The Diagnostic Discipline

If you have ever left a client meeting reassuring someone that you run a rigorous process, then sat in the car afterward wondering what you are actually going to do next, this paper is for you. That gap, between the process you describe in the room and the improvisation that follows, is the whole subject of it.

I wrote it because I kept hearing the same thing. Consultants are being pressed to advise on AI, and the market has quietly conflated two different things: using AI to run an audit, and running a rigorous diagnostic. They sound identical in a sales conversation. They are not the same act. The most common way people close that gap right now is to drop a client's material into a general-purpose model and prompt it to do an audit. That returns one ungrounded answer. It cannot cite its sources, it compounds nothing across an engagement, and it falls apart the moment scope or scale grows.

What the paper argues

The core claim is that a rigorous diagnostic and an improvised one are different in kind, not in quality. Not a better version of the same thing. A different thing.

A real diagnostic is a structured discipline: a sequenced, gated, source-traceable process where each phase is a prerequisite for the next, provenance is required rather than optional, and an enforced mechanism, not a flowchart, is what makes the rigor real. A sharper prompt cannot manufacture any of that, because the failure it fixes is structural, not verbal. You cannot prompt your way out of a blindness to the whole.

The record backs this up, and it is blunt about AI specifically. More than 80 percent of AI projects fail, roughly twice the rate of non-AI IT work, and the leading root cause is not the model. It is people, problem-framing, and readiness (RAND, 2024). Adoption is not value: nearly nine in ten organizations use AI, but only 7 percent have fully scaled it and only 39 percent report enterprise EBIT impact (McKinsey, 2025). Activity is everywhere. Results are not. The gap between them is judgment, and judgment is the thing a consultant actually sells.

That leads to the flag I am willing to defend for five years: the claim that AI is going to kill consulting is backwards. AI drives the cost of activity toward zero and never touches the value of expertise. You were always buying the mind, not the work. When activity is nearly free, the scarce good becomes knowing which work is worth doing, and a structured diagnostic is exactly how you sell that.

What is inside

The paper is 34 pages, 17 exhibits, and a full works-cited list. It works whether or not you ever touch our software. Here is what it covers.

The five principles a rigorous diagnostic has to hold, each one something a one-shot prompt structurally cannot keep: provenance over speculation, reality beats intent, readiness as a gate rather than a footnote, specificity as a quality gate, and implementability as discipline.

The five-phase methodology, argued at the level of principle rather than product. Discovery, then readiness, then solution architecture, then financial modeling, then roadmap. The point is not that these are five tidy buckets. It is that they form a dependency chain: reorder any phase and you break the one before it. Recommending a solution before discovery is malpractice, the same way prescribing surgery before ordering a test is malpractice, whichever coat you are wearing. Every phase here borrows from a recognized tradition. What is new is the decision to run them in this one forced order, with a gate between each that the next phase cannot cross until the previous one has produced a real answer.

The frameworks the diagnostic leans on are the ones you already know, adapted rather than reproduced: a maturity-model-like staged readiness view to place a firm and name what is blocking it, a 10-20-70-like resource lens to show where investment should actually go, an impact-effort matrix to force real prioritization, and a phased assess-pilot-scale roadmap to sequence the work. The paper cites the traditions it stands on and claims none of them as invented.

The evidence

The thesis was not asserted from one seat. Over roughly a year my cofounder Jeremy and I ran more than 200 conversations across the whole market: traditional and mid-market consultancies moving into AI, dev shops pivoting to AI consulting, boutique AI-consulting founders, an active paying customer, and finance and ERP rigor voices. In the paper, people are attributed by role and firm type only, never by name. A theme counts as a finding only when independent voices raised it unprompted. Three cleared that bar, and they were not what I expected going in.

"The most common thing I heard wasn't 'here's my AI strategy.' It was 'I don't even know what to tell a client who asks what AI to use.'"

The surprise was that skilled consultants are not behind by choice. They are frozen by pace. They are excellent at their craft and focused on clients and delivery, and the hype cycle is simply not something they have been able to keep up with. That is not a knowledge gap. It is paralysis, and that paralysis is the whole market.

"They weren't cutting corners. They were building a structure out of 31 cloud skills and an Excel sheet and praying it was rigorous."

"The best compliment we get in a demo isn't 'wow.' It's 'that's exactly how I would have done it.'"

Who wrote it and why

I lived the problem before I tried to solve it. I ran the hand-rolled version myself, a sprawl of cloud skills and shared docs and a spreadsheet held together by hand, with no real confidence the outputs were correct. Then I interviewed more than 200 consultants and heard my own frustration repeated back to me across firm after firm. Jeremy and I spent about a year turning the answer into working software, and the hardest part was not the intelligence. It was the discipline: making the shortcuts a tired human takes on a Friday structurally impossible. This paper is the thinking underneath that year, written for the people who are living the same thing.