← Writing

MBA · STO · ADHD

Invert, always invert: the founder's cheapest edge

Most founders spend their planning time asking how to win. The higher-leverage question is the opposite one: what would guarantee that I lose? Charlie Munger built a large part of his reputation on that inversion — and it is the closest thing to a free tool a founder has. It is also wrong often enough that using it without knowing its three failure modes is its own kind of risk.

A rule borrowed from a mathematician

The line Munger repeated for decades — "invert, always invert" — was not his. He took it from the German mathematician Carl Gustav Jacob Jacobi, who told his students man muss immer umkehren: many hard problems yield only when you turn them around and attack them backwards. Jacobi meant it about differential equations. Munger applied it to money, careers, and human misjudgement, and in doing so turned a technical heuristic into a general operating discipline.

The core claim is simple. For a large class of problems, describing the goal directly is hard, but describing its negation is easy. You may not be able to specify what makes a marriage happy, but you can list, quickly and with confidence, the things that reliably wreck one. You may not know how to build a category-defining company, but you can name the moves that kill startups: run out of cash, found with the wrong person, build something nobody wants, ignore the one regulation that ends you. Munger's advice is to write that second list first, and then spend your energy not doing what is on it.

He was explicit that this is a thinking tool, not a philosophy of life. In his 2007 commencement talk at USC Law he framed it as method: "a lot of success in life and business comes from knowing what you really want to avoid — like early death and a bad marriage." The order matters. Avoidance comes first, aspiration second.

Why the backward question is easier

Inversion is not a motivational trick. It works because of an asymmetry in the structure of most problems, and that asymmetry is worth spelling out from first principles rather than taking on Munger's authority.

Consider the space of possible outcomes for a new company. The set of states we would call "success" is small and specific: a particular product, at a particular price, reaching a particular market, at a particular time. The set of states we would call "failure" is enormous — nearly everything else. This is the same asymmetry a physicist sees between order and disorder: there are very few arrangements of a system that count as ordered and vast numbers that count as chaotic. A plan that tries to hit the small target directly has to get many things right at once. A plan that tries to avoid the large failure region only has to eliminate the biggest ways of landing in it.

There is a second reason, and it is about evidence. Failure modes are more observable than success paths. Thousands of companies have died of the same handful of causes, and those causes are documented. The specific combination that produced one outlier success is, by definition, rare and hard to copy — and usually over-explained after the fact. So the base rates you can actually trust are the ones on the failure side. Inversion points your attention at the part of the problem where the data is good.

The third reason is psychological, and Daniel Kahneman named it: we are loss-averse and we are overconfident, in that order. Asked to forecast our own project, we generate an optimistic, coherent story and stop. Asked instead to imagine the project has already failed, we suddenly get specific. The framing does the work.

The operational version: the pre-mortem

The most useful founder-grade application of inversion is not a mindset. It is a meeting. Gary Klein formalised it as the pre-mortem: before committing to a plan, assume it is a year later and the plan has failed completely, then have everyone write down why. Kahneman cites it as one of the few practical corrections for overconfidence he knows of, precisely because it legitimises doubt. In a normal review, the person who raises a risk looks disloyal. In a pre-mortem, naming the failure is the assignment, so the quiet objections that people were sitting on come out.

The mechanics are worth keeping literal, because the value leaks out if you soften them:

  • State the plan as decided. The pre-mortem runs on a plan you are about to commit to, not a vague direction. Ambiguity gives people room to avoid the uncomfortable answers.
  • Assume failure as a fact, not a risk. The prompt is "it failed — why," not "it might fail." The past tense is what unlocks the specifics.
  • Write before speaking. Everyone lists causes independently, then reads them out. This defeats the anchoring that would otherwise collapse the room onto the first, loudest opinion.
  • Sort by controllability. Separate failure modes you can cheaply prevent now from ones you can only monitor. Spend on the first; instrument the second.

For a company of one — which is how I run mine — the pre-mortem loses its group-debiasing benefit, so it needs a substitute for the missing second brain. Mine is to run the inversion against written base rates rather than my own memory: a checklist of the specific ways ventures like mine have died, kept in the wiki and re-read before any irreversible move. The tool is the same; the source of the failure list moves from a room full of people to a document.

Where it earns its keep: pricing, growth, resilience

Inversion is most valuable exactly where intuition is most confident and most wrong — in decisions whose second-order effects arrive on a delay.

Take pricing. The forward question — "what price maximises revenue?" — invites a growth hack: drop the price, add a discount, run the promotion. The inverted question — "what pricing move would quietly destroy this business a year from now?" — surfaces the real hazards: anchoring the market so low you can never raise, training your best customers to wait for the sale, selling below the cost to serve a segment you then can't fire. The forward move often looks good this quarter and is a debt that comes due later. Howard Marks calls this second-level thinking, and it is inversion wearing a different hat: the first-level answer is the obvious one, and the second-level answer asks what happens after everyone acts on the obvious one.

Or take resilience, which is my day job. Enterprise risk work is almost entirely inverted by construction. You do not ask how the system succeeds; you ask how it fails, in what order, and what the failure of one part does to the next. That is inversion applied to structure rather than to a single decision, and it is why Munger's rule feels native to anyone who has worked in risk: we were already writing the failure list for a living.

DecisionForward question (first-order)Inverted question (what would guarantee failure)
PricingWhat price maximises revenue now?What price would train the market to never pay more, or make my best segment unprofitable to serve?
Hiring / co-founderWho is the most talented person available?What partner dynamic reliably ends companies — and does this one have it?
RoadmapWhat feature would delight users?What are we building that nobody asked for, while the thing that kills us goes unfixed?
ComplianceHow do we move fast?Which single rule, broken once, ends the whole venture regardless of everything else?

The pattern across the right-hand column is that each inverted question points at a small number of high-consequence, low-visibility errors — the ones that do not show up in this month's metrics and are therefore the ones a forward-looking dashboard will miss.

A worked pre-mortem, start to finish

Abstract advice is cheap, so here is the tool run once, concretely. Suppose you are about to launch a paid tier on a product that has only ever been free. The forward question — "what will make people upgrade?" — produces a feature list and a pricing page in an afternoon, and it feels like progress. Invert it instead: assume it is a year later, the paid tier is a visible, named failure, and write down every reason it went wrong.

The list comes fast, and the speed is the evidence that the knowledge was already there. The free tier was good enough that nobody felt the pull to pay. The paywall landed on the one capability your most active users had quietly come to consider a right, and they revolted in public where prospects could see it. Support load tripled, because billing questions are the most expensive kind of ticket and the least deflectable. A competitor stayed free on exactly the feature you chose to gate, and made your paywall their marketing. You discounted in month two to hit a board number, and taught the market the real price was the lower one. None of these required foresight. They are the standard ways this specific move dies, and every one of them was available, for free, to anyone who asked the inverted question before writing a line of code.

Now do the part that turns a list into a decision: sort by controllability. "The free tier is too generous" is a design choice you still own — you can move the line now, before the revolt, not after it. "Support load spikes" you can instrument and staff ahead of time. "A competitor undercuts on the gated feature" you cannot control at all, so you assign it to monitoring rather than spending a week defending against it today. The output is not a mood of caution or a longer to-do list. It is a short, ranked set of preventable causes to act on now and an even shorter set of things to merely watch. That is the entire deliverable, and it cost twenty minutes and no capital.

Notice what the exercise did not do. It did not tell you whether to launch the paid tier at all — inversion is silent on the upside, and a founder who ran this and concluded "too risky, don't ship" would have misused it. What it did was strip the foreseeable, self-inflicted failures out of a plan you had already decided to pursue, which is exactly the layer where inversion is strongest and the layer most plans skip.

Where inversion breaks — the three failure modes

A tool sold as universal is being sold dishonestly, and inversion has real limits. Three of them matter enough that ignoring any one turns the tool against you.

One: avoiding failure is not the same as winning. Inversion is a member of the via negativa family — improvement by removal, which Nassim Taleb argues is more robust than improvement by addition because we know harm more reliably than we know benefit. That is true for survival. It is not true for outcomes that require a positive, specific bet. Some games are won only by doing the improbable right thing, not by avoiding the obvious wrong ones. A startup that flawlessly avoids every known way to die can still fail the only test that matters — building something people want badly enough. James Carse's distinction is useful here: you can play not to lose in a finite game, but the infinite game rewards moves that have no template. Inversion protects the downside; it will not author your upside.

Two: the failure list only contains known failure modes. This is the sharpest limit, and it is Taleb's black swan in different words. A pre-mortem enumerates the ways you can imagine failing. The event that actually ends you is disproportionately likely to be one nobody put on the list, because the imaginable risks get priced and defended while the unimagined one arrives undefended. Inversion reduces variance against the failures you can foresee and does close to nothing against the ones you can't. Treating a completed failure list as safety is precisely the overconfidence the tool was supposed to cure.

Three: for some minds, inversion is a trap door into paralysis. The tool asks you to hold every way a thing can go wrong in working memory. For a founder with an anxious or ADHD-shaped attention system, that is not a neutral request — it can tip from clarifying into rumination, where naming failures stops being analysis and becomes a loop. The discipline that makes inversion safe is a hard stop: you invert to produce a finite list of preventable causes, you act on them, and then you close the question. Inversion without that stop is just dread with a business-school vocabulary.

It is remarkable how much long-term advantage people like us have gotten by trying to be consistently not stupid, instead of trying to be very intelligent.Charlie Munger

Munger's line is usually read as humility. Read it instead as scope. "Consistently not stupid" is a claim about the survival layer — the failures you can foresee and forestall. It is the right ambition for the base of the stack, and the wrong ceiling for the top of it. The founder's job is to run inversion ruthlessly on the things that can kill the company, and then to switch tools entirely for the thing that can build it.

So invert first. List what would guarantee you lose, and refuse to do it. Then notice the moment the failure list is finished — and ask the forward question you were avoiding.

Sources

  1. primaryCharlie T. Munger, Poor Charlie's Almanack: The Wit and Wisdom of Charles T. Munger (2005) — the "invert, always invert" discipline and the Jacobi attribution.
  2. primaryCharlie Munger, USC Gould School of Law commencement address (2007) — "knowing what you really want to avoid."
  3. primaryGary Klein, "Performing a Project Premortem," Harvard Business Review (Sept. 2007).
  4. primaryDaniel Kahneman, Thinking, Fast and Slow (2011) — overconfidence, loss aversion, and the pre-mortem as a corrective.
  5. primaryNassim N. Taleb, The Black Swan (2007) and Antifragile (2012) — unknown failure modes and via negativa.
  6. secondaryHoward Marks, The Most Important Thing (2011) — second-level thinking.
  7. secondaryJames P. Carse, Finite and Infinite Games (1986) — playing not to lose vs. playing to continue.