The AI Implementation Report is a fortnightly newsletter for leaders navigating AI strategy. Each issue highlights the research, data and trends that matter most for implementation, not the hype, not the doom, just what's actually working and what isn't. Curated by Mark Phelps and Drew Horton. A Mark Phelps LLC production.
Last issue we looked at AI sprawl, everyone getting the tools and nobody redesigning the coordination layer. This one is about what happens next, once the ungoverned output starts flowing back into the organization.
THE ONE TO READ
If you only read one AI implementation article before your next leadership meeting, make it this one.
KPMG published a report where its own AI contradicted its own published research. Nobody noticed until an outside firm went looking.
KPMG's AI Report Becomes an Accidental Demo of AI Hallucinations — Carly Page, The Register (Jun 12, 2026) | FREE
KPMG published a report on agentic AI last October. Somewhere in it was a claim that 55% of CEOs rank AI as their top investment priority. KPMG's own CEO Outlook survey, released the same month, put that number at 71%.
The AI made up a statistic that contradicted the firm's own research. Nobody caught it before publication.
The citation problem is worse in volume. GPTZero's forensic review alleges that only five of the report's 45 citations correctly pointed to their sources. The rest were mangled, misleading, partially fabricated, or too vague to check. GPTZero coined a term for it, "vibe citing," which is the citation equivalent of vibe coding. KPMG has since pulled the report and says it's reviewing how it got published.
Easy to enjoy this one at KPMG's expense. Deloitte had to refund the Australian government last year after AI-generated content turned up in a taxpayer-funded report, so the consulting industry has form here. But the schadenfreude wears off quickly once you ask the obvious follow-up, which is whether anything remotely like this is happening in your own organization, minus the outside firm doing a forensic review and the trade press writing it up.
The mechanism has a name. Matthias Holweg at Oxford and Thomas Davenport at Babson call it knowledge decay, and their recent HBR piece (linked below, though it's behind their wall) explains why it spreads. Their key insight is the thought running through the head of the person who stops verifying: if AI is reading what I'm sending, I'll just use AI to create it. That person isn't being lazy. They're being rational, because what they received was probably unverified too, and whoever gets their output will likely run it through a model anyway. The quality-control step doesn't get skipped once. It gets skipped all the way down the chain.
They break the cost into three parts, and the third one is the one nobody wants to hear. Verification, where the labor of checking AI output often cancels out the productivity gain that justified the tool. Validation, where clients stop paying premium fees for work they can't confirm a human produced. And entropy, which is a property of the architecture rather than a discipline problem. Content degrades with every pass through a model, and no amount of process rigor fixes that with the technology we have now.
We think the useful takeaway is that KPMG almost certainly has AI policies. Big Four firms have policies about everything. What they didn't have was anyone verifying the output before it became someone else's input. Policies don't catch this. Process design does.
LEADERSHIP LOAD OUT: BRING THIS TO THE MEETING
Ask your team one question: who verifies the AI output before it becomes someone else's input? If the answer is "nobody, really," you have a knowledge decay problem. You just haven't priced it yet.
ALSO ON OUR RADAR
The mechanism behind the failure, why it happens at the team level, and what doing it properly looks like.
Don't Let AI Slop Muck Up Your Company's Processes — Matthias Holweg (Oxford) and Thomas Davenport (Babson), Harvard Business Review (Jun 16, 2026) — The full academic treatment of knowledge decay, and the source of the framework above. They close by warning that unmanaged AI proliferation will produce a rerun of the productivity paradox, which is where this newsletter started in Issue #1. Two researchers arriving at our thesis from the opposite direction. | PAYWALLED*
Thoughtful AI Implementation for UXR Leaders — UX Collective / Ashlee Edwards (May 1, 2026) — AI can make UX research faster, but the term "ResearchSlop" has entered the lexicon because too many teams aren't checking whether the outputs are actually worth anything. | FREE w/ SIGNUP*
Designing With Uncertainty: How AI Supercharges Probabilistic Thinking — Smashing Magazine (Jun 2026) — AI doesn't give you the answer. It gives you a high-confidence starting point in a field of probable ones. Designers who understand that are using AI effectively to build better products. | FREE
*We try to only link to free, no-obligation reads. Two exceptions this issue. The HBR piece sits behind their subscriber wall and the UX Collective piece asks for a signup. Both are good enough that you should know they exist, and we've written the commentary above so you get the substance either way.
PARTING THOUGHT
KPMG's report had a citation problem, and somebody wrote an article about it. Your organization probably has the same problem, spread across a few hundred documents nobody is ever going to audit. The difference is that nobody's writing an article about yours.
Know someone navigating AI implementation? Forward this to them.
Hit reply and tell me what you think. I read every response.
Published for AI implementation leaders in the US and UK.
Mark Phelps is the founder of Mark Phelps LLC, an AI implementation consulting practice and Fractional VP of Product Design service based in the Boston metro area, serving product, design and technology leaders in the US and UK. He brings more than 20 years of Fortune 100 experience at GE, Fidelity, TIAA, Bank of America and Marsh McLennan, and for the last three years has held fractional VP roles at Rockefeller Capital Management and See All AI. He holds a Stanford AI Certificate earned with Distinction and an MBA in Digital Strategy from Suffolk University (Beta Gamma Sigma). He is a named inventor on a pending design patent for an AI-powered intraoperative 3D surgical navigation UI and a United States Marine Corps veteran.
20 Portsmouth Ave., Suite One #161, Stratham, NH 03885