Inversion
Stop asking how to succeed. Ask what would guarantee failure, then refuse to do it.
The Model
Forward reasoning asks what would make this work. Inversion asks what would make it fail, and then treats the avoidance of those things as the plan.
It is not a mood. It is a direction of travel.
The reason it works is arithmetic. Success has a thousand causes and most of them are unknowable in advance. Ruin has maybe six, they are usually visible, and they repeat across industries.
So the inverted question has a smaller and more honest answer set. You cannot reliably list what will make a company great.
You can list what ends it: running out of cash, one customer holding the revenue, a founder who stops sleeping, a product nobody misses, a legal exposure taken to save a fortnight.
Jacobi was talking about hard problems in mathematics, where the direct proof resists and the reversed one opens. The transfer to decisions is exact.
You are not looking for the best path. You are removing the paths that end.
Why operators get this wrong
Operators treat inversion as an exercise instead of a constraint.
It gets run once, at an offsite, and produces a slide titled Risks. The slide has six bullets. Nothing in the operating rhythm changes, and eight months later one of those six bullets happens exactly as written.
That is the common failure and it is not a thinking failure. It is a conversion failure — nobody turned the finding into a rule that a calendar enforces.
The second mistake is inverting the wrong altitude. "How do we not fail" is too high to act on. "What would make this launch convert at zero" is specific enough to produce a decision: the price is above what the segment can approve without a second signature, the page loads in six seconds on mobile, the offer requires a behaviour change nobody has time for.
The third is treating the exercise as pessimism and quietly softening it. If your list of failure modes contains nothing you are currently doing, you did not run the exercise. You performed it.
Applied
Body
Do not design the ideal programme. List what guarantees you stop training within twelve weeks.
The honest list is boring: a session that needs ninety minutes, a gym twenty-five minutes away, a schedule that collides with the school run, an ambitious loading scheme that aggravates the shoulder you already have, and a plan with no version for a bad week.
Now build the smallest programme that avoids all five. It will look unimpressive on paper and it will still be running next spring.
Then run the version of the list you have actual data for. What did your last three attempts have in common in the week they ended?
Not the reason you gave at the time. The conditions — a travel week, a sick child, a deload that quietly became a month, a Monday session you have never once completed.
Those are your measured failure modes, already collected, sitting in your calendar history for free. Almost nobody looks, because looking means reading a record of stopping.
The inverted plan is not a better programme. It is a programme carrying a named response to each of the four things that have genuinely stopped you before.
Business
Point it at cash first, because cash is the only failure that ends the company outright.
Then point it at the process where you are currently optimising the wrong end. Hiring is the clearest one. Most founders write a scorecard describing the ideal candidate, which produces a list of qualities everyone claims to have.
Invert it. Write the six behaviours that would make you let this person go in month four — misses commitments quietly, needs the brief three times, cannot be disagreed with, does not read, will not touch unglamorous work, treats the customer as an interruption. Then design the interview to detect those specific six, because they are observable and the ideal ones are not.
Retention takes the same treatment. Stop asking how to increase it and ask what an account does in the thirty days before it cancels.
Logins fall. The champion stops replying, and a support ticket goes unanswered for four days.
Those three are measurable today. The forward version of the question has been sitting on your quarterly plan for two years without moving.
AI Leverage
A language model will always give you an answer. That is its function and it is also the danger.
So use it in the reversed direction. Before a prompt goes into anything a customer touches, ask the model to generate the ten worst outputs it could plausibly produce under that prompt. You now have your guardrails, written by the thing you are guarding against.
Then red-team the pipeline itself. What input makes this workflow produce a confidently wrong answer that nobody catches?
An AI failure is almost never loud. It is fluent, well-formatted, and wrong in the third paragraph.
The best single use is adversarial: paste your own plan and instruct the model to argue that it fails, with the strongest available case and no hedging. You will get a better critic than most boards, in nine seconds, with no politics attached to the feedback.
The Drill
Pick the one bet the quarter depends on. One.
On Wednesday, block a single hour. Write one page, dated twelve months from today, that opens with the sentence: this failed, and here is why. Past tense throughout — the tense is the whole trick, because it removes the option of hedging.
Force yourself to name six causes. Not three. The first three are the ones you already worry about; the useful ones arrive at four and five.
Then mark which of the six are inside your control. Take the top two and convert each into a rule rather than a task.
A task gets done once and forgotten. A rule gets checked weekly and holds — no customer above twenty-five percent of revenue, no launch without a mobile load test, no hire without two reference calls you placed yourself.
Put the rules where you already look on Mondays. That is the entire drill.
Stoic parallel
Seneca practised premeditatio malorum — the deliberate rehearsal of loss. He would spend days living as though poor, on hard bread, so that poverty would arrive as a familiar guest rather than a catastrophe.
The purpose was never to feel bad. It was to price the thing accurately while calm, so that when it came he was not making decisions in the first hour of shock.
Marcus opened his mornings the same way, telling himself he would meet the ungrateful and the arrogant that day. He was not being bleak. He was removing surprise, because surprise is what makes an otherwise capable person act badly.
Seneca inverted his life. Munger inverted his investments. Same instrument, two thousand years apart, aimed at the same weakness in the same animal.
One model per week.
Applied to training, business, and AI leverage. No fluff.
Related models
- Arguing from First PrinciplesReduce the problem to what is irreducibly true, then rebuild without borrowing anyone else's conclusions.
- Second-Order ThinkingAnd then what? The first consequence is obvious and priced in. The second one is where the money is.
- Opportunity CostsThe real price of anything is the best thing you gave up to get it, and it never appears on an invoice.
Origin: Carl Gustav Jacobi, who told mathematicians to invert; carried into decision-making by Charlie Munger