or first-pass tax research. Set a baseline before you start, or you’ll never know whether it worked.
• Days 61–90: Measure, then decide. Hours per return. Days to close. Review notes per engagement. Realization rate. If the numbers didn’t move, kill it and try the next one. If they did, document the process and train the team properly before expanding.

Then the question that actually determines your return on AI in accounting: what are you doing with the recovered time? Capacity that isn’t deliberately redeployed into advisory work, client relationships, or reduced busy-season burnout simply evaporates.

What AI won’t do

Professional skepticism. Materiality judgments. The conversation where you tell a client their expansion plan doesn’t pencil out. Ethical decisions with no clean answer. Reading a board’s discomfort in a room. Signing your name to something and standing behind it.

AI for accounting is very good at the first 80% of a task and unreliable at the last 20%, which happens to be the part clients pay for. The firms getting the most out of it aren’t the ones with the biggest tech budgets. They’re the ones that were clear-eyed about which 80% they were handing over.

People Also Ask:

Conclusion

AI in accounting stopped being a prediction somewhere around 2025. It’s now a set of ordinary operational decisions: which tools, which workflows, whose data, who reviews, what policy, what training.

The firms that will look good in three years aren’t the ones that moved fastest. They’re the ones that moved deliberately, picking real problems, measuring honestly, protecting client data, investing in their people’s skills, and keeping professional judgment exactly where it has always belonged. That, more than any tool, is what successful AI in accounting looks like.

Curiosity is the right posture. Verification is the right practice. You need both.

Working through the capacity side of this? Corient works exclusively with U.S. CPA firms as a white-labeled extension of your team, covering bookkeeping, tax preparation, compliance audit support, and financial reporting under ISO 27001 and GDPR-aligned controls. Explore our CPA outsourcing services or talk to our team about where your bottleneck actually sits.