Jeff Bezos divides all decisions into two types. He calls them doors.
Type 1 decisions are one-way doors. You walk through, and the door locks behind you. Choosing a co-founder. Selling the company. Going public. Signing an exclusive ten-year distribution deal. These decisions are irreversible or nearly so — the cost of unwinding them is so high that it functions as permanence. They demand careful deliberation, broad consultation, and the willingness to slow down when every instinct says move.
Type 2 decisions are two-way doors. You walk through, look around, and if you don't like what you see, you walk back. Launching a feature. Testing a pricing tier. Hiring for a new role before the role is fully defined. Opening a small office in a new market. These decisions are reversible at low cost. They should be made quickly, by individuals or small groups with good judgement, without committee review or multi-week approval cycles.
Bezos introduced the framework publicly in his 2015 letter to Amazon shareholders — the "Day 1" letter that became one of the most referenced documents in modern business strategy. The context wasn't abstract philosophy. It was an operational diagnosis of what was killing Amazon's speed as the company crossed 250,000 employees. His argument was precise: as organisations grow, they develop a systematic bias toward treating every decision like a Type 1. They apply the heavy machinery of analysis, consensus-building, and executive approval to choices that could be made in an afternoon and reversed by Tuesday. The result is a large company that moves like a large company — slowly, cautiously, and with the kind of deliberate thoroughness that is exactly right for irreversible bets and exactly wrong for everything else.
"Day 2 is stasis," Bezos wrote. "Followed by irrelevance. Followed by excruciating, painful decline. Followed by death." The Type 1 / Type 2 classification was his prescription for staying in Day 1 — the operating mode where speed and experimentation haven't been suffocated by process.
The insight cuts both directions. The more common error — and the one Bezos diagnosed at Amazon — is Type 2 decisions treated as Type 1. This is the disease of scale. A product manager wants to test a new onboarding flow. Instead of shipping an A/B test to 5% of users on Monday, she writes a six-page proposal, schedules three stakeholder reviews, waits for legal sign-off, and launches six weeks later. The cost of delay exceeded the cost of being wrong by an order of magnitude. The decision was a two-way door wearing a one-way door's costume.
But the inverse error is just as dangerous and far less discussed. Type 1 decisions treated as Type 2. The founder who fires a co-founder over a weekend argument without consulting the board. The CEO who accepts a term sheet with aggressive liquidation preferences because she wants to close the round before a competitor. The startup that open-sources its core IP on a whim, signalling a strategic direction that can't be credibly reversed. These are one-way doors kicked open at two-way-door speed.
The damage from this error compounds for years. A bad term sheet governs every subsequent financing round. A fired co-founder creates legal exposure, cultural shockwaves, and a narrative that follows the company through every future fundraise. The asymmetry is important: Type 2 decisions treated as Type 1 cost you time. Type 1 decisions treated as Type 2 cost you options.
The framework's value isn't in the classification itself — almost anyone can label a decision as reversible or irreversible after five minutes of thought. The value is in the organisational behaviour the classification unlocks. Once you name a decision as Type 2, you've given the team explicit permission to move fast, skip the approval chain, and accept the possibility of a wrong answer. Once you name a decision as Type 1, you've signalled that this one gets the full apparatus: data, debate, devil's advocacy, and the CEO's direct attention.
What makes the framework operationally powerful — rather than merely conceptual — is that it produces different behaviour at scale. A 10-person startup doesn't need a formal classification system; the founder's intuition handles both types. A 10,000-person company does. Without explicit classification, large organisations default to the highest-stakes decision protocol for every choice, because no individual wants to be responsible for a fast decision that goes wrong. The framework gives individuals and teams cover to move fast by making reversibility an institutional judgment, not a personal risk.
The distinction explains a paradox that confuses observers of Amazon. The company is simultaneously one of the fastest-moving large organisations on Earth (launching hundreds of experiments per week, entering new markets with minimal deliberation) and one of the most methodical (spending years and billions on irreversible infrastructure bets like AWS, Kindle, and fulfilment centre networks). It isn't schizophrenic. It's operating two different decision protocols, matched to two different decision types. Speed where speed is cheap. Caution where mistakes are expensive.
The metaphor of doors is deliberate. Doors are binary — open or closed — which forces a classification that gradient language ("somewhat reversible," "mostly permanent") avoids. Bezos wanted a forcing function, not a spectrum.
The door metaphor makes it psychologically harder to stall: if you've agreed that a decision is a two-way door, the question "why haven't we walked through yet?" becomes uncomfortable. The metaphor carries its own urgency. And it scales across cultures and languages within a global organisation — everyone understands doors.
Section 2
How to See It
The framework surfaces wherever an organisation's decision speed is mismatched to the reversibility of the decision at hand. The tell is either paralysis on low-stakes choices or recklessness on high-stakes ones.
Technology
You're seeing Reversible vs Irreversible Decisions when a startup ships a feature to 1% of users before the design is finalised, measures the response, and iterates — rather than spending three months perfecting it behind a feature flag. The team has correctly classified this as a Type 2 decision. If the feature fails, they revert the code. Facebook's early mantra "move fast and break things" was a Type 2 decision culture applied to product development. It worked until it didn't — the breaking started extending to infrastructure and trust, which are Type 1 domains.
Business
You're seeing Reversible vs Irreversible Decisions when a CEO insists on personal involvement in a specific category of choice — strategic partnerships, equity grants, major pricing architecture — while delegating everything else to team leads with wide latitude. The CEO isn't micromanaging selectively. She's applying different decision protocols to different reversibility classes. Bezos reportedly required personal sign-off on any Amazon decision involving more than $50 million in capital commitment or any deal that constrained future optionality. Everything else moved at team speed.
Investing
You're seeing Reversible vs Irreversible Decisions when a venture capitalist structures a deal with participation rights and pro rata follow-on provisions rather than a simple equity stake. The VC is preserving optionality — making the initial investment a two-way door by retaining the ability to increase or decrease exposure as information arrives. The founder who signs away board control or accepts full-ratchet anti-dilution in exchange for a faster close has turned a two-way door into a one-way door, often without realising it.
Personal life
You're seeing Reversible vs Irreversible Decisions when someone takes a six-month sabbatical to test a business idea rather than quitting outright. The sabbatical converts what looks like a one-way door (leaving a career) into a two-way door (leaving with a return ticket). Paul Graham has argued that one of the underappreciated advantages of being young is that almost all your decisions are Type 2 — you have decades to recover from bad choices, which means the rational move is to bias heavily toward action and experimentation.
Section 3
How to Use It
Decision filter
"Before analysing what to decide, classify the decision: can I reverse this at reasonable cost within a reasonable timeframe? If yes, make it fast and delegate it. If no, slow down and bring in the best thinking available. The classification itself is the most valuable step."
As a founder
Build the Type 1 / Type 2 classification into your company's operating system — not as a philosophy but as a process. Every decision that enters the executive team should be tagged: is this a one-way door or a two-way door?
One-way doors get the full treatment: written analysis, explicit assumptions, pre-mortem, senior review. Two-way doors get a bias toward action: a single owner makes the call, ships it, and reports back. The tagging itself is the intervention — it creates a shared vocabulary that makes decision speed an explicit, deliberate choice rather than an accidental byproduct of organisational culture.
Amazon formalised this at scale. New product launches, feature experiments, and pricing tests were explicitly labelled as Type 2 and delegated to two-pizza teams. Strategic platform decisions — AWS architecture, Prime membership structure, marketplace policy — were Type 1 and required Bezos's direct involvement and the famous six-page narrative memo process. The classification wasn't a suggestion. It was the organisational architecture.
The trap for early-stage founders is different: you'll default to treating everything as Type 2 because speed feels like your only advantage. It is — until it isn't. Choosing a co-founder is Type 1. Selecting a technical architecture that your first twenty engineers will build on for three years is Type 1. Signing a term sheet is Type 1. The velocity advantage of startups should be concentrated on the hundreds of Type 2 decisions you make weekly, not extended to the handful of Type 1 decisions that define your trajectory.
As an investor
Evaluate founders by watching how they classify decisions. A founder who agonises for weeks over logo colours and website copy is misclassifying Type 2 decisions — and will run the company slowly. A founder who signs binding agreements without reading the terms is misclassifying Type 1 decisions — and will accumulate structural damage.
The best founders display a visible gear-shift. They make dozens of small calls per day with minimal deliberation, then slow to a crawl when the stakes are irreversible. Watch for this pattern in due diligence conversations. Ask: "What's the last decision you reversed?" If the answer comes easily, the founder operates two-way doors comfortably. Ask: "What's the last decision you spent more than a month on?" If the answer involves a genuinely irreversible commitment — a key hire, a strategic partnership, equity structure — the founder understands the classification.
As a decision-maker
The highest-leverage application of this framework inside a large organisation is speed recovery. Most established companies are slow not because their people are slow but because they've wrapped every decision in Type 1 process. The cure is an audit: list the twenty most recent decisions that required executive approval, and classify each as Type 1 or Type 2. In most organisations, 70–80% will be Type 2 decisions that consumed Type 1 resources. Push those back down to team-level authority with explicit permission to move fast and be wrong.
Andy Grove at Intel used a version of this when he distinguished between decisions that affected Intel's strategic direction (Type 1 — Grove's personal attention) and decisions that affected execution within an established direction (Type 2 — delegated with aggressive timelines). The distinction allowed Intel to be both strategically deliberate and operationally fast during the critical 1985–1995 period when the company transitioned from memory to microprocessors.
Common misapplication: The most frequent error is treating reversibility as binary when it's actually a spectrum. Very few decisions are purely one-way or purely two-way. Most sit somewhere in between — partially reversible, with the cost of reversal varying by how long you wait.
Hiring is a good example: it looks reversible (you can fire someone), but the cost of a bad hire compounds with tenure — lost productivity, team disruption, severance, institutional knowledge walking out the door. A technology stack choice is similar: selecting a database on Day 1 is a two-way door; migrating after two years of accumulated data, integrations, and team expertise is substantially closer to a one-way door, even though the same decision is in play.
The framework works best when you ask not "can I reverse this?" but "what does reversal cost, and how does that cost change over time?" That second question — how cost escalates — is what separates skilled classification from naive classification.
Section 4
The Mechanism
Section 5
Founders & Leaders in Action
The framework reveals itself most clearly at moments when a leader consciously shifts decision speed — accelerating where reversibility permits, decelerating where permanence demands. The leaders below didn't just make good decisions. They made the right kind of decision at the right speed.
What's consistent across these cases is the visible gear-shift. The same person who agonises for months over a strategic commitment will make a product decision in minutes — not because they're inconsistent, but because they've internalised the classification and calibrated their process accordingly. The speed differential itself is the signature.
Bezos built the Type 1 / Type 2 classification into Amazon's operating DNA. The two-pizza team structure — small, autonomous groups with end-to-end ownership of a product or feature — was designed specifically to maximise Type 2 decision speed. Each team could ship, test, and revert without escalating. Thousands of experiments ran simultaneously because the organisational architecture matched the decision type.
The contrast with Type 1 decisions was stark. When Amazon was evaluating whether to build its own delivery network — a multi-billion-dollar, largely irreversible infrastructure commitment — Bezos spent over a year modelling the economics, visiting FedEx and UPS facilities, and running pilot programmes. The decision to launch Amazon Logistics wasn't made until the data was overwhelming. Similarly, the decision to create the Kindle required Bezos's direct involvement over multiple years — it meant Amazon was betting that digital reading would replace physical books, a thesis that, if wrong, would waste hundreds of millions and signal a strategic direction the company couldn't easily reverse.
The operational principle was explicit: the faster you make Type 2 decisions, the more experiments you run. The more experiments you run, the more information you generate. The more information you generate, the better your Type 1 decisions become. Speed on reversible choices compounds into wisdom on irreversible ones.
Bezos reinforced the classification through Amazon's memo culture. The six-page narrative memo — required for senior leadership meetings — was a Type 1 decision tool: it forced rigorous thinking for irreversible commitments. For Type 2 decisions, the expectation was the opposite. A two-pizza team didn't write a six-pager to launch an A/B test. They launched it. The process architecture mirrored the decision architecture — heavyweight tools for one-way doors, lightweight autonomy for two-way doors.
Hastings treated Netflix's 2007 streaming launch as a Type 2 decision — a low-cost experiment layered on top of the profitable DVD business. The initial streaming library was small. The feature was bundled free with existing DVD subscriptions. If it flopped, Netflix could pull it back without structural damage.
The decision to split DVD and streaming into separate businesses in 2011, however, was Type 1 — and Hastings misclassified it. The "Qwikster" rebranding attempted to separate the DVD business into a standalone entity, forcing customers to manage two accounts and two billing relationships. The backlash was immediate: Netflix lost 800,000 subscribers in a single quarter, and the stock dropped 77% from its July peak to its November trough. Hastings reversed the decision within weeks, but the damage — subscriber loss, reputational harm, stock collapse — demonstrated that what looked like a two-way door was functionally one-way. Customer trust, once broken at scale, doesn't reset by reverting a product decision.
The episode became a reference case inside Netflix for why classification matters. Hastings later described the error in terms Bezos would recognise: he'd applied Type 2 speed to a Type 1 choice. The streaming launch itself was the correct template — small, reversible, fast. The Qwikster split was the wrong template applied to a decision that affected millions of customer relationships simultaneously.
The lesson wasn't that Netflix should slow down. It was that Netflix should slow down on one-way doors and maintain speed on everything else. Hastings drew the line more carefully after 2011, and Netflix's subsequent decade — original content investment, international expansion, the shift to an ad-supported tier — showed a company that had internalised the distinction between decisions that deserve deliberation and decisions that deserve velocity.
Grove's most famous decision — exiting the memory business in 1985 — was a textbook Type 1 choice made with appropriate gravity. Intel had invented the DRAM memory chip in 1970. Memory was the company's identity, its founding product category, and the expertise of its engineering culture. Japanese competitors, led by NEC and Hitachi, had driven prices below Intel's cost of production. The memory division was losing $173 million annually.
Grove and co-founder Gordon Moore walked through what Grove later called "the revolving door test." He asked Moore: "If we got kicked out and the board brought in a new CEO, what would he do?" Moore answered immediately: "He'd get us out of memories." Grove replied: "Why shouldn't you and I walk out the door, come back in, and do it ourselves?" The framing was deliberate — they were acknowledging that this was a one-way door (no path back to memory leadership once they exited) and using the thought experiment to strip away the emotional attachment that was preventing the decision.
The exit from memory took two years of painful execution — plant closures, layoffs, and a complete reorientation of engineering culture toward microprocessors. Intel's revenue grew from $1.9 billion in 1986 to $25 billion by 1999. Grove treated the decision with the weight it deserved: irreversible, identity-defining, and worthy of the most rigorous analysis available. He didn't move fast. He moved right.
Marc AndreessenCo-Founder, Andreessen Horowitz, 2009–present
Andreessen Horowitz was built on a structural insight about decision reversibility in venture capital. Most individual investment decisions — writing a seed check, passing on a Series A — look irreversible but aren't. You can invest in the next round. You can find a similar company. The opportunity cost of a missed deal, while painful, rarely compounds into permanent damage.
Andreessen applied this to the firm's decision velocity. a16z was designed to make investment decisions faster than traditional VC firms — often within days rather than weeks. The thesis: in competitive deal environments, the speed of the commitment is itself a signal of conviction, and the cost of occasionally backing the wrong company is lower than the cost of consistently losing deals to faster-moving competitors. Each individual investment is partially reversible (through follow-on discipline, portfolio construction, and the option to let a position attrite), so the right move is to bias toward action.
The exception was platform-level decisions — the firm's strategic positioning, its expansion into crypto and bio, its media and marketing architecture. These were treated as one-way doors. Andreessen reportedly spent over a year evaluating the crypto thesis before launching the dedicated crypto fund in 2018, knowing that the positioning would define the firm's brand and talent pipeline for a decade. The firm's speed on deal flow coexisted with deliberation on strategic direction — two decision speeds matched to two reversibility classes.
When Nadella became CEO of Microsoft in February 2014, the company was structured around a Type 1 decision that his predecessor Steve Ballmer had made and couldn't reverse: the $7.2 billion acquisition of Nokia's phone business, closed just months before Nadella took office. The bet — that Microsoft could compete with Apple and Google in mobile hardware — was a one-way door. By July 2015, Nadella wrote down essentially the entire acquisition, recording a $7.6 billion impairment charge. The irreversible decision had already been made. All Nadella could do was stop the bleeding.
His own strategic signature was a different application of the framework. The shift from "Windows first" to "cloud first, mobile first" was a Type 1 decision — it redefined Microsoft's identity, restructured its engineering priorities, and signalled to enterprise customers that Azure would receive the company's best talent and investment. Nadella spent his first year building internal consensus, reorganising the leadership team, and laying the cultural groundwork before making the strategy public. He treated it with appropriate gravity.
Simultaneously, he pushed Microsoft's product teams toward Type 2 velocity on everything else. Office was released on iOS and Android — a move that the Windows-centric old guard would have debated for years. The decision was Type 2: if it didn't work, Microsoft could pull the apps. It did work. Office 365 became the entry point for millions of users who later migrated to Azure. LinkedIn integration, GitHub acquisition pricing, and Teams feature launches were all treated as high-speed experiments inside a strategic framework that was deliberately slow. Microsoft's market capitalisation grew from $300 billion to over $3 trillion in a decade. Two speeds. One company.
Section 6
Visual Explanation
Section 7
Connected Models
The reversibility framework doesn't exist in isolation. It connects to models that govern decision speed, decision quality, and the psychological traps that distort both. Understanding these connections is essential — the framework is most dangerous when applied without the guardrails that adjacent models provide, and most powerful when combined with the analytical tools that sharpen the classification.
The six models below form the immediate neighbourhood — some sharpen the classification, some check its misapplication, and some represent the natural next step after classification is complete.
Reinforces
Regret Minimization Framework
Bezos's regret minimization framework asks: "At 80, which choice would I regret more?" The reversibility framework asks: "Can I undo this choice?" Together, they form a complete decision protocol for the hardest calls. Regret minimization identifies which irreversible decisions are worth making despite the risk. The reversibility framework identifies which decisions are irreversible in the first place.
The sequence matters. Classify the decision first. If it's Type 2, make it fast — regret analysis is overkill for a reversible choice. If it's Type 1, then apply the regret framework: given that this door locks behind me, which direction will I regret more at the end of my life? Bezos used both at the founding moment of Amazon: the decision to start the company was Type 1 (he couldn't un-resign from D.E. Shaw and recapture the 1994 internet window), and the regret framework resolved the direction.
Reinforces
Second-Order Thinking
Second-order thinking — tracing consequences past the first link in the chain — is the analytical engine that makes reversibility classification accurate. A decision that looks reversible at the first order may become irreversible at the second. Hiring someone is reversible (you can fire them). Hiring someone who recruits their entire former team, reshapes the engineering culture, and builds architecture around their preferred stack is substantially less reversible — and that second-order chain takes six to twelve months to become visible.
The reinforcement runs both ways. The reversibility framework tells you when to invest in second-order analysis (Type 1 decisions, where the stakes justify the cognitive expense) and when to skip it (Type 2 decisions, where speed matters more than depth). Without second-order thinking, you'll misclassify decisions. Without the reversibility framework, you'll apply second-order thinking everywhere and drown in analysis.
Section 8
One Key Quote
"Some decisions are consequential and irreversible or nearly irreversible — one-way doors — and these decisions must be made methodically, carefully, slowly, with great deliberation and consultation. But most decisions aren't like that — they are changeable, reversible — they're two-way doors."
— Jeff Bezos, 2015 Letter to Amazon Shareholders
Section 9
Analyst's Take
Faster Than Normal — Editorial View
The genius of this framework isn't the insight that some decisions matter more than others. That's obvious. The genius is giving organisations a shared vocabulary — "Type 1" and "Type 2," "one-way door" and "two-way door" — that makes decision speed a deliberate, discussable variable rather than an emergent property of culture and hierarchy.
Most companies don't have a speed problem. They have a classification problem. They apply a single decision process — calibrated for their most consequential choices — to every choice they make. The result is an organisation that takes three weeks to approve a landing page change and three weeks to approve a strategic acquisition. Both get the same number of meetings, the same stakeholder reviews, the same slide decks. One of those timelines is appropriate. The other is organisational malpractice.
The framework's sharpest application is as a diagnostic tool. If you want to understand why a company is slow, don't look at its people or its processes. Look at its decision classification. Count the number of decisions that required VP-level approval in the last quarter and classify each as Type 1 or Type 2. In every organisation I've examined, the ratio is lopsided: 80% of escalated decisions were Type 2, consuming executive bandwidth that should have been reserved for the 20% that genuinely required it. The executives aren't the bottleneck. The classification is.
The inverse error — Type 1 decisions made at Type 2 speed — is rarer but far more damaging. I've watched founders accept term sheets with punitive provisions because they didn't want to "slow down momentum." I've seen CEOs fire co-founders impulsively and spend years dealing with the legal and cultural fallout. I've seen engineering teams adopt a core infrastructure dependency over a weekend Slack conversation and regret it for three years. Each case follows the same pattern: the decision felt reversible because the act was fast, but the consequences were permanent.
The hardest part of applying this framework is honest classification. Most people classify based on how the decision feels rather than what reversal would actually cost. A decision to enter a new market feels big and scary (loss aversion tags it as Type 1), but the pilot programme can be shut down in three months for minimal cost (it's actually Type 2). A decision to grant a key employee a board seat feels collegial and reversible (the relationship makes it seem Type 2), but unwinding a board seat requires legal action and destroys trust (it's functionally Type 1). The feeling and the reality diverge constantly. The framework only works when classification is based on the , not on the emotional weight of the choice.
Section 10
Test Yourself
The framework sounds straightforward. Applying it correctly — especially distinguishing genuine reversibility from the illusion of reversibility — is where most people stumble. The scenarios below test the edges: decisions that look reversible but aren't, decisions that look irreversible but are, and classification errors that produce real organisational damage.
Is this mental model at work here?
Scenario 1
A SaaS company's product team wants to test a new checkout flow. The VP of Product writes a detailed proposal, schedules reviews with legal, finance, and the CEO, and targets a launch date eight weeks out. A junior PM asks: 'Can't we just A/B test it with 5% of traffic this week and revert if metrics drop?' The VP responds: 'We need to get this right the first time.'
Scenario 2
A startup founder receives two term sheets. One offers better valuation but includes full-ratchet anti-dilution, participating preferred with a 3x cap, and a board seat for the lead investor. The other offers a lower valuation with standard terms. The founder is tempted to take the higher valuation and 'deal with the terms later.' Her advisor says: 'The valuation is a Type 2 decision — it'll get reset next round. The terms are Type 1 — they'll govern your company for its entire life.'
Scenario 3
The CEO of a 500-person company personally approves every new hire, reviewing each offer letter before it goes out. She explains: 'People are our most important asset. Every hire is a one-way door.' Average time-to-offer is 4.2 weeks. The company is losing candidates to competitors who move in 5 days.
Section 11
Top Resources
The best material on this framework starts with Bezos's own words — the shareholder letters are short and operational, not theoretical. From there, the decision theory and organisational design literature provides the intellectual scaffolding for why the classification works and how to implement it at scale.
The source document. Bezos introduces the Type 1 / Type 2 framework as part of a broader argument about how to maintain "Day 1" speed at scale. The letter is short, direct, and operational — not theoretical. Read it before reading anything else on the topic. Every subsequent treatment of the framework derives from these paragraphs. Pay particular attention to Bezos's warning about "Day 2" companies — his description of how Type 1 process applied universally leads to organisational death by slowness.
Grove's account of navigating Intel through its memory-to-microprocessor transition is the most detailed case study of a Type 1 decision made correctly. His concept of "strategic inflection points" — moments when the fundamentals shift enough to demand irreversible commitment — complements Bezos's framework by describing the conditions under which Type 1 decisions become unavoidable. The "revolving door test" described in the book is itself a decision-making tool for leaders paralysed by the irreversibility of the choice in front of them.
Hastings describes Netflix's culture of distributed decision-making, where Type 2 decisions are explicitly delegated and Type 1 decisions are escalated. The "context, not control" management philosophy is a direct operationalisation of the reversibility framework — give people the context to classify decisions correctly, then trust them to move at the appropriate speed. The Qwikster chapter is especially valuable as a case study in misclassification.
The academic foundation for why irreversible decisions deserve more deliberation. Dixit and Pindyck formalise the "option value of waiting" — the economic benefit of preserving the ability to choose later when facing uncertainty and irreversibility. Dense but essential for anyone who wants to understand the economic theory underneath Bezos's intuitive framework. The mathematical proof that "wait and see" has quantifiable value when decisions are irreversible — and negligible value when they're reversible — is the academic backbone of the Type 1 / Type 2 distinction.
A curated collection of Bezos's shareholder letters and public speeches, including the 2015 letter where the Type 1 / Type 2 framework appears. Isaacson's introduction contextualises the framework within Bezos's broader decision-making philosophy, connecting it to the regret minimization framework, the "disagree and commit" principle, and Amazon's memo culture. Reading the letters chronologically reveals how Bezos refined his thinking about organisational speed over two decades — the Type 1 / Type 2 distinction was the culmination, not the starting point.
Leaders who apply this model
Playbooks and public thinking from people closely associated with this idea.
Two decision types, two protocols — Type 1 (one-way doors) require deliberation; Type 2 (two-way doors) require speed. Most organisations err by applying Type 1 process to Type 2 decisions.
Tension
Loss Aversion
Loss aversion is the psychological engine behind the most common misapplication of the framework: treating Type 2 decisions as Type 1. Kahneman and Tversky showed that losses feel roughly twice as painful as equivalent gains. When a product team considers launching an imperfect feature, loss aversion amplifies the imagined cost of a bad user experience and mutes the real cost of delay. The team asks for more data, another review cycle, one more design iteration — all in service of avoiding a loss that, for a truly reversible decision, would cost almost nothing.
The tension is structural. Loss aversion makes every uncertain decision feel like it should be Type 1. The framework says most of them are Type 2. Resolving the tension requires overriding the emotional signal with a process-level classification: "Yes, this feels risky. Is the risk actually irreversible? No? Then ship it." Organisations that can't override loss aversion at the team level will default to Type 1 process for everything and wonder why they can't innovate.
Tension
Sunk [Cost](/mental-models/cost) Fallacy
The sunk cost fallacy distorts reversibility assessment in the opposite direction. Once an organisation has invested heavily in a decision, the sunk costs make reversal feel psychologically impossible — even when the decision is technically reversible. A company that has spent $20 million building a product nobody wants can, in theory, kill the product tomorrow. In practice, the sunk cost creates gravitational pull that converts a two-way door into a one-way door through sheer emotional weight.
The tension: the reversibility framework says "if you can undo it, undo it and move on." The sunk cost fallacy says "we've come too far to turn back." The framework is an antidote to this specific trap — by forcing an honest assessment of reversal cost as it exists today, not as it's inflated by past investment. The question isn't "how much have we spent?" It's "what does it cost to reverse right now, and is that cost lower than the cost of continuing?"
Leads-to
Iteration [Velocity](/mental-models/velocity)
The reversibility framework is the precondition for high iteration velocity. You can only iterate fast if you've identified which decisions permit fast iteration — the Type 2 decisions where wrong answers are cheap and information gain is high. Companies that move fast without classifying reversibility will eventually make an irreversible mistake at two-way-door speed. Companies that classify correctly can push Type 2 decision speed to its maximum, running more experiments per unit of time and generating the learning loops that compound into competitive advantage.
The causal chain is direct: correct classification → permission to move fast on Type 2 → more experiments → faster learning → better products. Amazon's thousands of simultaneous A/B tests, Spotify's squad model of autonomous feature teams, and Netflix's culture of "highly aligned, loosely coupled" teams all depend on the reversibility classification operating in the background. Without it, autonomy degrades into chaos or calcifies into bureaucracy.
Leads-to
Forcing Function
Requiring explicit reversibility classification before any decision reaches the executive team is itself a forcing function — it compels clarity about stakes before anyone begins debating options. The classification question ("Is this a one-way door or a two-way door?") forces the team to articulate what makes the decision consequential, what reversal would look like, and what reversal would cost. These are exactly the questions that sloppy decision processes skip.
The framework also functions as a forcing function for delegation. Once a decision is labelled Type 2, the executive team has created a structural expectation that it will be resolved at the team level without escalation. This forces team leads to develop their own decision-making capability rather than routing everything upward. The classification becomes a training mechanism: every Type 2 decision resolved without executive input builds the organisation's decision-making muscle at the level where speed matters most.
economics of reversal
One nuance that Bezos's original framing underplays: reversibility degrades over time. A decision that's fully reversible on Day 1 may be only partially reversible by Month 6 and effectively irreversible by Year 2. A pricing change is a two-way door on launch day. After twelve months, customers have anchored to the price, contracts reference it, and sales teams have built pipelines around it. The reversal cost has quietly compounded from trivial to substantial. The implication: even Type 2 decisions have a shelf life on their reversibility. The faster you evaluate and reverse a bad Type 2 decision, the cheaper the reversal. Delay converts two-way doors into one-way doors through the simple accumulation of dependencies.
There's a cultural dimension that Bezos's framing doesn't address directly. Some organisations punish wrong decisions regardless of reversibility. If a product manager ships a Type 2 experiment that fails, and the failure is visible to leadership, and leadership treats it as evidence of poor judgement rather than efficient learning, the classification system collapses. No one will make fast Type 2 decisions in a culture that punishes fast Type 2 failures. The framework requires a cultural foundation: the organisational willingness to treat reversed Type 2 decisions as information, not as mistakes. Amazon built this explicitly — the "Institutional Yes" principle meant that a manager needed a reason to say no, not a reason to say yes. Most companies operate the reverse.
My honest assessment: this is the single most operationally useful decision framework available to founders and executives. Not the most intellectually deep. Not the most theoretically elegant. The most useful — because it produces an immediate, concrete change in organisational behaviour. The day you start classifying decisions by reversibility is the day you stop treating your approval process as one-size-fits-all. That single shift — matching process weight to decision weight — recovers more speed than any agile methodology, reorg, or motivational offsite.
The test I give founders: pick the last ten decisions that went to your executive team. Classify each as Type 1 or Type 2. If more than three of the ten were genuinely Type 1, your classification is probably honest. If seven or more were Type 1, your organisation is either facing an existential crisis or — far more likely — you're applying the wrong process to the wrong decisions. Fix the classification and you fix the speed.
Scenario 4
A direct-to-consumer brand has been selling exclusively through its own website. The team proposes listing on Amazon's marketplace. The CEO pauses: 'If we list on Amazon, our customers learn to buy there instead of from us. Even if we delist later, that behaviour doesn't reverse. This looks like a two-way door, but it's a one-way door with a time delay.' She approves a limited pilot with a 90-day evaluation window.
