The method you never wrote down

Why this matters to you now

Three surgeons were filmed teaching a cricothyrotomy, then put through a cognitive task analysis on the same procedure. Maura Sullivan, Kenneth Yates, Kenji Inaba, Lydia Lam and Richard Clark published the result in Academic Medicine. Measured against a reference list built with three other surgeons, the experts' teaching left out 71% of the clinical knowledge steps and 51% of the action steps. The number that matters to you is the third one: 73% of the decision steps. And across the twenty-seven action steps, the how-to was described in only 13% of cases.

These are not careless people. They are experts teaching on purpose, on camera, knowing they are being recorded. The explanation the authors give is automation: a skill that has become automatic runs outside conscious awareness and faster than it, so it is not available for introspection. You cannot recall a decision you no longer really make.

A second number in the same paper tells you what to do with the first. In free recall, the experts described 44% of the steps. Under structured questioning, 66%. Being asked questions closes nearly half the gap. Nothing closes it all the way.

Weigh the sample before you believe me: three experts, one surgical procedure, no consultants anywhere in it. What transfers is not the percentage, it is the mechanism: the automation of a skill is a fact about learning, not about surgery.

Edition #33 on the client who arrives with an answer ended on what sounded like good news: what you have left to sell is your judgment. Here is the question it left open. Where is it written?

Your method is not unwritten because you are busy. It is unwritten because the part that matters does not come out through introspection.

Pricing: the price of the tenth rule

The move: Write the two or three decision rules your clients argue with you about, and put your price there. Keep your procedure for your collaborators: it gets work executed, it does not get work sold.

Why it works: In What's Your Strategy for Managing Knowledge?, Morten Hansen, Nitin Nohria and Thomas Tierney describe two knowledge strategies, each carrying its own economics. Reuse: invest once in a knowledge asset and reuse it many times, with large teams, a high ratio of juniors, and total revenue as the goal. Expertise: charge high fees for highly customized solutions to unique problems, with small teams, and margin as the goal. The authors measured the gap between the two. In 1997, daily fees for a McKinsey consultant averaged more than $2,000 while an Andersen Consulting consultant came in a little above $600, and the ratio of consultants to partners ran around seven at both McKinsey and Bain. Those are U.S. dollars from almost thirty years ago and the amounts mean nothing today. The ratio, a little over three to one, still frames the question.

Hold on to the exact definition of the expensive column: high fees for a unique problem. A unique problem is precisely the one for which no rule has been written yet. So it is not your documentation that holds your price up, it is your ability to settle the situation your documentation does not yet cover.

The trap: Writing your procedure while believing you are writing your method. A written procedure switches your column, the wrong way: you inherit reuse economics without the volume, the teams or the juniors that make it pay. The real asymmetry sits elsewhere, and it is sturdier than any argument about secrecy. Your procedure can be rebuilt without you. Your tenth rule cannot: you have to have seen the nine situations that came before. What you sell is not the document, it is the fact that you can extend it. Edition #6 said your methodology is your product, and it also told you to name it and present it as a proprietary framework in your proposals. That half has not aged well. What happened in nine months is that a named sequence in a proposal became the part the committee can compare. What holds from #6, and I stand by it, is its step 4: write the decision rules. One word was missing, and the word is "because."

This week: Your last three engagements. In at least one of the three, find a decision where you went against what the data suggested. Write it as a single sentence in this shape: when this shows up, I do that, because of this. If you only find one, one is enough. That is your first asset.

Sales and business development: sell the rule, not the steps

The move: In your next proposal, replace the phase diagram with two or three decision rules, spelled out. Name what you would do differently from a competent internal team with good tools.

Why now: In Teaching Smart People How to Learn, Chris Argyris argues that the people in an organization everyone assumes are best at learning are in fact not very good at it, and he says it of those people in particular: the well-educated, high-commitment professionals in leadership roles. His explanation is defensive reasoning. When you are rarely made to fail, you have never had to account for the reasoning behind your own successes, so a question about that reasoning lands as an attack rather than a request.

A buying committee asking how you will decide is asking for exactly what you have never examined. You hear suspicion. It is a substantive question, and it is the only one left, because the procurement grid asking for your "methodology" now invites the answer that is worth nothing.

The trap: Answering with the steps. As of this year, the steps are precisely the part the committee can compare against what its own team produced last week with its own tools. You are choosing to be evaluated on the one ground where you no longer hold an advantage. The methodology section of a proposal is, to my mind, the most expensive paragraph in the document: it occupies the page where the reason to choose you should be.

This week: Open your last proposal. Find the section describing your approach, count its sentences, then count the ones that would survive if the client's team had already done the steps. If fewer than half survive, you have your rewrite.

Collaboration networks: get yourself interviewed

The move: Find a peer who does not serve your clients. Two hours each, taking turns. They interview you about ONE past engagement, specific and dated. Not about your method. They do not need to know your trade: none of the three questions in the format requires any, and that is exactly why they work.

The mechanism: In 1989, Gary Klein, Roberta Calderwood and Donald MacGregor published the critical decision method, a protocol built to pull knowledge out of a practitioner. The principle is a reversal: you never ask someone to describe what they do in general, you walk them through one specific incident and probe each turning point on its own. The protocol exists because we already knew that a practitioner asked to describe their method does not describe it. The 2014 measurement put a number on it twenty-five years later: 44% in free recall, 66% under questioning, among surgeons. For you, the size of the gap is unknown and its direction is not. And nobody can be their own interviewer, because the omission is invisible from the inside.

The format: Three questions are enough, looped over the same engagement. Where did you hesitate? What settled it? What would a good consultant in your field have done in your place? The third one does the most work: it makes you name a rule while describing someone else. One two-hour session on a single engagement yields one rule, two if the engagement went badly. That is enough to know whether the format works for you.

The trap: The conversation about "our methods." Two consultants comparing methods produce two tidy procedures and zero rules, because each one describes the presentable version of their work. Edition #6 already proposed this exchange between peers, and its format was wrong on the one point that decides everything: it had you describe the typical engagement. An interview about the typical produces a procedure, because the typical has no moments of hesitation. You need one engagement, a single one, with its date. Edition #27 made codification the precondition for cloning, and it was right about that; it did not say how to codify what you cannot introspect, and that is where its plan stops for most people.

This week: Pick the engagement you want to be questioned about, and write down the dates of the three moments where you hesitated. That is all your peer needs to know, and it is half the work.

Value creation: the deliverable that carries the rule

The move: Put the rules IN the deliverable, named and visible, instead of letting them operate underneath the recommendation.

Why it changes everything: Edition #30 already asked you to name what your deliverable ruled out. The middle column was missing: why. Without it, a rejected option reads as a preference. In The Reflective Practitioner, Donald Schön starts from an observation his publisher states plainly: the best professionals know more than they can put into words. They lean less on the formulas learned in school than on improvisation acquired in practice, and that unarticulated process is what he calls reflection-in-action. Follow the commercial consequence. What you bring your client that is worth the most is, by its nature, the thing they cannot see. So they price what they can see: pages, days, meetings.

The test: If a reader of your report cannot tell which recommendation came from a rule and which came from the data, they will conclude the data did the work. And the data is theirs now. A rule applied without being named is the elegant version of burying your value under volume. I would be surprised if the best-paid judgment calls of your past year appeared anywhere in the documents you handed over.

This week: One more page in your next deliverable, titled "the three judgment calls." For each one: the decision taken, the rule applied, and what the rule ruled out. Write what the rejected option would have cost, in one line: an option named without its price becomes a request for free extra work. Do not justify the rule in the document, state it. The justification is the conversation you want to start.

AI: the interviewer, not the writer

The move: Use a model to interrogate you about your past engagements. Never to write your method. Five uses.

Before the five: an engagement file belongs to your client. Work from closed engagements, strip the names of people, companies and places. What you are after is your reasoning, not their file.

1. The incident interview. Describe a finished engagement, out of order, however it comes back to you. Then ask to be questioned about the moments of hesitation only, one question at a time, and explicitly forbid questions about your process. You get a transcript where your rules are already phrased, in your own words.

2. The missing-step detector. Submit your written procedure, the one you hand to your collaborators, together with a closed file where you applied it. Ask what the file shows you doing that the procedure never mentions. The gap is the list of your unwritten rules.

3. The rule test. Give it one rule in a single sentence and three past files. Ask which of the three would have come out wrong under that rule. If it fails in none of them, it is not a rule, it is a platitude, and a platitude does not sell.

4. The committee's advocate. Give it the rule and the makeup of the committee that will hear it. Ask for the objections, not the arguments. You will learn which of your rules does not survive a simple question before it gets asked by the only person in the room who could sign.

5. The cutter. Take the transcript from use 1 and ask for the list of sentences that already have the shape "when, then, because," without rewording a single one. You get your rules in your own words, and you see at the same time how many of your hesitations never produced one.

The warning: Never take a model's explanation for the reason behind its answer. And do not ask it to write yours. In Language Models Don't Always Say What They Think, Miles Turpin, Julian Michael, Ethan Perez and Samuel Bowman showed that a step-by-step explanation can systematically misrepresent the real reason behind an answer. By adding a bias to the input, for instance reordering the choices so the right answer is always the first one, the authors changed the answers while the models never mentioned that bias in their explanations. Pushed toward a wrong answer, they rationalized it.

The bias dropped accuracy by as much as 36% across thirteen BIG-Bench Hard tasks. The measurements are on GPT-3.5 and Claude 1.0, 2023 models. What transfers is the mechanism, and the authors name it exactly right: plausible yet misleading.

Recognize the kinship, and no more: you omit without knowing it, the model invents without knowing it. In both cases, the account of the decision is not the decision. Here is a prediction, mine and not the authors': a model you ask to write your method will hand back something fluent, credible, organized into phases, and containing none of your rules. You will read it over and find yourself well organized.

This week: Use 1, on a single engagement. Thirty minutes. You are looking for one sentence, not a document. You will know it worked when one of the questions makes you answer "it depends," because that is exactly where your rule lives.