UNDERTOW 017: The Zero-Deductible Organization
AI just made one hundred percent coverage available for the first time. That is the problem.
Your car insurance has a deductible. Some first slice of every accident, a thousand dollars, maybe five hundred, that the company will not cover and that you would not want to pay them to cover. You have probably never thought of it as a designed object. It is one of the most carefully designed objects you own.
It exists for two reasons, and both of them matter here. The first is arithmetic. Insuring the last layer of a risk is a terrible trade. The premium for that final thousand dollars runs close to a thousand dollars, because small claims are frequent and costly to process, so you would be paying a dollar to protect a dollar. The second reason is about you. A driver with zero deductible becomes a slightly worse driver. Not on purpose. But if scratches cost nothing, you park closer. The industry has a name for this: moral hazard. Full coverage changes the person it covers.
So every sane policy on earth leaves a layer of the risk with you, on purpose. The industry has a name for that too. It is called retention. The risk you keep.
Hold that, because now we read your company’s policy.
Look at what your organization has built around its decisions and ask what it is. The pre-test before the campaign airs. The dashboard that settles the argument. The committee that approves the work. The model that now drafts, scores, and predicts, for pennies. None of this comes with a policy number, but it is all the same purchase. You are paying premiums, in money, in weeks, in process, so that no single decision can hurt you too much. And so that no single person has to answer for one. That second part is not a side effect. Ask anyone who has ever said “the research supported it” in a bad quarter. That sentence is a claim check. It pays out in blame not taken.
So your company insures its judgment. Fine. Insurance is a good product. This essay is not against insurance.
The problem is a question nobody in the building has asked: where is the deductible?
Judgment has a last layer, the same as a car. It is the part no test can settle. Whether the idea is actually good. Whether the work should run even though the numbers came back mediocre. Whether the safe version is the expensive one. Economists have understood since 1921 that the measurable part of any decision earns nothing, because whatever can be measured can be copied and priced by everyone else. The only part of judgment that was ever worth anything is the part with no number on it. The last layer is not a residue your process hasn’t cleaned up yet. It is the entire reason your company is paid.
The only part of judgment that was ever worth anything is the part with no number on it.
And here is the fact the whole essay rests on, so take it slowly. You cannot get rid of that layer. Retention is conserved. When you insure a decision, the uninsurable part does not shrink. It moves.
Watch where it goes. When a creative director says “we are running this, on my call,” the retained layer is small, visible, and has a name on it. If it goes wrong, you know what went wrong and who answers. When the same decision goes through the test, the dashboard, and the committee, the layer does not disappear into the machinery. It passes through the machinery and comes out the other side as something worse: risk held by everyone, owned by no one, watched by no one. It stops appearing in any report. Unwatched risk does not come due in installments. It comes due all at once.
You have seen this at civilizational scale. That is what 2008 was. Every institution had insured, hedged, sliced, and sold its risk, and each of them was, on paper, covered. And the firm at the center of the wreckage was, fittingly, an insurer. AIG had spent a decade selling coverage on the safest layer of the mortgage market, the tranche called super senior, the piece everyone agreed could never fail, hundreds of billions of it, with almost nothing held in reserve, because the models said that layer would never come due. But risk passed around is not destroyed. The system as a whole was holding all of it, with no name on any of it, and it came due on a single morning. That was not a failure of insurance. That was insurance working perfectly on the measurable layer while the retained layer collected, unnamed, in the dark.
The scale you live at is smaller and slower, which is what makes it hard to see. A brand that has been beige for a decade did not decide to be beige. It insured every decision, one at a time, each time correctly, and the retained layer drained out of every named person and pooled in the organization, where it compounded as sameness. Nothing in the coverage failed. What failed had no line item anywhere. The people who would have signed were not fired. They were covered, until signing was no one’s job, and then they left, and their replacements never expected to sign. Ten years on, the brand has never been safer. Cause of death: coverage.
Ten years on, the brand has never been safer. Cause of death: coverage.
Now the audit, and it takes about a minute. Take the last significant creative decision your organization made. Not the biggest. Just the last one that mattered. Write down what the coverage was: the test it passed, the data that supported it, the people who approved it. Now find the place where one person’s name is attached to the part no coverage reached. The part where someone said, in effect, the numbers do not settle this, and I say we do it, and I will answer for that.
For most organizations, that line is blank.
The blank line is the whole diagnosis. An insurance claim pays against a signature. Somebody underwrote it, and you can find out who. A retained risk with no signature is not retained by anyone. It is simply loose. Your organization is full of decisions like this: fully covered on every measurable surface, and blank exactly where the name should be. Nothing about it feels dangerous from the inside. It feels like governance. That feeling is what full coverage is designed to produce.
You will want to file this under bravery. Take more risks, trust your gut, fortune favors the bold. We know that speech, and this is not it. Courage is not the variable. You can be extremely cautious and still hold a named layer; a pilot is not brave for being the pilot. You can also be reckless and hold nothing, gambling with risk that belongs to everyone and answering for none of it, which describes half the boldness on LinkedIn. The variable is never how much risk you take. It is whose name is on the risk you already carry.
The most rule-bound industry on earth understands this. Commercial aviation has insured flying to within an inch of its life: checklists, procedures, redundancies, automation. And at the center of all that coverage, federal law keeps one human holding the last layer, by name. The regulation says that in an emergency, the pilot in command may deviate from any rule to the extent required. Not the airline. Not the manual. The named person in the left seat. The most insured system we have ever built made its deductible a matter of law, because it learned, expensively, that the last layer has to live somewhere, and the only safe place is a person everyone can see.
The Catholic Church learned the same thing, then unlearned it, and the unlearning is the best natural experiment on record. For about four centuries, every candidate for sainthood had to get past a formal officer whose whole job was doubt: the Promoter of the Faith, better known as the Devil’s Advocate. One named man, holding the Church’s retained layer, with real power to stop the process. He was, in the terms of this essay, the deductible. In 1983 the office was effectively abolished, part of a reform to make canonization faster and cheaper. It worked the way full coverage always works. The pope who retired the office went on to canonize 482 saints, more than every pope before him combined going back four centuries. And the meaning drained out at the same rate the volume rose. Try to name three of them.
And when organizations pretend to retain, the pretense measures itself. Google’s famous twenty percent time was, on paper, a retained layer: one day a week held back from the measurable work, protected for unprovable bets. In practice, employees called it 120 percent time, because the insured work still had to fit in the other hundred, and the exceptions got approved less and less. When another company measured its own version honestly, the retained fifth of the calendar came out to roughly one percent. A layer that is written down but not protected is not retention. It is retention’s paperwork.
The assigned devil’s advocate fails the same way. The skeptic who does not mean it entrenches the room instead of testing it. A name only counts if the person behind it can lose something.
If none of this is new, why is it urgent? Because the price of coverage just collapsed. Testing a campaign used to cost real money and real weeks. Testing a product cost more: you had to build at least some of the thing before anyone could tell you it was wrong, which meant someone had to decide what to build on judgment alone. That expense kept the insurance partial by default. There were always decisions too small or too fast to cover, and the retained layer survived in the gaps, the way weeds survive between paving stones.
The model ends that. It will score the campaign, simulate the user, run the focus group without the focus group, mock the flow and grade its own mock, instantly, for almost nothing. When premiums fall, coverage expands. That is not a moral failure. It is demand behaving normally. Every creative organization on earth is now expanding its coverage at once, gratefully, and squeezing its retained layer toward zero. Not because anyone decided to. Because nobody is pricing what the coverage removes.
Which brings us to the turn, and it is worth saying flatly. You believed coverage reduced your exposure. Run the arithmetic from the top. The uninsurable layer cannot shrink. It can only move from named people into the unnamed organization. So every increase in coverage increases the share of your risk that nobody holds, nobody watches, and nobody prices. There is a name for what your company is becoming, and you already know it as a perk, the upgrade every benefits brochure sells: zero deductible. The zero-deductible organization is the most thoroughly insured institution in the room and the one carrying the largest invisible exposure. Full coverage is not the opposite of the disaster. Measured in the only units that matter, full coverage is the disaster, prepaid.
Full coverage is not the opposite of the disaster. Measured in the only units that matter, full coverage is the disaster, prepaid.
And the AI model adds the part that makes this the present tense. Final judgment used to be uninsurable for a boring reason: no machine could reach it. Covering it was impossible at any price, so a deductible survived by default. Somewhere in every process, a human name stuck to the last layer because there was nothing else for it to stick to. That default just ended. For the first time, the last layer will take coverage. The machine will happily hold the opinion, make the call, own the taste, for pennies. Which means that for the first time, one hundred percent coverage is available. The zero-deductible organization used to be impossible to build. Now it builds itself. And one hundred percent is precisely the point at which the retained layer, still conserved, has no name left to sit under. The machine does not remove the risk you were in business to take. It removes the last name from it.
The machine does not remove the risk you were in business to take. It removes the last name from it.
That sentence closes too neatly to be the end, because the symmetry only holds if something else is true. A name on a risk is not a label. It is a stake. The signature works because the signer can be ruined a little: their standing, their next budget, their next job. That is what made the pilot’s authority real and the assigned devil’s advocate fake. So the blank line in your audit is not asking for a name to be written in, which any org chart can supply by Friday, and which would be coverage wearing a lanyard. It is asking a colder question: who in your organization is still positioned to lose anything by a judgment call? And what have you been doing, all this time, to everyone who was?
Your policy has been comprehensive for years. The premiums cleared. Every layer that could take a number took one. What is left is the layer that never could, the one you were founded to carry, sitting exactly where you left it. Unpriced. Unwatched. Unsigned.
The blank line is still blank. It is the only line that was ever yours.


