Reversible vs. Irreversible: A Founder’s Decision-Speed Framework

The single most correctable drag on founder velocity is not bad strategy — it is mismatched deliberation: spending three weeks on a pricing test and three days on a co-founder split. Jeff Bezos codified the antidote in Amazon’s 2015 Letter to Shareholders: sort every decision by its reversibility, then calibrate the process accordingly. Two-way doors get speed; one-way doors get rigor. The framework is deceptively simple and almost universally misapplied.

Why founders get decision speed exactly backwards

There is a persistent and costly cognitive error at the heart of most founding teams: the conflation of emotional weight with structural irreversibility. A decision feels large, so it receives a large process. A decision feels routine, so it receives a quick nod. The problem is that emotional weight and structural irreversibility correlate poorly. Choosing a brand name feels momentous; it is, in practice, almost entirely reversible — companies rebrand routinely, from Backrub to Google, from AuctionWeb to eBay.¹ Choosing a technical architecture feels like an engineering detail; depending on vendor lock-in and switching costs, it can be a genuine one-way door.

The result of this inversion is a founder who convenes a two-hour all-hands to debate a landing-page headline and signs a five-year office lease after a single walkthrough. Both errors are expensive. The first wastes the scarcest resource in any early-stage company — focused attention. The second forecloses optionality at precisely the moment when optionality is most valuable.

The corrective is not to move faster on everything. It is to build a classification system that routes each decision to the right process before deliberation begins. Bezos built exactly that system, and it is worth studying in its original form before operationalizing it.

The original framework: Type 1 and Type 2 decisions

In his 2015 Letter to Amazon Shareholders, Bezos drew a clean distinction between two categories.² Type 1 decisions are consequential and irreversible or nearly irreversible — one-way doors. Type 2 decisions are changeable and reversible — two-way doors. ³ “If you’ve made a suboptimal Type 2 decision, you don’t have to live with the consequences for that long. You can reopen the door and go back through.”

The prescriptions that follow from this classification are asymmetric. ⁴ Type 1 decisions “must be made methodically, carefully, slowly, with great deliberation and consultation.” Type 2 decisions “can and should be made quickly by high-judgment individuals or small groups.” The failure mode Bezos identified is not recklessness — it is bureaucratic overcaution applied to the wrong category. ⁵ “As organizations get larger, there seems to be a tendency to use the heavyweight Type 1 decision-making process on most decisions, including many Type 2 decisions. The end result of this is slowness, unthoughtful risk aversion, failure to experiment sufficiently, and consequently diminished invention.”

This is not a startup-specific pathology. It is the default trajectory of any organization that does not actively resist it. The question for founders is whether to install the resistance mechanism early, before scale makes the habit structural.

The information threshold: why 70% is enough

The framework has a natural companion rule on information sufficiency. In his 2016 Letter to Shareholders, Bezos stated that ⁶ “most decisions should probably be made with somewhere around 70 percent of the information you wish you had. If you wait for 90 percent, in most cases, you’re probably being slow.” The corollary is equally important: ⁷ “if you’re good at course correcting, being wrong may be less costly than you think, whereas being slow is going to be expensive for sure.”

The 70% rule is not a license for sloppiness. It is a recognition that the marginal value of additional information declines steeply once a decision is reversible. The cost of gathering the remaining 30% — in time, attention, and opportunity cost — almost always exceeds the expected improvement in outcome quality. This is the economic logic that underpins the entire framework: for reversible decisions, the cost of delay is real and certain; the benefit of extra deliberation is speculative and small.

Research in decision science supports this intuition. ⁸ A 2024 study published in the Journal of Neuroscience by Kira, Zylberberg, and Shadlen at Harvard Medical School and Columbia University demonstrated that human decision-makers flexibly adjust the balance between decision speed and accuracy in response to changes in the cost of deliberation time — modulating the amount of evidence required before committing to a choice. The implication for organizational design is direct: when the cost of delay is made explicit and visible, teams naturally calibrate toward faster, appropriately confident decisions.

The academic parallel: fast decisions are not low-quality decisions

The intuition that speed trades off against quality is widespread and largely wrong. ⁹ Kathleen Eisenhardt’s landmark 1989 study, “Making Fast Strategic Decisions in High-Velocity Environments,” published in the Academy of Management Journal, found that fast decision-makers in the microcomputer industry used more information, not less, than slow decision-makers, and developed more alternatives simultaneously. ¹⁰ Eisenhardt found strong evidence of a link between decision speed and firm performance: the quick-deciding firms performed well; the slowly deciding firms failed or did poorly.

The mechanism Eisenhardt identified — real-time information, parallel alternatives, experienced counselors, active conflict resolution — maps cleanly onto what Bezos operationalized at Amazon. Speed is not achieved by skipping steps; it is achieved by running the right steps in parallel and eliminating the wrong steps entirely. For Type 2 decisions, the “wrong steps” are the consensus-building rituals that exist to manage the consequences of irreversibility — consequences that, by definition, do not apply.

The spectrum problem: reversibility is not binary

The framework’s most dangerous misreading is treating reversibility as a binary property. It is not. Most decisions sit on a spectrum, and the classification itself requires judgment.

Consider the canonical counterexample from Amazon’s own history. ¹¹ Bezos has noted that neither Amazon Prime nor AWS were one-way door decisions at launch — in both cases, Amazon launched without significant fanfare and could have wound either down without significant fanfare. The reversibility was a function of scale and commitment at the moment of decision, not of the eventual size of the business. Today, shutting down either would be anything but easy. The door changed classification as the company walked through it.

This points to a subtler principle that decision theorists have formalized: ¹² organizations should be especially careful about decisions that become irreversible through accumulation. Any single decision might be reversible, but a pattern of decisions can lock in a trajectory. A company that makes a series of individually reversible choices to expand in a particular market may find that the accumulated investment, relationships, and organizational focus have made exit effectively impossible. The one-way door was assembled incrementally, one two-way door at a time.

The practical implication is that classification is not a one-time act performed at the moment of decision. It is a periodic audit of the decision portfolio. Founders who institutionalize this audit — asking quarterly which reversible bets are quietly becoming irreversible commitments — maintain optionality that their competitors surrender by default.

Operationalizing the framework: three design choices

1. Classify before you deliberate

The classification question — “Is this a one-way or two-way door?” — must precede the substantive discussion, not follow it. ¹³ Amazon teams explicitly ask this question before any major decision. When the classification happens after deliberation has begun, anchoring bias ensures that the process already underway will be treated as appropriate regardless of the decision’s actual reversibility. The sequence matters: classify, then calibrate the process, then deliberate.

A useful operational heuristic: if the decision can be substantially reversed within 30 days without major financial, legal, or reputational consequence, treat it as a two-way door and assign it to the smallest competent team available. If reversal would require more than 30 days or would trigger significant switching costs, escalate the classification and the process.

2. Delegate two-way doors aggressively

The second design choice concerns who makes the decision, not just how fast. ¹⁴ Amazon’s “Bias for Action” leadership principle states that “many decisions and actions are reversible and do not need extensive study” and explicitly values “calculated risk taking.” The organizational implication is that reversible decisions should be pushed to the lowest level of competent judgment — not because senior leaders lack the capacity to make them, but because concentrating reversible decisions at the top creates a bottleneck that compounds across every layer of the organization.

AWS’s own published guidance makes the mechanism explicit: ¹⁵ “leaders can accelerate innovation by empowering teams to make more two-way door decisions independently.” When teams closest to the customer have autonomy over reversible choices, the organization runs experiments at the rate of its teams rather than at the rate of its executive calendar. For a fifty-person startup, this is the difference between ten experiments per quarter and one hundred.

3. Make irreversible decisions as late as possible

The asymmetry runs in both directions. ¹⁶ The correct timing rule is: make reversible decisions as soon as possible; make irreversible decisions as late as possible. For one-way doors, delay is not procrastination — it is optionality preservation. Every additional week of information gathering before a genuinely irreversible commitment has positive expected value, because the cost of reversal is high and the cost of delay is comparatively low.

This is the logic Apple applied when transitioning the Mac lineup from Intel to its own M-series silicon — a genuine one-way door that required rearchitecting macOS, retooling developer workflows, and securing ecosystem commitments before the first chip shipped.¹⁷ Senior leadership deliberated for months, conducted extensive performance testing, and locked in developer commitments before the public announcement. The deliberation was not bureaucratic friction; it was appropriate process for an irreversible architectural commitment.

The “disagree and commit” extension

The framework has a third element that founders frequently omit: the resolution mechanism for two-way door decisions where consensus is absent. Bezos addressed this directly in the 2016 shareholder letter. ¹⁸ “If you have conviction on a particular direction even though there’s no consensus, it’s helpful to say, ‘Look, I know we disagree on this but will you gamble with me on it? Disagree and commit?'” The phrase saves time precisely because it decouples intellectual agreement from operational commitment — a separation that is only safe when the decision is reversible.

“Disagree and commit” applied to a one-way door is reckless. Applied to a two-way door, it is the correct mechanism for breaking deliberative deadlock without false consensus. The reversibility classification is what makes the phrase safe to use.

What this means

The compounding return on decision architecture

The framework’s deepest value is not the individual decision it improves — it is the compounding effect of running it consistently across hundreds of decisions per year. A team that correctly classifies and accelerates two-way door decisions runs more experiments, generates more learning, and builds more institutional knowledge than a team of equal intelligence that applies uniform deliberation to everything. Over eighteen months, the gap in accumulated learning between these two teams is not incremental; it is structural.

This is what founder agency looks like at the systems level: not the heroic individual decision made under pressure, but the architecture that ensures the right decisions get the right process, every time, without requiring heroism. The one-way door framework is not a thinking tool for exceptional moments. It is an operating system for ordinary ones — and ordinary decisions, made well and fast, are where the compounding actually happens.

Founders who build this classification habit early — before scale makes deliberative bureaucracy feel normal — will find that the framework does something more valuable than saving time on any single choice. It trains the entire organization to distinguish between the cost of being wrong and the cost of being slow, and to treat those two costs as the distinct and asymmetric things they are.