Audit Firms Adopt AI Faster Than They Can Train Staff

Only 28% of audit and accounting firms say they are very or extremely prepared to retrain staff for AI, according to an IDC survey of 1,005 professionals. The same study found that 66% already have AI in firm strategy, in select functions, or in pilots.

Cost and a shortage of technical talent were the two largest barriers. The people who have to check the machines are the people many firms still cannot hire or teach.

The Staffing Gap Landed in the March Release

IDC fielded the work in December 2025 and published it as February 2026 IDC InfoBrief #US54248126-IB, sponsored by Caseware. Caseware sells audit software used by more than 29,000 firms in 130 countries. The first public cut, on February 17, 2026, led with adoption and the human-in-the-loop rule. Two later cuts, on March 10 and March 31, carried the staffing and training numbers that sit underneath that rule.

THE THREE CASEWARE CUTS

1. February 17, 2026: Caseware publishes the adoption and validation findings, including the 66% in-strategy share and the 64% who want every relied-on AI output checked.
2. March 10, 2026: The second release adds the security trade-off, the call for a global AI framework, and the bias warning.
3. March 31, 2026: The third release ranks the skills firms now want and states that only 28% feel very or extremely prepared to reskill staff.

Mickey North Rizza, group vice-president for Enterprise Software at IDC, treated that last gap as the constraint that will decide who gets value from the tools.

The profession is approaching AI with a clear strategy for creating value. However, success in the AI era will depend on organizations that invest in data and AI literacy, set expectations for explainable and auditable AI and develop the skills needed to work confidently alongside intelligent systems.

Mickey North Rizza, Group Vice-President, Enterprise Software, IDC, March 31, 2026

David Marquis, chief executive officer at Caseware, said on February 17 that what stood out was “not hesitation about AI, but clarity.” The later chapters of his own sponsored study show how costly that clarity is. Firms can still download the full IDC study from Caseware.

Why 64% Want Every AI Output Checked

Nearly two-thirds of respondents said auditors should always validate AI outputs used to reach professional conclusions. That demand for human validation of AI output turns every model suggestion into another review step. It is a labor rule, not a slogan.

Eighty-eight percent agreed to some degree that AI tools could undermine professional judgment. The share who agreed or strongly agreed sat at 36%, a harder cut on the same worry. Respondents also split on trust in the profession itself: 48% said AI carries a significant risk of eroding that trust, and 44% said it does not.

A slim majority, 53%, agreed or strongly agreed that AI tools can enhance audit quality. Seventy-six percent said AI will fundamentally transform audit over the next decade. Those two numbers describe a long horizon and a qualified yes on quality. They do not describe a profession ready to take the model at its word.

One anonymous respondent put the fear in a single line: “Excessive trust in AI outputs could undermine professional skepticism.” Another asked for transparency in how AI arrives at its outputs. Checking is the work that grows when the model writes faster than a reviewer can read. Firms that skip that check tend to discover accuracy problems after rollout, not in the vendor demo.

Implementation Cost Still Outranks Every Other Barrier

When IDC asked for the single biggest barrier to AI adoption, the answers added to 100%. Money came first. People came second. Rules, culture, and ethics filled the rest. Caseware later posted the full list of adoption barriers from the same questionnaire.

THE FIVE ADOPTION BARRIERS

An InfoBrief comment on cost was blunt: “High implementation costs can make it challenging for firms to fully embrace AI.” The talent line was just as direct. “Ensuring that people have the skills and confidence to work alongside AI. Without the right talent mindset, adoption will lag no matter how good the technology is.”

Always-on review is what turns those two barriers into one bill. A firm that buys a model still needs reviewers who can catch a bad classification, an incomplete source, or a biased risk score. Those reviewers are the scarce technical talent. Their time is the implementation cost that does not show up on a software invoice.

Accountants Rank Data Analysis First, Skepticism Third

Asked which skill will matter most in an AI-driven practice, respondents did not put professional skepticism on top. They put the work of reading the machine.

SKILLS FIRMS NOW WANT MOST

• Data analysis: 33% ranked it first.
• Technology and AI literacy: 28% put this second, a separate share from the 28% who feel ready to reskill.
• Critical thinking: 14% ranked it third.
• Ethics, governance and oversight: 13% put this fourth.

Those four add to 88%. Other skills split the remaining 12%. Data analysis and AI literacy together are the job description for the person who has to validate the model. Critical thinking, the old core of audit skepticism, sits behind them. The remaining 72% of respondents said they were unprepared, slightly prepared, or only moderately prepared to upskill staff for that work.

Sixty-seven percent agreed or strongly agreed that the profession needs a transformative execution rethink that reorganizes procedures and embeds AI at every stage of the workflow. The largest path they named for new client value was integrating AI and advanced analytics to improve risk detection, chosen by 39%.

As AI advances, the definition of audit expertise is changing.

A new global study from IDC, sponsored by Caseware, reveals that data analysis and AI literacy are emerging as the most important capabilities for accounting professionals as AI continues to reshape the audit… pic.twitter.com/FAF6s23Su2

— Caseware (@Caseware) March 31, 2026

Marquis said on March 10 that the question is no longer whether to adopt, but how to deploy AI “in a way that the profession can truly depend on.” Dependence, in this sample, means a reviewer who can read a data pull, interrogate a model, and still sign the file. Most firms in the sample do not claim they can train that person yet.

Nearly Half the Sample Works at Firms of 100 or Fewer

This is not a Big Four-only poll. Senior decision-makers filled it out: directors, vice presidents, C-level officers, and partners. Organization size ran across the market. Firms with 11 to 50 staff were 24.4% of the sample. Firms with 51 to 100 staff were 25.4%. Together that is 49.8%, nearly half, at 100 people or fewer. Firms with 500 or more staff were 24.8%.

A 34% cost barrier lands differently in a 40-person practice than in a global network. IDC did not publish a barrier breakout by headcount, so the survey does not prove that smaller firms are more blocked. It does prove that the voices calling AI a strategy item include a large bloc of shops that cannot spread a new specialist team across hundreds of engagements.

Geography is lopsided toward North America. The United States supplied 39.8% of respondents, the United Kingdom 15%, and Canada 14.9%. Australia, Germany, and the Netherlands each contributed about 10%. Later country cuts from the same study show the global averages hiding local orders. Australian respondents put a lack of technical talent first, then cost, then regulation. US respondents were more willing than the global sample to say AI can raise audit quality, and more insistent that every relied-on output be checked.

The IAASB Has Not Written an AI Audit Rule

Two-thirds of the IDC sample, 66% on a separate question from the adoption share, said there is an urgent need for a globally harmonized AI framework for audit and assurance. Fifty-five percent were willing or very willing to trade some AI performance for stronger security or safety. Seventy-nine percent rated the risk of algorithmic bias in systems used for risk assessment, fraud detection, or decision support as moderately, very, or extremely significant.

Standard setters have not closed that request. On February 10, 2026, the International Auditing and Assurance Standards Board published its IAASB roundtable summary on technology after sessions in the second half of 2025 with more than 240 stakeholders on six continents. Participants said the existing quality-management standards, especially ISQM 1 and ISA 220 (Revised), still provide the base for managing technology risk. They also asked for practical guidance as tools change.

At a time when the use of AI in audit and assurance is expanding rapidly, stakeholders told us clearly that global consistency and practical clarity matter.

Tom Seidenstein, Chair, International Auditing and Assurance Standards Board, February 10, 2026

In December 2025 the IAASB approved work on non-authoritative material. That is guidance, not a new International Standard on Auditing that tells a partner when an AI output counts as evidence. Until that line exists, the 64% “always validate” finding is professional opinion sitting on top of older rules about sufficient appropriate evidence and the auditor’s own judgment. North Rizza said value will come from how well firms fold the tools into methodologies, quality controls, and governance, not from the tools alone.

Vendor Tools Fail the Depth and Training Test

The same respondents who want data skills are unhappy with the help they get from software makers. The InfoBrief asked about vendor tools and training in several ways, and the scores ran in one direction.

WHERE VENDORS FALL SHORT

• Depth: 65% said their vendor’s AI tools lack depth and do not cover the latest advances.
• Usability: 63% did not believe those tools were comprehensive and user friendly.
• Features: 78% saw room for improvement on advanced features and customization.
• Training scale: 47% said the AI training tools their vendor provides do not scale as required.

A firm that cannot hire data analysts is supposed to retrain auditors. A firm whose vendor training does not scale is then stuck buying a product it cannot teach. That is the second bill inside the 34% cost barrier: software, plus the review hours, plus a syllabus that half the buyers already doubt.

Frequently Asked Questions

How Many Professionals Did IDC Survey, and From Where?

IDC canvassed 1,005 senior audit and accounting decision-makers in December 2025. The United States accounted for 39.8% of the sample, the United Kingdom 15%, Canada 14.9%, Australia 10.1%, Germany 10%, and the Netherlands 10%. Titles included directors, vice presidents, executive and senior vice presidents, C-level officers, and partners.

What Share of Firms Plan to Expand AI Use in the Next Two Years?

Fifty-nine percent of the global sample said they plan to adopt or expand AI technologies within two years. The US figure was 53%. Australia was 50%. In the UK and Ireland only 42% said they plan to adopt within two years, a lower share that tracks a market already further into deployment, with 71% reporting AI in strategy, functions, or pilots against the 66% global average.

Did Australian Firms Name the Same Top Barrier as the Global Sample?

No. Australian respondents put a lack of technical talent first at 28%, then cost of implementation at 26%, then regulatory uncertainty at 23%. Globally those three were 34% cost, 30% talent, and 17% regulation. Australian respondents were also less likely to call algorithmic bias a moderate-to-extreme risk (67% against 79% globally) and less likely to say auditors should always validate AI outputs (51% against 64%).

How Did US Respondents Differ on Quality and Validation?

Sixty-three percent of US respondents agreed or strongly agreed that AI can enhance audit quality, ten points above the 53% global figure. Seventy-three percent said auditors should always validate AI outputs used in professional conclusions, nine points above the 64% global share. Thirty-eight percent of US respondents ranked data analysis as the top future skill, five points above the 33% global rank. The 36% who agreed or strongly agreed that AI could undermine professional judgment matched the global figure on that harder cut.

Do Audit Standards Require Human Checks of AI Output?

No AI-specific International Standard on Auditing requires a partner to validate every model output. The IAASB has said ISQM 1 and ISA 220 (Revised) still govern technology risk, and in December 2025 it approved non-authoritative material rather than a new standard. The 64% “always validate” finding is a survey result, not a legal duty. Firms that treat it as a duty are writing a review step into their own files while the rulebook catches up.

On September 9, 2026, Caseware unveiled Verity for Excel at CwX Germany, putting agentic steps inside the workpapers where reviewers already sit. The InfoBrief’s own heading said AI will not replace auditors and will redefine what they do. The survey’s third release left the training numbers attached to that claim, and they have not moved.