The Misunderstood One-Way Door: Decision Making Is a Classification Problem, Not a Deliberation Skill

The short answer. Jeff Bezos’s 2015 Amazon shareholder letter split decisions into Type 1, one-way doors that cannot be reversed, and Type 2, two-way doors that can. The passage is almost always quoted as a warning about big decisions. It is mostly the opposite. Bezos gives irreversible decisions one sentence and spends the rest warning that large organizations apply heavyweight Type 1 process to Type 2 decisions, producing “slowness, unthoughtful risk aversion, failure to experiment sufficiently, and consequently diminished invention.” The capability that matters is therefore classification, telling the doors apart, rather than deliberation. Almost no company has evidence about which of its managers can do it.

Everyone quotes the one-way door. Almost nobody finishes the paragraph.

I have heard “one-way door” in four executive conversations this month. It comes from Jeff Bezos, the 2015 Amazon letter to shareholders, and it has become the polite corporate way of saying that a decision is important.

So I read the actual paragraph.

---

“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 Shareholders · Amazon

That is the part everyone knows. Bezos gives irreversible decisions one sentence. One. The rest of the passage is about the opposite failure.

“As organizations get larger, there seems to be a tendency to use the heavy-weight 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.”

Jeff Bezos · 2015 Letter to Shareholders · Amazon

He was not asking leaders to be more careful. He was warning them that they are being careful about the wrong things.

He made the same argument again the following year. The 2016 letter, under the heading high-velocity decision making, says most decisions should be made with about 70 percent of the information you wish you had, and that if you wait for 90 percent you are probably being slow.

The skill is classification, not deliberation

If you read the whole passage, the capability changes shape. It is not the ability to deliberate well. Deliberation is the easy part, and organizations are already oversupplied with it.

The capability is classification. Knowing which door you are standing in front of, and then matching your process to the answer. There are two ways to get it wrong and both are expensive.

Why smart organizations get this wrong on purpose

The misclassification is not stupidity, and treating it as stupidity is why the memos telling people to move faster never work.

It is an incentive asymmetry. A fast decision that goes wrong is visible, attributable, and has a name attached to it. A reversible decision slowed to a crawl is invisible, shared across a committee, and belongs to nobody. One failure ends up in a performance review. The other ends up in a calendar invite.

Given those odds, and given no evidence that would protect a manager who moves quickly, routing everything through the heavy process is the individually rational choice. It is only collectively catastrophic.

The cost is measurable. Paul Nutt tracked 356 real organizational decisions over two years and found that half of them failed. Fewer than 20 percent of the processes developed more than one option, and where they did, success rose from 56 percent to 70 percent. “Failure to experiment sufficiently” is not a slogan. It has a citation behind it.

The uncomfortable part: attendance is not evidence

Nearly every company I speak to can show me who attended the workshop on decisive leadership. Not one can tell me which of their managers can tell the doors apart under real pressure.

That is not a curriculum problem. An attendance record confirms that a session happened. It says nothing about behavior, and it was never designed to. Completion rates, satisfaction scores and confidence surveys all share the same limitation: they measure whether the event occurred, not whether anyone got better at the work.

It is an evidence problem. And it does not belong to whoever owns the learning calendar. It belongs to whoever owns succession, transformation and enterprise risk, because those are the people who have to bet on a manager and currently do it on instinct and reputation.

What we do about it at RCM ThinkLabs

Full disclosure, I build an instrument for exactly this problem, so I am not a neutral party. Here is how it works, and where it stops.

RCM ThinkLabs puts people inside an immersive narrative where interests genuinely conflict, information is incomplete, and no outcome is clean. About fifteen minutes a session, over weeks. While they work, the system reads a multitude of microskills invisibly.

• The measurement itself stays invisible. People see their own progress inside the world they are playing; what they never see is the read being taken on their behavior. The moment people know which behavior is being assessed, they perform an archetype instead of behaving.
• Every read is against that person’s own baseline, taken before anything starts. Not against a norm group, and not against their peers.
• The output is behavioral evidence, not a completion rate. It is designed to survive a skeptical CFO asking one follow-up question.

What that produced in a live deployment. At a US defense software company, active listening moved from 22 to 41 on our index in 39 days. 70 percent practiced daily, voluntarily, with nobody assigned to it. Measured at day 39 of a 66 day cycle, not at completion. Every published figure ships with its anchors attached, exclusions included.

The honest constraint

We do not measure the real decisions made in your live environment, and I will not pretend otherwise. We measure behavior inside a constructed world where interests conflict. That is the claim and it stops there.

It is also, as far as I can find, more than anyone currently has.

A diagnostic you can run this week

You do not need us to start. Three questions, and the discomfort is the point.

• How many of the decisions your team escalated last quarter were actually two-way doors? Pick five at random and check.
• What did the speed cost? Not the wrong answers. The right answers that arrived six weeks late.
• Do you have behavioral evidence of how your managers decide, or do you have attendance records? If it is the second, you are in the majority, and you are already asking a better question than your dashboard was built to answer.

Common questions

What did Jeff Bezos actually mean by one-way and two-way doors?

In the 2015 Amazon shareholder letter, Bezos called irreversible decisions Type 1, or one-way doors, and reversible decisions Type 2, or two-way doors. The passage is usually quoted as a warning about big decisions. It is mostly the opposite. Bezos gives one-way doors a single sentence and spends the rest of the passage warning that large organizations apply heavyweight Type 1 process to Type 2 decisions, producing slowness, unthoughtful risk aversion, failure to experiment and diminished invention.

What is the difference between a Type 1 and a Type 2 decision?

A Type 1 decision is irreversible or nearly irreversible, so the cost of being wrong is permanent and it warrants slow, methodical deliberation and consultation. A Type 2 decision is reversible, so the cost of being wrong is the cost of reversing it, and it should be made quickly by an individual or a small team. The difficulty is not applying the right process. It is correctly identifying which type you are facing before you choose a process.

Why do managers misclassify Type 2 decisions as Type 1?

Because the incentives are asymmetric. A fast decision that goes wrong is visible, attributable and career damaging. A reversible decision slowed to a crawl is invisible, shared across a committee and nobody’s fault. Without evidence that protects a manager who moves quickly, routing everything through the heavy process is the individually rational choice, even though it is collectively expensive.

How do you measure decision-making ability in managers?

Attendance records, completion rates and self-assessments measure whether an event happened, not whether behavior changed. Behavioral measurement instead observes what a person actually does in situations where interests conflict and information is incomplete, reads specific microskills from that behavior, and compares each person against their own baseline rather than against a peer ranking.

What is the difference between decision-making courses and behavioral measurement?

A course delivers content and reports who completed it. Behavioral measurement reports what people did. The first produces an attendance record, which tells a succession or risk owner nothing about who can be trusted with a consequential call. The second produces evidence against a baseline, which is the thing those owners actually need and rarely have.

Is RCM ThinkLabs a leadership simulation?

No. A simulation asks someone to role-play a scenario and usually scores the answer they give. RCM ThinkLabs is an immersive narrative that runs over weeks, where characters hold genuinely conflicting interests, no outcome is clean, and nothing is scored in front of the participant. The measurement is invisible and continuous, which is what stops people performing for the assessment instead of behaving.

How is this different from a 360 review or a psychometric assessment?

A 360 asks colleagues for their opinion of someone once or twice a year, and everyone involved knows an assessment is happening. A psychometric instrument measures traits in a single sitting. RCM ThinkLabs reads behavior continuously while a person is doing something they chose to do, and compares them to their own earlier behavior rather than to a norm group.

Can behavioral evidence be used for succession planning?

That is the closest fit. Succession decisions are themselves one-way doors, and they are usually made on reputation, tenure and instinct. Behavioral evidence about how a person acts when interests conflict gives a succession or risk owner something checkable to put next to the instinct. It does not replace the decision, it informs it.

So: how many of the doors your team walked through last quarter were actually one-way?

See what behavioral evidence looks like. RCM ThinkLabs measures how people act when interests conflict, against their own baseline, with the exclusions published. rcmlabs.io

Sources. Jeff Bezos, 2015 Letter to Shareholders, Amazon. Jeff Bezos, 2016 Letter to Shareholders, Amazon. Paul C. Nutt, “Surprising but true: half the decisions in organizations fail,” Academy of Management Executive, 1999.