广告
加载中

毕马威百个AI代理并行审计 提问比审计师更深入

亿邦AI 2026-11-19 10:54
亿邦AI 2026/11/19 10:54

邦小白快读

EN
全文速览

总1:先看试验规模与结果。

1. 毕马威今年首次对一家英国富时250指数成分公司完成AI代理全程参与的并行审计,超过100个AI代理在年度审计各阶段与常规团队同步作业,最终审计意见仍基于常规审计。

2. 内部称Project Exponential。AI可独立完成数据收集、分析与记录并起草审计结论,部分情形比常规团队走得更深,提问有时比审计师更好。

总2:再看人工与限制。

1. 人工必须介入,曾有一个AI代理无法把监管规定与公司文件正确匹配;员工负责设目标、给背景、定期复核并在定稿前质疑结论。

2. 系统提供完全可见性与可追溯性,每次交互记录到单个字符和时间戳;计划扩展到最多20项更大审计,但技术不足以支撑数千项,且只有少数公司数据足够干净。

总3:监管与收费。

1. FRC将独立性规则适用于AI,关注与客户分享AI发现是否损害独立性,咨询销售说辞可能构成明显威胁,重大内控缺陷仍须报告。

2. AI带来明显时间节约,但未明确是否降低收费;毕马威国际已以AI降本为由协商更低费用,普华永道也有并行审计项目。

总1:AI审计会改变专业服务产品的交付与定价。

1. 毕马威用超过100个AI代理在审计各阶段同步作业,代理独立收集数据、分析记录并起草结论,部分技术提问比审计师更深入。

2. 这带来明显时间节约机会,但毕马威不愿透露是否带来更低收费;其国际部门已以AI降低工作成本为由与自身审计师协商更低费用,说明服务定价面临重估。

总2:客户洞察可能成为新卖点,也可能反噬品牌信任。

1. AI可分析远多于传统抽样交易,发现低效流程、潜在成本节约和常规审计难见的模式,这些发现可能被客户视为额外服务机会。

2. FRC担心这类发现被当作咨询业务销售说辞,分享时可能对审计独立性构成明显威胁;品牌若处理不好,会把技术能力变成信任风险。

总3:产品化落地有门槛。

1. 技术尚不足以支撑数千项审计,只有少数公司拥有足够干净、可供AI可靠使用的数据;先试点再扩展更稳妥。

2. 人工仍需设定目标、提供背景、复核和质疑结论,品牌传播应强调人机协同与完全可追溯,而非纯无人审计。

总1:政策风向要先看清。

1. 英国FRC近日发布指引,把现行独立性规则适用于AI使用,核心是审计师何时可在不损害独立性时与客户分享AI审计发现。

2. 对审计客户的某些非审计服务有严格限制或全面禁止,因为事务所可能日后复核自己的建议,存在不愿让客户不满的动机;AI使冲突更尖锐。

总2:机会与风险并存。

1. 机会在于AI代理可分析远多于抽样的交易,浮现低效流程、潜在成本节约和常规审计不会出现的模式,并带来明显时间节约;毕马威计划扩展到最多20项更大审计,普华永道也已开展并行审计项目。

2. 风险在于分享这些发现可能被视为额外服务宣传或咨询销售说辞,对独立性构成明显威胁;重大内控缺陷仍须报告,强制披露不计入非审计服务。

总3:可学习点。

1. 人必须介入,曾有一个代理无法正确匹配监管规定与公司文件;员工要设目标、给背景、复核并质疑结论。

2. 规模化受技术和数据限制,只有少数公司数据足够干净;系统全程可见可追溯,交互记录到字符和时间戳,可作为合规运营参考。

总1:数字化试点要有人机协同。

1. 毕马威让超过100个AI代理在审计各阶段与常规团队同步作业,代理独立完成数据收集、分析和记录并起草结论,部分环节比常规团队走得更深。

2. 但人工仍至关重要:员工设定目标、提供背景、定期复核并质疑结论;曾有一个代理无法把监管规定与公司文件正确匹配,说明关键匹配和判断仍需人介入。

总2:数据与流程是落地前提。

1. 系统提供完全可见性与完全可追溯性,代理与文件的每次交互都被记录,细化到单个字符和时间戳,这对工厂推进数字化和可审计流程有启示。

2. 目前技术不足以支撑数千项审计,只有少数公司拥有足够干净、可供AI可靠使用的数据;工厂应先治理数据质量,再扩大AI应用。

总3:AI能发现流程改善机会。

1. AI可分析远多于传统抽样的交易,发现低效流程、潜在成本节约和常规审计难见的模式,可作为生产与运营降本线索。

2. 收费结构可能变化,毕马威看到时间节约但未明确降价,其国际部门已以AI降本为由协商更低费用,提示企业采购AI服务时要关注成本传导。

总1:行业出现AI并行审计新趋势。

1. 毕马威今年首次让超过100个AI代理全程参与一家英国富时250成分公司的并行审计,代理在每一阶段与常规团队同步作业,独立收集数据、分析记录并起草结论。

2. 部分情形下AI比常规团队走得更深,技术提问有时比审计师更好;普华永道也已开展多个并行审计项目,说明专业服务的技术竞赛在升温。

总2:客户痛点集中在数据、责任与独立性。

1. 技术尚不足以支撑数千项审计,只有少数公司拥有足够干净、可供AI可靠使用的数据;一个代理曾无法正确匹配监管规定与公司文件。

2. FRC把独立性规则适用于AI,担心AI发现被用作咨询销售说辞,分享时构成明显威胁;审计师仍须解释如何使用AI、质疑结果并为证据负责。

总3:解决方案是人机协同加可追溯。

1. 员工设定目标、提供背景、定期复核并在定稿前质疑结论,项目由合伙人和总监带高级经理及初级员工推进。

2. 系统提供完全可见性与可追溯性,交互记录到字符和时间戳;先试点再扩至最多20项更大审计,是服务商可借鉴的落地路径,同时收费模式面临重估。

总1:平台化AI服务要先解决数据与规模问题。

1. 毕马威的试验显示,超过100个AI代理可在审计各阶段同步作业,但技术尚不足以支撑数千项审计,且只有少数公司拥有足够干净、可供AI可靠使用的数据。

2. 这意味着平台若想承载大规模AI代理,需要先把客户数据治理、接入标准和可靠使用条件做扎实,否则难以复制扩张。

总2:运营管理要保留人工复核与全程留痕。

1. 员工为代理设定目标、提供背景、定期复核工作,并在审计意见定稿前质疑结论;曾有一个代理无法正确匹配监管规定与公司文件,说明关键节点必须人工介入。

2. 系统提供完全可见性与完全可追溯性,代理与文件的每次交互都被记录,细化到单个字符和时间戳,可作为平台运营和风险管理的参考。

总3:监管风向要提前规避。

1. FRC将独立性规则适用于AI,关注与客户分享AI发现是否损害独立性,咨询销售说辞可能构成明显威胁;重大内控缺陷仍须报告。

2. 收费结构也在变化,毕马威看到时间节约但未明确降价,其国际部门已以AI降本为由协商更低费用;平台招商和定价需预留合规与成本传导空间。

总1:产业新动向是AI代理进入高责任专业服务。

1. 毕马威首次对一家英国富时250成分公司完成AI代理全程参与的并行审计,超过100个代理在每一阶段与常规团队同步作业,最终意见仍基于常规审计,项目名为Project Exponential。

2. AI能独立收集数据、分析记录并起草结论,部分环节比常规团队更深,提问有时优于审计师;但一个代理无法正确匹配监管规定与公司文件,显示责任边界仍是问题。

总2:新问题集中在独立性、责任与收费。

1. FRC发布指引,将独立性规则适用于AI,核心是审计师何时可与客户分享AI发现;AI可分析远多于抽样的交易,发现低效流程、成本节约和常规审计难见的模式,可能被视为咨询销售说辞并构成明显威胁。

2. 审计师仍须解释如何使用AI、质疑结果并为证据负责;重大内控缺陷仍须报告,强制披露不计入非审计服务。

总3:商业模式与政策启示。

1. 规模化受技术不足和客户数据不够干净限制,毕马威计划扩至最多20项更大审计;系统全程可见可追溯,交互记录到字符和时间戳。

2. 收费结构面临重估:毕马威看到时间节约但未明确降价,其国际部门已以AI降本为由协商更低费用;德勤内部图表显示到2035年按小时计费咨询收入占比将缩至极窄,提示研究按结果或价值收费的替代模式。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

Summary 1: First, the scale and results of the trial.

1. KPMG this year completed, for the first time, a parallel audit of a UK FTSE 250 company with AI agents involved throughout; more than 100 AI agents worked alongside the regular team at every stage of the annual audit, while the final audit opinion was still based on the conventional audit.

2. Internally called Project Exponential. AI can independently collect data, analyze and document it, and draft audit conclusions; in some cases it went further than the regular team, and its questions were sometimes better than the auditors'.

Summary 2: Human involvement and limits.

1. Humans must intervene; one AI agent failed to correctly match regulatory rules with company documents. Staff set objectives, provide context, review regularly, and challenge conclusions before final sign-off.

2. The system provides full visibility and traceability, with every interaction recorded down to individual characters and timestamps. There are plans to expand to up to 20 larger audits, but the technology cannot yet support thousands, and only a few companies have sufficiently clean data.

Summary 3: Regulation and fees.

1. The FRC applies independence rules to AI, focusing on whether sharing AI findings with clients undermines independence; consulting sales pitches may pose a clear threat, and material internal control weaknesses must still be reported.

2. AI brings clear time savings, but it is unclear whether fees will fall; KPMG International has negotiated lower fees on the grounds of AI-driven cost reduction, and PwC also has parallel audit projects.

Summary 1: AI audit will change the delivery and pricing of professional services.

1. KPMG used more than 100 AI agents working in parallel at every audit stage; the agents independently collected data, analyzed records, and drafted conclusions, and in some cases their technical questions went deeper than those of auditors.

2. This creates clear opportunities for time savings, but KPMG declined to say whether it would lead to lower fees; its international arm has already negotiated lower fees with its own auditor on the grounds that AI reduces work costs, suggesting service pricing faces revaluation.

Summary 2: Client insights could become a new selling point, but could also backfire on brand trust.

1. AI can analyze far more transactions than traditional sampling, identifying inefficient processes, potential cost savings, and patterns that conventional audits rarely see; clients may view these findings as additional service opportunities.

2. The FRC worries that such findings could be used as consulting sales pitches and that sharing them may pose a clear threat to audit independence; if handled poorly, a brand could turn technical capability into a trust risk.

Summary 3: Productization has thresholds.

1. The technology is not yet sufficient to support thousands of audits, and only a few companies have data clean enough for reliable AI use; piloting first and scaling later is the more prudent path.

2. Humans still need to set objectives, provide context, review work, and challenge conclusions; brand communications should emphasize human-machine collaboration and full traceability, not fully unattended audits.

Summary 1: First understand the policy direction.

1. The UK FRC recently issued guidance applying existing independence rules to AI use, with the core question being when auditors can share AI audit findings with clients without compromising independence.

2. There are strict restrictions or outright bans on certain non-audit services for audit clients, because firms may later review their own advice and have an incentive not to displease clients; AI makes conflicts sharper.

Summary 2: Opportunities and risks coexist.

1. The opportunity is that AI agents can analyze far more transactions than sampling, surface inefficient processes, potential cost savings, and patterns that conventional audits would not reveal, and deliver clear time savings; KPMG plans to expand to up to 20 larger audits, and PwC has also launched parallel audit projects.

2. The risk is that sharing these findings may be seen as promotion of additional services or consulting sales pitches, posing a clear threat to independence; material internal control weaknesses must still be reported, and mandatory disclosures are not counted as non-audit services.

Summary 3: Lessons to learn.

1. Humans must be involved; one agent failed to correctly match regulatory rules with company documents. Staff need to set objectives, provide context, review work, and challenge conclusions.

2. Scaling is limited by technology and data, and only a few companies have sufficiently clean data; the system is fully visible and traceable, with interactions recorded down to characters and timestamps, which can serve as a reference for compliant operations.

Summary 1: Digital pilots need human-machine collaboration.

1. KPMG had more than 100 AI agents work alongside the regular team at every audit stage; the agents independently completed data collection, analysis, and documentation and drafted conclusions, and in some steps went further than the regular team.

2. But humans remain critical: staff set objectives, provide context, review regularly, and challenge conclusions; one agent failed to correctly match regulatory rules with company documents, showing that key matching and judgment still require human intervention.

Summary 2: Data and processes are prerequisites for implementation.

1. The system provides full visibility and full traceability; every interaction between the agent and documents is recorded down to individual characters and timestamps, offering lessons for factories advancing digitalization and auditable processes.

2. Current technology cannot support thousands of audits, and only a few companies have data clean enough for reliable AI use; factories should govern data quality first, then expand AI applications.

Summary 3: AI can identify process improvement opportunities.

1. AI can analyze far more transactions than traditional sampling, identifying inefficient processes, potential cost savings, and patterns rarely seen in conventional audits, which can serve as leads for production and operational cost reduction.

2. Fee structures may change; KPMG sees time savings but has not specified fee cuts, and its international arm has negotiated lower fees on the grounds of AI-driven cost reduction, suggesting that companies procuring AI services should watch cost pass-through.

Summary 1: A new trend of AI parallel audits is emerging in the industry.

1. KPMG this year, for the first time, had more than 100 AI agents participate throughout a parallel audit of a UK FTSE 250 company; the agents worked in parallel with the regular team at every stage, independently collecting data, analyzing records, and drafting conclusions.

2. In some cases AI went further than the regular team, and its technical questions were sometimes better than the auditors'; PwC has also launched several parallel audit projects, indicating that the technology race in professional services is heating up.

Summary 2: Client pain points center on data, accountability, and independence.

1. The technology is not yet sufficient to support thousands of audits, and only a few companies have data clean enough for reliable AI use; one agent failed to correctly match regulatory rules with company documents.

2. The FRC applies independence rules to AI and worries that AI findings could be used as consulting sales pitches, posing a clear threat when shared; auditors must still explain how AI is used, challenge results, and take responsibility for evidence.

Summary 3: The solution is human-machine collaboration plus traceability.

1. Staff set objectives, provide context, review regularly, and challenge conclusions before final sign-off; the project was led by partners and directors with senior managers and junior staff.

2. The system provides full visibility and traceability, with interactions recorded down to characters and timestamps; piloting first and then expanding to up to 20 larger audits is an implementation path service providers can learn from, while fee models face revaluation.

Summary 1: Platform-based AI services must first solve data and scale issues.

1. KPMG's trial shows that more than 100 AI agents can work in parallel at every audit stage, but the technology cannot yet support thousands of audits, and only a few companies have data clean enough for reliable AI use.

2. This means that if a platform wants to host large-scale AI agents, it must first make client data governance, access standards, and conditions for reliable use solid; otherwise replication and expansion will be difficult.

Summary 2: Operations management must retain human review and full record-keeping.

1. Staff set objectives for the agents, provide context, review work regularly, and challenge conclusions before the audit opinion is finalized; one agent failed to correctly match regulatory rules with company documents, showing that human intervention is essential at key nodes.

2. The system provides full visibility and full traceability; every interaction between the agent and documents is recorded down to individual characters and timestamps, which can serve as a reference for platform operations and risk management.

Summary 3: Regulatory trends should be anticipated and avoided.

1. The FRC applies independence rules to AI and focuses on whether sharing AI findings with clients undermines independence; consulting sales pitches may pose a clear threat, and material internal control weaknesses must still be reported.

2. Fee structures are also changing; KPMG sees time savings but has not specified fee cuts, and its international arm has negotiated lower fees on the grounds of AI-driven cost reduction; platform recruitment and pricing need to leave room for compliance and cost pass-through.

Summary 1: The new industry development is AI agents entering high-accountability professional services.

1. KPMG for the first time completed a parallel audit of a UK FTSE 250 company with AI agents involved throughout; more than 100 agents worked alongside the regular team at every stage, while the final opinion was still based on the conventional audit; the project is called Project Exponential.

2. AI can independently collect data, analyze records, and draft conclusions, and in some steps it went deeper than the regular team, with questions sometimes better than auditors'; but one agent failed to correctly match regulatory rules with company documents, showing that accountability boundaries remain an issue.

Summary 2: New issues center on independence, accountability, and fees.

1. The FRC issued guidance applying independence rules to AI, with the core question being when auditors can share AI findings with clients; AI can analyze far more transactions than sampling, identify inefficient processes, cost savings, and patterns rarely seen in conventional audits, and may be viewed as consulting sales pitches posing a clear threat.

2. Auditors must still explain how AI is used, challenge results, and take responsibility for evidence; material internal control weaknesses must still be reported, and mandatory disclosures are not counted as non-audit services.

Summary 3: Business model and policy implications.

1. Scaling is limited by insufficient technology and client data that is not clean enough; KPMG plans to expand to up to 20 larger audits; the system is fully visible and traceable, with interactions recorded down to characters and timestamps.

2. Fee structures face revaluation: KPMG sees time savings but has not specified fee cuts, and its international arm has negotiated lower fees on the grounds of AI-driven cost reduction; internal Deloitte charts show that by 2035 hourly-billed consulting revenue will shrink to a very narrow share, suggesting research into alternative outcome- or value-based fee models.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

毕马威今年首次对一家英国富时250指数成分公司完成了一场由人工智能代理全程参与的并行审计。超过100个AI代理在年度审计的每一个阶段与常规审计团队同步作业,最终的审计意见仍基于常规审计得出。毕马威内部将这一试验命名为"Project Exponential"。

毕马威英国审计主管凯丝·伯内特介绍试验结果时称,在若干情形下,AI"能够比常规审计团队走得更深",部分环节的技术提问"比事务所自己的审计师更好"。这些代理独立完成数据收集、分析与记录,并起草审计结论。但在一个领域,人工必须介入,因为一个AI代理无法将一套监管规定与公司文件正确匹配。

毕马威方面强调,人的参与仍然"至关重要"。员工为代理设定目标、提供背景信息、定期复核工作,并在审计意见定稿前对结论提出质疑。该项目由两名合伙人和一名总监与高级经理及初级员工共同推进。事务所计划将试验扩展至最多20项更大规模的审计,但目前技术尚不足以支撑数千项审计所需的规模,且只有少数公司拥有足够干净、可供AI代理可靠使用的数据。毕马威称,系统提供"完全可见性与完全可追溯性",代理与文件的每一次交互都被记录,细化到单个字符和时间戳。

监管层面,英国会计监管机构财务报告委员会(FRC)近日发布指引,将现行独立性规则适用于AI的使用。核心问题在于,审计师何时可以在不损害独立性的前提下,与客户分享AI审计产生的发现。对审计客户的某些非审计服务适用严格限制或全面禁止,因为一家为某公司提供咨询的事务所日后可能复核自己的建议,存在不愿让客户不满的动机。

FRC认为,AI代理使这一冲突更加尖锐。传统审计师抽取样本,而AI代理可以潜在地分析多得多的交易,由此浮现的发现可能超出审计本身,包括低效流程、潜在成本节约以及常规审计中可能永远不会出现的模式。FRC担心,这类发现可能被视为对额外服务的宣传,或被用作咨询业务的销售说辞,在特定情形下分享这些发现可能对审计独立性构成"明显威胁"。

FRC监管标准执行董事马克·巴宾顿说,无论事务所使用何种技术,审计师仍然负有责任。审计师必须能够解释自己如何使用AI、质疑其结果,并为证据负责。某些发现,如内部控制中的重大缺陷,仍须向客户报告,这类强制披露不计入非审计服务。

审计独立性问题在英国尤为敏感。建筑与外包集团Carillion于2018年倒闭后,一项议会调查认定,作为其长期审计师的毕马威在19年间未能质疑管理层"越来越离谱的数字",两家毕马威实体因此合计被罚款2100万英镑。

毕马威的试验还引出一个更根本的问题。如果AI代理能够接管初级审计师团队的大量劳动密集型工作,行业的收费结构将如何变化。毕马威看到"明显的时间节约机会",但不愿透露这是否会带来更低收费。与此同时,毕马威国际已经以AI降低工作成本为由,与其自身的审计师协商了更低费用。竞争对手普华永道也已开展多个"并行审计"项目。

德勤今年5月的一场内部会议显现出同样的压力。管理层展示的一张图表显示,到2035年,按小时计费的咨询收入在持续增长的整体市场中占比将缩至极窄的一小条。一名德勤顾问总结这一信息时说,他们"强烈暗示我们的模式已经完蛋了,我们基本上会被机器人取代"。

本文首发于 亿邦动力 官方网站

文章来源:亿邦动力

广告
微信
朋友圈

FAQ回顾

什么是AI代理并行审计?

AI代理并行审计指多个AI代理在年度审计各阶段与常规审计团队同步作业。毕马威试验中超过100个代理独立完成数据收集、分析与记录并起草审计结论,最终审计意见仍基于常规审计得出,项目名为Project Exponential。

审计行业有哪些公司在探索AI代理并行审计?

毕马威今年首次完成超100个AI代理参与的并行审计试验,并计划扩展至最多20项更大规模审计;普华永道也已开展多个“并行审计”项目;德勤内部会议则显示,到2035年按小时计费咨询收入占比将缩至极窄。

AI代理在审计中能完全替代审计师吗?

不能完全替代。毕马威强调人的参与“至关重要”,员工为代理设定目标、提供背景、定期复核并质疑结论。试验中AI在若干环节提问比审计师更深入,但有一个领域无法正确匹配监管规定与公司文件,仍需人工介入。

AI用于审计为什么可能影响审计独立性?

英国财务报告委员会(FRC)将现行独立性规则适用于AI使用,核心问题是审计师何时可与客户分享AI审计发现。AI代理可分析更多交易,发现低效流程、潜在成本节约等,若分享可能被视为额外服务宣传或咨询销售,对独立性构成“明显威胁”。

AI代理会如何影响审计行业收费模式?

毕马威看到“明显的时间节约机会”,但未透露是否带来更低收费;毕马威国际已以AI降低工作成本为由与自身审计师协商更低费用。德勤内部图表显示,到2035年按小时计费咨询收入占比将缩至极窄,传统收费模式面临压力。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0