KPMG has withdrawn a report on agentic AI after organisations named in the document disputed claims about their use of the technology, raising fresh questions about evidence standards in professional services research.
The report, titled Redefining excellence in the age of agentic AI, was removed after questions were raised over exaggerated adoption stories, inaccurate case studies, and apparent AI hallucinations. The withdrawn report had included claims about organisations including UBS, the NHS, Swiss Federal Railways, and Transport for London.
The withdrawal followed scrutiny of the report’s citations and case studies, with several named organisations rejecting the way their activity had been described. The case is especially sensitive because the report concerned the benefits and adoption of AI.
Professional services businesses rely heavily on reports, white papers, benchmarking studies, and market insight pieces to demonstrate expertise, influence clients, support sales, and shape boardroom conversations. If underlying evidence fails, reputational damage can exceed the value of the material itself.
The issue is not whether AI tools can support research and drafting. They can. The governance problem is whether organisations have controls to verify claims, check citations, confirm case studies, obtain approval from named third parties, and distinguish between plausible output and evidenced fact.
AI governance has already entered boardroom planning. In directors put AI governance on agenda, Institute of Directors polling showed companies beginning to formalise oversight of AI use. The KPMG withdrawal shows how the same discipline applies to knowledge work and external communications, not only customer-facing automation or operational systems.
AI-generated errors can be difficult to detect because they are often fluent, specific, and formatted like legitimate research. Hallucinated citations, misattributed claims, and synthetic case studies may appear credible unless every source is checked, named organisations are contacted, and claims are compared with primary evidence.
That creates a different quality control burden from conventional editorial review. A standard review process may catch tone, grammar, argument structure, and legal risk, but not necessarily fabricated sources or unsupported case studies. Organisations using AI in research workflows need explicit checks: source verification, quote approval, case study confirmation, third-party sign-off, version control, and named human accountability for final publication.
The case may also influence wider AI adoption. Executives are being encouraged to use AI to improve productivity, accelerate analysis, and reduce cost. Cases involving false citations or invented claims give cautious leaders more reason to slow deployment, especially in regulated or high-trust sectors.
Advisory businesses face a particular conflict. They are selling AI transformation services while adopting AI internally. Clients may judge their credibility not only by what they advise, but by how rigorously they control their own use of the tools.
The market is already putting more weight on assurance. Boards want to know where AI is being used, what data it touches, who reviews outputs, what audit trails exist, and how errors are escalated. A published report is only one use case, but it is a visible one because errors become public and traceable.
Procurement teams may also respond. Clients commissioning research, consulting, or thought leadership may ask suppliers to disclose whether AI was used, how sources were checked, and whether named case studies were independently verified. That could become a standard part of professional services assurance, particularly where reports influence investment, risk, policy, or transformation decisions.
The KPMG episode points to a simple operating principle: AI can speed up knowledge work, but it cannot weaken evidential standards. In professional services, the output is only as strong as the verification behind it.
You must be logged in to post a comment.