Enterprise & Work

KPMG Puts AI Agents Inside U.S. Audit Workflows With Clara

KPMG's KCi AI agents now assist more than 125 U.S. audit teams, with full private-company deployment planned for 2027 and auditors retaining final judgment.

KPMG Brings AI Agents Into U.S. Audits With Clara Intelligence
KPMG Brings AI Agents Into U.S. Audits With Clara Intelligencemwitt1337 / pixabay
By Rebecca Stone5 min read

Why it matters

  • KPMG announced KCi on November 17, with rollout to more than 125 U.S. teams and 500 teams globally following 2026 use
  • Full deployment across private company audits is planned for 2027, with a phased rollout for public company audits
  • On some engagements, KCi's transaction analysis shortened timelines by as much as three and a half weeks
  • Nearly 1,000 Audit and Assurance interns trained at KPMG's Lakehouse facility on evaluating AI-assisted work
  • Technology partners include Anthropic, Google Cloud, Microsoft, and OpenAI

KPMG is embedding specialized AI agents directly into U.S. audit workflows through KPMG Clara intelligence, or KCi, while keeping auditors responsible for reviewing the work and making final judgments. The firm announced the deployment on November 17, extending its existing KPMG Clara platform across activities that include risk assessment, transaction analysis, engagement coordination, and testing support.

The rollout has reached more than 125 U.S. teams and 500 teams globally following use during 2026, according to KPMG. The firm's next objective is deployment across all private company audits in 2027, with a separate, phased rollout planned for public company audits. KPMG presents these as deployment plans. The firm does not claim the platform already handles every engagement.

The announcement matters because it shows how professional services firms are moving beyond standalone AI assistants. Instead of asking auditors to transfer information between disconnected tools, KPMG is placing agents inside the processes where client data, accounting guidance, documentation, and review already meet. The audit profession carries high accountability and strict professional standards, which makes it a demanding test case for whether agentic AI can operate under governance rather than as an unsupervised shortcut.

What do the agents actually do?

For risk assessment and planning, KCi agents analyze engagement data, develop preliminary considerations, and prepare first-pass documentation. KPMG says the intended benefit is earlier visibility into potential risks, giving audit teams more time to examine complex areas before year-end. The firm adds a qualification of its own: preliminary assessments still need professional evaluation. Producing a plausible draft is not the same as establishing an audit conclusion.

Other agents handle coordination work:

  • Organizing milestones and information requests
  • Surfacing technical accounting guidance
  • Managing engagement activities through team dashboards

These functions may be less conspicuous than automated analysis, but they address the practical work of keeping an engagement moving and making requests visible to client accounting teams.

KPMG also describes capabilities that automate elements of control testing and sample evaluation. Auditors remain responsible for reviewing and evaluating those results. The operating model therefore shifts some effort from assembling work toward directing it, supervising it, and deciding whether the underlying evidence supports the findings.

How broad is the transaction analysis?

One of the more substantial claims in the announcement concerns transaction coverage. According to KPMG, automated data pipelines can analyze complete transaction populations in selected audit areas, identifying unusual items, trends, and anomalies for targeted investigation.

That qualification carries weight. Analyzing every transaction within a selected area does not mean every aspect of an audit has been automated. Flagged anomalies do not necessarily establish an error. What the approach gives auditors is a broader analytical starting point for deciding where to investigate.

KPMG reports concrete results from selected engagements. On some, the approach shortened timelines by as much as three and a half weeks. It also removed the need for clients to retrieve large volumes of individual supporting documents manually. The announcement does not present that result as an average across all engagements. It is a company-reported outcome from selected work, not a universal measure of the platform's performance.

Who is accountable for the output?

KPMG says KCi operates under its audit methodology, professional standards, and Trusted AI framework. Engagement teams can trace outputs back to underlying data sources, inspect the information, and independently corroborate findings before incorporating documentation into the firm's system of record.

This is the central governance distinction in the rollout. Agents can generate and organize material, but responsibility does not transfer to the software. Christian Peo, vice chair of Audit and Assurance at KPMG U.S., framed the opportunity as giving auditors more time to understand information and challenge what does not make sense, rather than automating their judgment.

The traceability design responds to a structural concern in AI-assisted professional work. If an auditor cannot trace a machine-generated finding to its underlying data, the finding cannot support an audit conclusion. KPMG's answer is to build that trace into the workflow itself rather than treat it as an afterthought.

The technology foundation combines proprietary KPMG capabilities with enterprise technologies available through relationships with Anthropic, Google Cloud, Microsoft, and OpenAI. The announcement does not specify which provider powers each individual KCi function, so those relationships should not be read as a component-by-component architecture.

How does the rollout change auditor roles?

Alongside KCi, KPMG is adjusting role definitions, moving supervision and review earlier in engagements, and standardizing how data is organized. The firm says nearly 1,000 Audit and Assurance interns attended training at its Lakehouse facility earlier this year. The training covered critical thinking, professional skepticism, and evaluation of AI-assisted work.

The scale of the intern training signals how KPMG views the shift. The firm is not only redefining the roles of current auditors. It is preparing its incoming workforce to supervise AI-assisted work rather than perform every assembly task manually. KPMG describes this as redefining roles and training staff to supervise AI-assisted work rather than just adding a model.

What is at stake for the profession?

The broader significance is that deploying agents in a high-accountability profession involves more than adding a model to existing software. Data preparation, review timing, traceability, and staff training all shape whether faster preparation translates into useful audit work. An agent that produces documentation quickly has limited value if the audit team cannot verify it or fold it into a defensible conclusion.

The timeline compression KPMG reports, up to three and a half weeks on some engagements, points to the commercial pressure behind the deployment. Audit firms compete on efficiency and on the burden they place on clients. Reducing manual document retrieval changes the client experience as much as the internal workflow, and competitors face the same incentives to automate preparation work.

KPMG's planned expansion in 2027 will test how consistently that model can operate across engagements while preserving the human scrutiny on which audit conclusions depend. The firm has staked its position on a specific claim: agents do the assembling, and auditors keep the judgment. Whether that boundary holds at full deployment scale is the question the next phase of the rollout will answer.

Source: unite.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. KPMG Ran 100 AI Agents Through a Full FTSE 250 Audit — and the Bots Dug Deeper
  2. New Relic's Autopilot Kubernetes Agent Hits Public Preview
  3. 94% Pilot Adoption Drove Wotton Kearney to Take Harvey Firmwide
  4. Uptiq launches GovernAI to govern AI agents in banking at no extra cost
  5. SAS: Corporate AI's 2027 Test Will Be Trust, Not Deployment Speed

« Previous articleNext article »