UNDERTOW 016: How to De-Risk Creativity
Your process didn't strangle the idea. It made sure no one had to have it.
Picture the most comprehensive insurance policy ever written. It covers every disaster that could befall your company, in full: the fire, the flood, the lawsuit, the downturn. There is one exclusion, buried on a back page, easy to miss. It is the single risk you are actually in business to take. Nobody would knowingly buy that policy. You already did, at considerable expense, and you are proud of it. It is called creative infrastructure.
The logic that builds it is sound and nearly universal. Companies are good at generating ideas and bad at converting them into anything, and that gap is the most discussed problem in the business of creativity. The fix is to make the unreliable input reliable, the unrepeatable process repeatable. To de-risk it. You install the forums, the gated funding, the idea markets, the innovation function, the seat at the table.
You have probably built some of this. You may be about to build more. This is written for you, in the voice of the thing you would normally read about this, and it will agree with you for exactly as long as it takes to show you what you bought.
Get the wrong reading out of the way first. You will want to file this under process strangles creativity. Bureaucracy bad, art good, you cannot put a gate in front of a muse. That is not the argument, and it is a weaker one than it sounds. Process is not the enemy. The andon cord is process. The problem is not that you systematized. The problem is what you chose to insure.
In 1921 an economist named Frank Knight drew a line the entire conversation about creativity has spent a century avoiding. He split the unknown into two kinds. The first he called risk. Risk is the unknown you can measure: you can put a number on it, and once you can put a number on it you can price it, hedge it, and insure it. The second he called uncertainty. Uncertainty is the unknown you cannot measure: no number, no odds, nothing to price, so it cannot be insured or handed to anyone. It can only be carried, by a person, making a call.
Then Knight closed the trap. In an open market, the returns to mere risk get competed away to nothing, because anything you can measure your competitor can measure too, and price, and match. The measurable is by definition the part everyone can already do. So profit, real profit, the return above the cost of capital, is never the reward for taking risk. It is the reward for bearing uncertainty. The part with the number pays nothing. The only part that pays is the part with no number, the part a human has to stand behind without knowing.
Hold that against the thing you built.
A creative judgment has two parts, and they sit on Knight’s line exactly. There is the part you can write down: the brief template, the scoring rubric, the stage gate, the criteria, the best practice. That part is risk. It is legible, so it is insurable, so an apparatus can absorb it, and so, in Knight’s terms, it was already worth nothing, because legible means everyone has it.
Then there is the other part: the unpopular call, the taste, the willingness to hold a standard when the quarter is bad and the room disagrees and there is no number that proves you right. That part is uncertainty. It cannot be made legible, which means it cannot be transferred, which means it can only be borne by an exposed person. It is also, by Knight’s law, the only part that was ever going to pay.
The part with the number pays nothing.
Now look at what the apparatus does. You built it to take the risk out of the judgment. It worked. It absorbed the legible, insurable, worthless part with real efficiency. And it could not touch the other part, because the other part is uninsurable by nature, so the other part went undone, and no one bore it, and the move is this: the apparatus is sitting exactly where the person who should have borne it would be standing. The forum met. The gate cleared. The function signed off. The seat at the table was occupied. Everything looks handled. The one move that was the entire source of value did not happen, and the machinery is positioned so that you cannot see it didn’t.
Call it nerve laundering. You had a judgment that needed someone personally exposed, and you ran it through enough apparatus that no individual is on the hook for having made it, the way you run money through enough accounts that no one is on the hook for where it came from. The output looks clean. It is clean. It is also empty, in the one place that mattered.
You do not have to imagine what this looks like at full scale, because an entire industry already ran the experiment. A credit rating is that policy in its purest form: a machine for taking an uninsurable judgment, will this borrower pay, will this pool of mortgages hold, and turning it into a letter you can hedge against. The whole apparatus exists so that no individual has to stand behind the call. In 2006 Moody’s stamped thirty mortgage securities triple-A every working day, and eighty-three percent of that year’s batch were later downgraded. The model handled the legible half beautifully. The other half, the exposed human saying this is garbage and I will not put our name on it, was the half the machine was built to remove.
We know it was removed because the people inside watched it go. In an instant-message exchange that later surfaced, two S&P analysts put it to each other plainly: the deal was ridiculous, the model did not capture half the risk, they should not be rating it. Then one wrote the line that is the whole essay: we rate every deal, it could be structured by cows and we would rate it. The chair where the exposed call should have sat was occupied by a process, the process produced a triple-A, and the triple-A looked exactly like a hundred years of diligence right up until the morning it did not.
The output looks clean. It is clean. It is also empty, in the one place that mattered.
This is where the anti-process reading we threw out gets it backwards. The word infrastructure covers two opposite machines. Toyota’s andon cord is a rope any worker can pull to flag a defect and, if it cannot be cleared in time, stop the line. It is pure infrastructure, and it does the opposite of absorbing the call: it puts one named human at the point of the uninsurable decision and obligates them to make it, on their own judgment, in front of everyone, with the cost of stopping a production line riding on the pull. It is built to manufacture exposed people, not to spare them.
American carmakers copied the rope and installed it faithfully, and their workers almost never pulled it, because pulling it meant being the person who held up the line and caught the blame. By one account a single Toyota plant pulled the cord thousands of times a week; the American equivalent, a handful. Same rope, opposite outcome. The variable was never the infrastructure. It was whether the company protected the human who bore the uncertainty.
So the question that separates the two machines is one question, and you can run it on anything you have built. After the apparatus is in place, is there still a specific person exposed at the point of the call no number can settle? If yes, you built an andon cord. If no, you built a laundry.
None of this is the familiar complaint that responsibility goes diffuse in big systems. That is true, and it is about decisions that get made and then cannot be pinned on anyone. This is the stranger thing one step earlier: the decision that does not get made at all, and leaves no visible gap, because the apparatus is shaped like the decision. And it is not quite the old story that rigid systems crush local know-how, because that story stops short of the part Knight supplies. The know-how the system flattened was not a nice-to-have. It was where the profit lived. The system did not just sand off the texture. It insured away the earnings.
If yes, you built an andon cord. If no, you built a laundry.
It is also not what Nassim Taleb named skin in the game, though that is the idea standing nearest. Skin in the game is about who eats the loss after a call is made: the builder made to live in the house he built, the trader whose own money rides on the position. This is one step earlier, at a place that idea does not reach, the uninsurable call itself, made before there is any downside to put a name against. You can have perfect skin in the game on every measurable decision and still never make the one with no measure, because exposure to a loss cannot tell you what to do when there is no loss yet to price.
You can watch both machines at the two ends of the business that sells creativity for a living. Apple’s “1984” was screened for the company’s board in December 1983, and they hated it. Most of them, the CEO later said, felt it was the worst commercial they had ever seen, and one director wanted the agency fired. The apparatus returned one verdict: kill it. Apple ordered the agency to sell off the Super Bowl airtime it had already bought. The agency refused. It sold the thirty-second slot, said it could not unload the sixty in time, and let the spot run, so the most famous ad ever made aired only because the people who made it disobeyed the client to keep the machine from killing their work. The board’s no was the insurable half, the safe read no one could be blamed for. The call a few exposed people made with their necks out, against the room, was the only thing that paid.
Then the industry spent thirty years building the machine that would have strangled that ad in the room: copy-testing, effectiveness scores, the procurement auction that buys creative by the pound, all of it tuned to the half you can measure. By its own effectiveness data, award-winning work is now no more effective than the forgettable kind, the output has converged into a sea of sameness, and the holding companies that mastered the measurable have watched their margins thin and their relevance drain, one of them dropping out of the FTSE 100 in 2025. They insured the half a client could buy from anyone, and now a model makes it for free. The apparatus won, and it commoditized the people who built it.
The apparatus won, and it commoditized the people who built it.
Publishing is the same mechanism, the exposed call ducked in plain sight. Houghton Mifflin had a French cookbook under contract for years, sent the authors back for cuts, and finally let it go: too long, too detailed, too formidable to the American housewife, who, an editor explained, did not want an encyclopedia, she wanted to cook something quick with a mix. That was the apparatus returning its legible verdict on what an American cook would buy. The manuscript landed on the desk of Judith Jones, a junior editor at Knopf and the person who had pulled Anne Frank’s diary out of a reject pile. Instead of judging it from the page, she cooked it, scouting the ingredients and working the recipes in her own kitchen to see whether they held. They held. Knopf still had low expectations and spent almost nothing to promote it. She put her name behind it anyway, and Mastering the Art of French Cooking reshaped how a country ate. The publisher’s verdict was the insurable half, the safe read on the cook who already existed. The bet a person made by cooking the food and standing behind it was the only part that paid. A process can walk you to the most defensible version of what already exists, the quick cookbook with the mix. It cannot bet on the thing that is not there yet, because there is nothing there yet to measure.
Now the part that is hard to hear, so we will say it flatly. None of this is a malfunction. The machine is working perfectly. It is doing precisely what it was built to do, which is to make sure no person ever has to stand exposed behind an unprovable call. From outside it does not look like cowardice or decline. It looks like maturity. It looks like governance, rigor, process discipline, grown-up institutional behavior, the very things you are praised for installing. An organization can insure itself, comprehensively and at great expense, against the only thing that would have paid, and the more complete the apparatus, the more it reads as an achievement, and the less anyone can tell. That is the comedy of it. The better you do this, the more impossible it becomes to notice you have done it.
And you did not learn this at work. Work only gave you the budget for it.
The same policy covers the rest of your life, and the premiums come out monthly. There is a conversation you are managing instead of having, because having it means saying the true thing and standing there while it lands, with no process to point to and no one to share the blame. There is work you could make in your own voice, with your name on it and nothing behind it, and instead you make the version that will test well. There is a person you could tell, first, before you have any proof of what they will say back. Each is a call no number can settle, which is the only kind that was ever going to pay, and each you have run through enough asking around and waiting for the room that the decision belongs to no one and the failure to make it cannot be pinned on you.
From the outside it passes for wisdom. It has respectable names. It looks like being reasonable, like a person who has their life together. And what it builds is a life fully insured and defensible at every point, signed off by everyone you consulted. Every seat filled but one. The one that was always going to be yours.
And there is one more turn, because you have been reading this the way you read these things. Somewhere in the last few minutes a part of you started converting it into something usable. A principle. A slide. A diagnostic you could run on your org, a framework you could brief to your team, a thing to apply. That impulse, right there, the reach to take the uninsurable point and make it legible, transferable, repeatable, is the exact move the essay is about. It is the laundering, happening again, now, to this. The piece you are reading is the same kind of artifact as the ones it describes, made in the same voice, and the proof that the mechanism has you is that you want to operationalize it.
Every seat filled but one. The one that was always going to be yours.
And the entire apparatus you are so proud of building was designed, expensively and with the best intentions, to ensure that the someone is never you.
That is the coverage. Everything is insured. You are paid for the exclusion.
What this leaves you holding
You will want a checklist. Resist it. The checklist is the laundry. There is exactly one thing to do with this, and it is not something you install.
Find the call in your organization that currently belongs to no one. Not the decision that gets made and then can’t be pinned on anyone after, the one that never gets made at all, because the forum and the gate and the function are arranged in the shape of it and everyone reads the shape as the decision. It is the brief that tests fine and means nothing. The hire everyone has an opinion about and no one will own. The standard you keep aligning on instead of holding. You already know which one it is. The discomfort you felt at a specific sentence above was the address.
Then do the one thing the apparatus was built to prevent. Put a name on it. Yours, or one person you will protect when they pull the cord and hold up the line. Not a committee, not a RACI, not a new function, because those are more apparatus, which is more laundry. A named human, exposed, at the point of the call no number can settle, with cover from the blame that exposure invites. That is the whole andon cord. Everything else you might build this quarter is the other machine.
The stakes are not abstract. The legible half of your craft, the half your apparatus got so good at, is the half a model now does for nothing. Your remaining margin is the uninsurable call and the people willing to bear it. Insure that away and you have, precisely, insured away your earnings.
So the brief is one line. This week, find the call no one owns, and give it a name before the room weighs in.



this:
It was whether the company protected the human who bore the uncertainty.