王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0359

文章 · 存档

深度解读美国CPA协会发布的2025年AI会计报告

CPA.com是美国注册会计师协会全资拥有的网站。

CPA.com在2025年6月发布了《2025 AI in Accounting Report》,基于AICPA和CPA.com的AI会计与金融研讨会、生态系统研究、供应商和从业者访谈,分析AI在会计领域的现状、主题、趋势与未来。

报告指出,AI已从实验应用转为基础设施,从自动化到增强会计师潜力。

AI会计目前的发展状况

当下AI技术已从2023年的ChatGPT热潮进化到更成熟的阶段。

大型语言模型(LLM)如GPT-4o、Claude 3.5等,不仅处理文本,还融入多模态能力(如图像分析、实时决策);

代理AI(Agentic AI)开始自主执行多步任务;

监管框架如欧盟AI法案(EU AI Act)和美国NIST标准日益完善。

同时会计发展趋势强调数字化转型:

云会计软件如QuickBooks、Sage Intacct的普及,数据隐私合规(GDPR、SOC 2)的强化,以及从合规导向向咨询服务的转变。

报告的核心观点是:

AI不是取代会计师,而是放大他们的潜力,帮助他们从重复任务中解放,聚焦高价值洞察。

报告开篇的执行摘要直击要害:

会计行业正经历深刻转型,AI已从技术进步转为战略必需。

3年CPA.com的AI倡议旨在“揭开”AI的神秘面纱,而两年后,报告指出AI已渗透到工作流自动化、智能审计助手、生成AI研究和代理任务执行等领域。

CPA.com总裁兼CEO Erik Asgeirsson强调:“AI是催化剂,提供前所未有的机会,让事务所和财务团队提升作为可信顾问的角色。”

当前状态分析显示,行业正处于拐点:

传统合规责任与客户期望、市场动态的平衡日益艰难。

人才短缺、复杂性上升使旧方法不可持续。

AI重塑运营,不仅自动化任务,还重新思考工作流、决策结构和服务产品。

关键指标包括:税准备工具自动化率超80%、LLM研究工具减少文档分析时间50%以上、代理AI支持客户互动和对账。

be radical联合创始人Pascal Finette指出:“战略要务不再是是否采用AI,而是多快、多周全地转向增强模型。”

结合当下AI发展这与现实高度吻合。

5年AI工具如Intuit的TurboTax AI或Thomson Reuters的Checkpoint Edge,已实现高自动化税申报;OpenAI的GPT系列和Google的Gemini在会计研究中广泛应用,减少手动检索时间。

会计趋势上,中国会计准则与国际接轨,强调数字化审计(如区块链辅助),而全球人才短缺问题凸显AI的“增员”作用。

关键战略七大主题

1、AI驱动的工作流自动化:事务所投资工具加速银行对账、交易编码、月末结账和报告,供应商报告时间节省30-70%。这释放产能转向咨询和分析,与当下Sage Copilot等工具一致,后者如“数字助理”路由文档、总结细节。

2、人机验证(Human-in-the-loop):自动化虽先进,但高风险领域如审计、税需人类审查。系统基于置信阈值标记交易,会计师从“执行者”转为“审查者”,降低风险、提升吞吐量。这反映了AI伦理趋势:避免“黑箱”决策,确保透明。

3、监管合规与风险管理:全球AI立法加速,如欧盟AI法案、美国行政命令和NIST标准。AI工具集成风险评分、文档追溯和审计轨迹。中国类似趋势见于《生成式人工智能服务管理暂行办法》,强调合规。

4、数据隐私与安全:客户信任依赖数据管理。供应商强调SOC 2、GDPR和选择性训练协议,“隐私优先设计”成为差异化点。当下,AI数据泄露事件频发(如2024年多家LLM模型漏洞),会计行业需警惕。

5、生成AI用于研究与决策支持:LLM工具如Blue J,总结复杂立法、提供税建议,减少研究时间。当下,Anthropic的Claude在专业领域细化模型,减少幻觉(hallucination)。

6、代理AI自主任务执行:代理系统执行多步工作流,如准备回报、对账、生成方差分析。人类监督仍关键,从推荐到执行的跃进正在进行。2025年,代理AI如Auto-GPT在会计中兴起,处理链式任务。

7、提升客户体验:AI提供实时仪表盘、按需支持和预测警报。产品嵌入客户端门户,实现现金流预测、异常检测。这与当下趋势契合:客户期望“即时洞察”,如Xero的AI分析工具。

这些主题显示,AI从孤立效率转向全面重构。

报告引用Sage CTO Aaron Harris:“你将做更多、更好、不同的事。”

当下会计趋势中,中国事务所如普华永道已采用AI提升客户门户,预示全球一体化。

六个趋势与五个发展

趋势分波次:

1. AI原生事务所:自动化优先,角色聚焦策略,减开销。

2. 领域特定AI模型:会计专用LLM,准确高、幻觉少、合规词汇。FinBERT优化金融。

3. AI上下文窗口扩展:处理超100万token,分析财务记录或税法。Gemini 1.5 Pro达200万token。

4. AI审计转型:DAS实时风险评估、模式检测。会计师转为风险导航者。EY工具实现实时检测。

5. 税合规向咨询转移:自动化准备,会计师聚焦规划、建模。

6. AI与财务系统集成:API连接QuickBooks,减延迟。API有限,期待标准。

五个发展:

- 簿记自动化:代理完成全流程,如分类、报告生成。

- 税务准备:从小时到分钟自动化,仅最终审查。

- 审计分析:缓慢采用,聚焦判断。

- 咨询服务:AI生成情景、预算,深化对话。

- 工作流效率:邮件分类、调度,成为基础设施。

EY全球AI保障领导Richard Jackson强调:“随着技术发展,我们设置训练轮、控制和保障,确保人类与技术同步。”

Caseware策略副总裁Danielle Supkis Cheek:“会计专业适合设计监督过程,批判评估输出。”

当下AI智能体在税软件如TaxJar中应用,审计工具如KPMG的Clara加速风险分析。

战略影响与人才转型

AI加速中高级需求,入门级减少。重构管道、提升团队、重塑轨迹。招聘聚焦批判思维、适应性和AI伙伴。

影响:

技能:AI素养必备,培训模型优缺点、提示、输出导航。

职业路径:新兴角色如AI运营经理、提示工程师。需跨职能技能。

竞争:成熟差距扩大,早采用获效率。延迟失份额、人才、信任。

客户:更快答案、丰富洞察、主动推荐。顾问角色强化。

人才转型:入门上移,要求AI素养;AI-ready程序结合培训;学术认证验证知识和技术。

培训:学徒复兴,如AICPA程序;逆向指导;模拟开发判断。

发展:早期AI监督;轮岗技术部门;评估转向价值分析、沟通、AI协作。

理解AI与未来轨迹

AI包括ML和知识系统。ML分类、检测、预测;知识处理规则,常优于LLM成本低。

混合:知识先、ML边缘、LLM建议。Jeff Seibert警告辨别供应商技术。

未来:系统学习互动、适应偏好、预见需求。

转变:任务到协调;静态到实时;决策到协作。

到2028,Gartner预测15%决策代理自主。Aaron Harris比喻逐步解锁信任。

当下:GPT-4V处理图像;LangChain建链。会计向持续审计,AI实时监控。

准备路线图与ROI洞察

四阶段:

发现(评估成熟、工作流审计、伦理);

试点(选用例如税研究、冠军、KPI);

操作化(嵌入工作流、责任、培训);

扩展(推广、产能分配、治理)。

成功关键:数据卫生、变革领导、可靠平台。

ROI早期主观:Gartner显示每周节省5.4小时,但69%用于重工。

实际获益:手动任务减70%、审查周期快5倍、容量增2-3倍。

BrooklynFi联合创始人Shane Mason称:“AI投资微小,节省30分钟即回本。”未来2026-2027,端到端重构将降低成本、缩短计费周期。

AICPA总裁Mark Koziel强调:“时间节省创造容量,服务更多客户。”

Hancock Dana合伙人Brian Gough问:“AI拿走三分之一工作后,能服务多少新客户?”

报告最后的反思

报告以反思练习结束,激发读者思考极端场景:

“你是否能想象出1百万智能体为你工作的场景?”

“合规自动化后,核心竞争力何在?”

“如果你的竞争对手用超越你使用AI的方式来使用AI,你将何去何从?”

这些问题提醒未来未定,但战略选择决定一切。

结合当下AI与会计趋势,我们可以借鉴的就是加速AI合规工具开发,推动教育改革,拥抱代理AI提升效率。

会计师们,AI不是威胁而是机会。

行动起来,从试点用例开始重塑你的事务所。

未来属于那些平衡创新与责任的先驱。

Deep Dive into the 2025 AI Accounting Report from the American CPA Association

CPA.com is a website fully owned by the American Institute of CPAs (AICPA).

In June 2025, CPA.com released the “2025 AI in Accounting Report.” It is based on AICPA and CPA.com AI accounting and finance workshops, ecosystem research, and talks with vendors and practitioners. It looks at the state, themes, trends, and future of AI in accounting.

The report says AI has moved from test use to infrastructure, and from automation to boosting the power of accountants.

Current state of AI in accounting

Today, AI has moved from the 2023 ChatGPT boom to a more mature stage.

Large language models (LLMs) like GPT-4o and Claude 3.5 handle not only text but also multi-modal tasks (like image analysis and real-time decisions).

Agentic AI has started to do multi-step tasks on its own.

Rules like the EU AI Act and U.S. NIST standards are getting better.

At the same time, accounting trends stress digital change.

Cloud accounting tools like QuickBooks and Sage Intacct are common. Data privacy rules (GDPR, SOC 2) are stronger. The focus is shifting from compliance to advisory work.

The core view of the report is:

AI does not replace accountants. It makes them stronger and frees them from repeat tasks so they can focus on high-value insight.

The executive summary at the start hits the key point:

The accounting field is going through deep change, and AI has moved from tech progress to a strategic must-have.

CPA.com's AI push in 2023 aimed to “unpack” AI. Two years later, the report says AI is now in workflow automation, smart audit helpers, gen-AI research, and agent task work.

CPA.com President and CEO Erik Asgeirsson says: “AI is a catalyst. It gives a new level of chance for firms and finance teams to grow their role as trusted advisors.”

The current state shows an inflection point:

It is harder to balance old compliance duties with client needs and market change.

Talent is short, and work is more complex, so old ways will not work.

AI is remaking operations. It not only automates tasks but also rethinks workflows, how we make decisions, and what services we sell.

Key numbers include: tax prep tools with over 80% automation; LLM research tools that cut doc review time by more than 50%; agentic AI that helps with client chats and account reconciliation.

Pascal Finette, co-founder of be radical, says: “The key is no longer whether to use AI, but how fast and how well you move to an augmented model.”

This matches today's AI state very well.

In 2025, tools like Intuit's TurboTax AI or Thomson Reuters Checkpoint Edge deliver high-automation tax filing.

OpenAI's GPT series and Google's Gemini are widely used in accounting research, cutting manual search time.

On trends, China's accounting rules are closer to global rules, with a push on digital audit (like help from blockchain). Worldwide talent gaps make AI a “staff booster.”

Seven key strategy themes

1. AI-driven workflow automation: Firms invest in tools to speed bank recs, transaction coding, month-end close, and reports. Vendors report 30–70% time saved. This frees time for advisory and analysis. This matches tools like Sage Copilot, which act as “digital assistants” to route files and sum up details.

2. Human-in-the-loop: Even with strong automation, high-risk areas like audit and tax need human review. Systems tag items by confidence level. Accountants move from “doers” to “reviewers,” cutting risk and raising throughput. This follows AI ethics trends: avoid “black box” decisions and keep things clear.

3. Regulation and risk: AI laws speed up worldwide, like the EU AI Act, the U.S. order, and NIST standards. AI tools add risk scores, doc trace, and audit trails. China has a similar path in its interim gen-AI rules that stress compliance.

4. Data privacy and security: Client trust rests on data control. Vendors stress SOC 2, GDPR, and selective training deals. “Privacy by design” is a key point. With many AI data leaks in 2024, the field must stay alert.

5. Gen-AI for research and decisions: LLM tools like Blue J sum up complex law and give tax views, cutting research time. Now Anthropic's Claude tunes pro models to cut hallucinations.

6. Agentic AI for tasks: Agent systems do multi-step flows like return prep, recs, and variance analysis. Human watch is still key. The jump from “advice” to “action” is on the way. In 2025, agent tools like Auto-GPT show up in accounting to run chained tasks.

7. Better client experience: AI gives real-time dashboards, on-demand help, and predictive alerts. Products embed in client portals for cash flow forecasts and anomaly flags. This fits the trend: clients want “instant insight,” like Xero's AI tools.

These themes show AI is moving from point efficiency to full rebuild.

The report quotes Sage CTO Aaron Harris: “You will do more, do better, and do different.”

In today's trends, firms in China like PwC use AI to upgrade client portals, hinting at global alignment.

Six trends and five developments

Trends in waves:

AI-native firms: automation first, roles shift to strategy, lower overhead.

Domain models: accounting-specific LLMs with higher accuracy, fewer hallucinations, and compliant terms. FinBERT helps in finance.

Bigger context windows: over 1 million tokens, to read long records or tax law. Gemini 1.5 Pro hits 2 million tokens.

Audit shift: DAS for real-time risk and pattern checks. Accountants become risk guides. EY tools do real-time checks.

Tax shift to planning: prep is automated, pros focus on planning and models.

System integration: APIs link to QuickBooks and cut delay. APIs are still limited; standards are hoped for.

Five developments:

Bookkeeping automation: agents do end-to-end work like classify and make reports.

Tax prep: from hours to minutes, with only final review.

Audit analytics: slower take-up, focus on judgment.

Advisory: AI makes scenarios and budgets, and deepens talks.

Workflow gains: email triage and scheduling become base tools.

EY Global AI Assurance leader Richard Jackson says: “As tech grows, we add training wheels, controls, and safeguards so people and tech move together.”

Caseware Strategy VP Danielle Supkis Cheek says: “The accounting field is great at building oversight steps and judging outputs.”

Today, agent AI shows up in tax tools like TaxJar, and audit tools like KPMG Clara speed risk analysis.

Strategy impact and talent shift

AI raises demand for mid- and senior-level work and cuts entry-level work. Pipelines, teams, and paths get rebuilt. Hiring looks for critical thinking, learning speed, and AI teaming.

Impacts:

Skills: AI literacy is a must. Train on model ups/downs, prompts, and output checks.

Career paths: new roles like AI ops manager and prompt engineer. Need cross-function skills.

Competition: the gap will widen. Early users gain speed. Late movers lose share, talent, and trust.

Clients: faster answers, deeper insight, and proactive tips. The advisor role grows.

Talent change: entry moves up, AI skills needed. “AI-ready” programs blend training. Academic badges prove knowledge plus tech.

Training: apprenticeships return (like AICPA programs), reverse mentoring, and sims to build judgment.

Development: early AI oversight, rotate into tech teams, and move reviews to value analysis, talk skills, and AI teamwork.

Know AI and the road ahead

AI includes ML and knowledge systems. ML does classify, detect, and predict. Knowledge systems use rules and can beat LLMs on cost.

Mix: knowledge first, ML at the edge, LLM for advice. Jeff Seibert warns to tell vendor tech apart.

Future: systems will learn from use, fit to your style, and see your needs early.

Shift: from tasks to orchestration; from static to real time; from single decisions to co-work.

By 2028, Gartner says 15% of decisions will be made by agents on their own. Aaron Harris says trust will unlock step by step.

Today, GPT-4V handles images; LangChain links steps. Accounting moves to continuous audit with real-time AI watch.

Readiness roadmap and ROI notes

Four stages:

Discover (check maturity, workflow audit, ethics).

Pilot (pick cases like tax research, choose champions, set KPIs).

Operationalize (embed in workflows, set roles, train).

Scale (roll out, shift capacity, govern).

Keys to win: clean data, change leaders, and solid platforms.

Early ROI is soft: Gartner shows 5.4 hours saved per week, but 69% is lost to rework.

Real gains: 70% less manual work, 5× faster review cycles, and 2–3× more capacity.

Shane Mason, co-founder of BrooklynFi, says: “AI takes small spend to pay back. If it saves 30 minutes, it pays for itself.” In 2026–2027, end-to-end rebuilds will cut cost and shorten billing cycles.

AICPA leader Mark Koziel says: “Time saved creates capacity to serve more clients.”

Partner Brian Gough at Hancock Dana asks: “If AI takes one-third of your work, how many more clients can you serve?”

Final reflections of the report

The report ends with reflection tasks to spark bold thought:

“Can you picture one million agents working for you?”

“After compliance is automated, what is your core edge?”

“If your rival uses AI in a better way than you, what will you do?”

These questions remind us the future is open, but strategy choices decide the path.

Given today's AI and accounting trends, we should speed AI compliance tools, push education change, and use agent AI to raise efficiency.

Accountants, AI is not a threat but a chance.

Take action and start with pilot use cases to remake your firm.

The future belongs to those who balance new ideas with duty.