The Quality Collapse in AI Advisory - What It Means For You

The second half of career advice that few of us are getting straight: where AI fluency actually matters, and where it is theater. The most useful second half of career advice on AI does not come from the people selling certifications.

Of the $684 billion enterprises spent on AI in 2025, more than $547 billion produced no measurable business value. That is the RAND number, corroborated by MIT, Gartner, BCG, and S&P Global, all of whom have spent the last twelve months arriving at the same uncomfortable answer from different directions. Eighty percent of AI projects fail to deliver. Ninety five percent of generative AI pilots produce zero P&L impact. Forty two percent of companies scrapped at least one AI initiative in 2025, up from seventeen percent the year before. The average sunk cost on an abandoned enterprise initiative is $7.2 million.

These are not the numbers of a technology that does not work. They are the numbers of a technology being sold and implemented by people who do not know what they are doing.

That is the part of the story the consulting industry would prefer the reader skip past, because it has direct implications for who the next phase of AI advisory belongs to.

The pattern in the failure data

The most useful finding in the RAND research is not the headline failure rate. It is the breakdown. Of 140 enterprise AI implementations they analyzed in detail, only twenty three percent of failures came from model performance, data quality, or integration complexity. The other seventy seven percent came down to strategy, governance, and change management. Seventy three percent of failed projects had no agreed definition of success before the work began. Sixty one percent were approved on projected ROI that was never measured after launch.

In other words, the failures are not technical. They are executive.

This matters because the entire AI advisory market has been priced and staffed as if the opposite were true. The premium has been paid for technical fluency. The Big Four and MBB collectively poured over ten billion dollars into AI initiatives since 2023, hired aggressively in machine learning, and built practices around being able to talk credibly about transformer architectures and agent frameworks. They sold the technology, then discovered, at the client's expense, that the technology was the easiest part.

Why the big firms misfired

The CIO trade press has been unusually direct about this. Major pharmaceutical and healthcare enterprises have publicly ended multi million dollar engagements after concluding that their internal teams were better positioned than consultants who appeared to be learning on the client's dime. The line that has been quoted most often, attributed to a CIO interviewed in late 2025, is the one to remember: "We are not going to keep paying $500K for a report that we suspect was generated largely by a machine."

The structural problem is straightforward. The big firms staffed AI engagements the way they staff everything else: a senior partner who sells, a small team of senior managers who orchestrate, and a layer of bright twenty eight year olds who do the work. In a domain where the questions are mostly about how to govern, sequence, and operationalize new technology inside a complex organization, the pyramid is the wrong shape. The senior people who could answer those questions are not in the room enough, and the people who are in the room have neither the operating scars nor the standing to challenge the client's bad assumptions.

The boutique alternative that has emerged is not really a firm. It is a single senior operator, charging $700 to $1,500 an hour with project floors of $50K to $250K, sitting in the room personally, and selling exactly what the pyramid cannot offer: judgment that has been earned somewhere it counted. The market has begun to price this. It has not yet figured out how rare the qualified supply actually is.

The qualification gap

Here is what is actually happening in the supply side of AI advisory in 2026. A large fraction of the people now selling enterprise AI strategy learned the vocabulary in 2023 or 2024. They are fluent in the current model lineup, comfortable with the agentic frameworks, and capable of running a workshop that produces a roadmap. What they do not have is twenty years of watching enterprises try to absorb new technology and fail. They have never sat in a steering committee meeting where a CFO killed a project that should have lived, or watched a perfectly good system die because nobody owned the change management. They are selling the part of the work that is easiest to learn, and the failure data suggests it is the part that matters least.

The reader of this newsletter has the opposite problem. The pattern recognition is there. Three or four decades of it, in many cases. The technical vocabulary is the part that needs to be acquired, and the acquisition curve for someone who already understands data, systems, governance, and the politics of enterprise change is a great deal shorter than the credentialed AI native would like to believe.

The market has not priced this in yet. The reason it has not is that the buyers are still recovering from the last cycle, the failure data is only six to twelve months old in its current form, and the boards now demanding AI ROI (ninety eight percent of directors, by recent CIO research) have not yet rebuilt their procurement reflexes around what actually works. They will. The window in which a senior operator with thirty years of credibility and twelve months of serious AI study can charge what an MBB partner charges, while delivering a product the MBB partner cannot, is open now and will not be open in three years.

What this means for the reader's next move

The temptation, for any reader of this newsletter who has spent the week thinking about AI positioning, is to update the LinkedIn headline and start describing themselves as an AI advisor. That move is wrong, and it is wrong in a specific way.

Generic credibility loses to specific case studies, even at this seniority level. A bio that says "thirty five years in enterprise data, now advising on AI strategy" reads to a sophisticated buyer the same way a thirty year old's bio with three AI certifications reads. Both are gestures. Neither is evidence.

The work, before the repositioning, is to be able to point to two or three specific engagements where you have shipped, governed, or salvaged something. Not theoretical exposure. Not "led the AI strategy conversation." A system you put into production, a governance framework you implemented and that survived contact with a real organization, a failed initiative you were brought in to diagnose and that you actually diagnosed correctly. If the case studies do not exist yet, the next twelve months are about manufacturing them, including, if necessary, taking one engagement at a discount in exchange for the right to write about the outcome.

The qualification gap is real. It will not stay open forever. The operators who walk through it will be the ones who can prove, in specifics, that they belong on the other side.