Two-Thirds of Audit Firms Run AI, but Most Demand Human Validation

Two-thirds of audit and accounting firms now run artificial intelligence in some form, according to an IDC study of 1,005 professionals across six country markets, sponsored by the audit-software firm Caseware. The same survey finds that 64% of respondents say human validation should always be required for any AI output used to support professional conclusions, a hard line the profession has drawn alongside the technology it is rapidly adopting.

The numbers come from the first wave of a planned three-part press series Caseware is rolling out from a December 2025 survey. The technology has clearly arrived. The data points to the layer of validation and governance the profession is now demanding around the technology, and to the cost and talent gap that will decide which firms get to build it.

The Adoption Moment

For global audit and accounting firms, AI has crossed the experimentation threshold. 66% of respondents say AI is already embedded in their firm strategy, deployed in select functions, or running in active pilots, per the IDC study of 1,005 audit and accounting professionals. The question has moved on from if to how deep adoption runs.

Optimism about the result is broad. 53% of global respondents strongly agree or agree that AI tools can improve audit quality, and 76% believe AI will fundamentally transform audit and accounting over the next decade. AI is increasingly embedded in everyday processes, including data analysis, documentation, and large-scale transaction review, per the April 28 US findings on AI validation. The US runs hotter than the global average on both adoption and validation. In the US, 63% agree AI can improve audit quality, ten percentage points above the global figure, and 53% of US firms say they plan to adopt or expand their AI use within the next two years.

Caseware, the Toronto-based software firm that sponsored the research and operates in more than 130 countries serving over 29,000 firms, sees the data as a moment of professional clarity.

What stands out in this research is not hesitation about AI, but clarity. Audit and accounting professionals recognize the value AI can bring, while setting firm expectations around validation, judgment and accountability.

David Marquis, chief executive officer of Caseware, made the comment in the February 17, 2026 release of the first wave of findings. The line sits inside an argument that the adoption question has been settled and the open question is governance.

The Hard Line on Human Validation

Adoption is not the same as trust. Nearly two-thirds of respondents, 64%, say auditors should always validate AI outputs before relying on them in professional conclusions. In the United States, the share rises to 73%, the most emphatic validation mandate of any country in the survey.

The same study shows the risk concern is real and not confined to skeptics. 36% of respondents globally say AI could undermine professional judgment, and the figure is identical for US respondents.

The validation requirement is a guardrail the profession is building in front of a technology it is already using, not a wall it is putting up against further adoption. The 36% worry is high enough to matter, low enough that the guardrail answer won. The mandate is consistent across the survey’s six country markets, with the United States the largest cohort at 39.8%, followed by the United Kingdom at 15%, Canada at 14.9%, Australia at 10.1%, Germany at 10%, and the Netherlands at 10%.

The mandate is also a directional signal. The firms that skip the validation step will be the ones whose AI-augmented opinions the market is least willing to accept.

• 66% of global respondents say AI is embedded in firm strategy, in select functions, or running in pilots.
• 64% of global respondents say auditors should always validate AI outputs in professional conclusions; 73% in the US.
• 76% believe AI will fundamentally transform the profession over the next decade.
• 36% say AI could undermine professional judgment (global equals US).
• 34% cite cost of implementation as the single biggest barrier; 30% cite a lack of technical talent.

Cost and Talent, Not Skepticism, Are the Real Ceiling

The barriers to AI adoption in audit and accounting are practical, not philosophical. When asked to name the single biggest obstacle to AI adoption in their practice, 34% of respondents pointed to cost of implementation. The runner-up was a lack of technical talent, cited by 30%. Together those two answers capture nearly two-thirds of the profession.

The same trade shows up in skill priorities. When the IDC survey asked what skill would become most critical for accounting professionals in an AI-driven world, US respondents picked data analysis and interpretation (38%), five percentage points ahead of the global figure of 33%. Technology and AI literacy was a distant second, at 28% in the US, exactly matching the global number. The profession is telling itself it needs analysts who can read AI output, not just engineers who can build it. The current wave of AI implementation, per the AICPA and CPA.com Audit Transformation Survey, is still mostly about saving time, with quality and role redefinition as the next agenda item.

The Bias Problem and the Framework Response

The validation mandate is one part of the new infrastructure the profession is asking for. The other parts are surfaced in the March 10 release on bias and global frameworks, the second wave of the same IDC study. 79% of respondents rate the risk of algorithmic bias in AI systems used for critical functions such as risk assessment, fraud detection, or decision-making support as moderately, very, or extremely significant. Two-thirds of respondents, 66%, believe there is an urgent need for a globally harmonized AI framework for audit and assurance, the same share that already have AI in their firm strategy.

The willingness to compromise for safety is the third signal. 55% of audit leaders say they are willing or very willing to trade some level of AI performance in exchange for stronger security or safety measures. The trade is a stated preference for slower, more controlled AI over faster, less controlled AI.

The profession is entering a defining phase of AI maturity. Adoption is accelerating, but the competitive advantage will come from how effectively firms embed governance, harmonize with standards and operationalize responsible AI at scale.

Mickey North Rizza, group vice-president of enterprise software at IDC, made the comment in the March 10 release, where he framed the next phase of AI in audit as a governance race. The same survey period is also the period in which the regulatory floor under AI in audit is being laid down, with the EU AI Act’s August 2 enforcement deadline bringing an accountability regime on top of the AI systems that touch European audit and accounting clients, and US firms with cross-border books set to operate against both regimes.

What an AI-Augmented Audit Actually Looks Like

The abstract numbers land on a concrete workflow. In practice, AI in audit and accounting today shows up in four overlapping functions, each of which is in active use at firms that have moved past the pilot stage. The four areas cover the work auditors are doing right now.

The first function is document work: reading and summarizing contracts, processing large volumes of supporting documents, and pulling the items most relevant to an audit workpaper.

The second function is communications. AI drafts or suggests language for memos, client letters, and internal notes, with the auditor reviewing and signing off. The third function is data analysis, where AI spots anomalies in general-ledger and sub-ledger data, flags patterns a manual review would miss, and creates the visualizations that support a finding. The fourth function is the new edge, agentic AI tasks: chained workflows that can access the general ledger, fill a workpaper, send a cash confirmation to a bank, and route a flagged discrepancy to a human reviewer.

Caseware’s own answer to this is AiDA, a digital assistant integrated into the Caseware Cloud platform, per a February 2026 Journal of Accountancy profile. AiDA retrieves information from a firm’s own body of documentation, analyzes documents, generates memos, and answers audit and accounting questions, with the explicit design choice that it pulls from the firm’s own policies and memos rather than the open web. The same shape of deal is showing up across enterprise software, with AI layered on top of established systems: Canals’ $35 million raise for an AI layer over ERP is the parallel example.

A Profession Splitting in Two

The data points to a profession sorting itself into two tiers. The first tier, made up of the firms that can pay the implementation cost, hire or train the technical talent, and build the validation, governance, and bias-control stack that the survey’s respondents now treat as non-negotiable, will set the professional standard for AI-augmented audit work. The second tier, made up of the firms that cannot or do not, will run pilots, fall behind on the validation mandate, and watch their work get re-performed or re-priced by the firms that did the build-out. The two-tier read is a logical extension of the cost and talent barriers the survey documents.

The two-tier shape is already visible in the pipeline. US firms are 53% likely to adopt or expand AI use within the next two years, and 73% of US respondents demand human validation, a combination that sets a high bar for both capability and governance. Smaller firms that already struggle with talent cost and training budgets will find the new floor difficult to clear without partnership, consolidation, or platform-level tooling that bakes in the validation step.

The IDC study, sponsored by Caseware, is itself a signal of where the industry thinks the next layer of value lives, with the firm’s framing in the press series being that the question has moved past whether to adopt AI to how to deploy it. “In the US, AI is no longer something firms are piloting on the side. It is being built into how work gets done,” said David Marquis, chief executive officer of Caseware, in the April 28 US release.

Frequently Asked Questions

What percentage of audit and accounting firms use AI?

66% of the 1,005 audit and accounting professionals canvassed in the December 2025 IDC survey say AI is already part of their firm strategy, used in specific functions, or working through pilot projects. The same share holds in the United States, where 53% of firms also say they plan to adopt or expand their AI use within the next two years.

Do auditors need to validate AI outputs?

64% of global respondents, and 73% in the United States, want auditors to validate every AI output before it factors into an audit conclusion. 36% of respondents globally and in the US add that AI could risk undermining professional judgment without that validation step.

What is the biggest barrier to AI adoption in audit firms?

Cost of implementation, named by 34% of respondents, tops the list. A shortage of technical talent, named by 30%, comes second. Together those two answers capture nearly two-thirds of the profession, and both point to practical constraints rather than skepticism about the technology itself.

Are accounting firms worried about AI bias?

79% of respondents to the IDC survey treat algorithmic bias in critical-function AI as a moderately, very, or extremely significant risk. They cite risk assessment, fraud detection, and decision support as the areas most exposed. 66% also want a globally harmonized AI framework for audit and assurance, an urgency call that pairs with the same share of firms that already have AI in their firm strategy.

What is AiDA?

AiDA is the AI assistant Caseware developed for its Caseware Cloud platform. It works from a firm’s own documentation to draft memos, analyze documents, and answer audit and accounting questions, with the source links auditors need to verify each answer. The product is positioned as a digital assistant rather than a co-pilot, designed to enhance the work of auditors without replacing their professional judgment.