Every explainer of the one-way door model stops in the same place: it tells you which kind of door you’re standing at. None of them tell you what to say next, out loud, with four other people in the room waiting on you. And none warn you that a door can swing both ways for you while opening only one way for the person it lands on.
There’s a three-question test that handles both. It runs in under a minute, without pausing the meeting.
What Bezos meant by one-way and two-way doors
Jeff Bezos, in Amazon’s 2015 letter to shareholders, split decisions into two kinds. Type 1 decisions are the one-way doors, the ones you can’t get back through once you’ve walked out, and his instruction for those is that “these decisions must be made methodically, carefully, slowly, with great deliberation and consultation.” Type 2 decisions reopen: “you don’t have to live with the consequences for that long. You can reopen the door and go back through.” Those he wanted made fast. “Type 2 decisions can and should be made quickly by high judgment individuals or small groups.”
That’s the model, and longer explanations of it exist. The habits around it sit in our guide for managers who want to decide better. What follows is the ninety seconds after you understand Bezos.
The three-question triage script you can run out loud
Frameworks for decision making teach you how to decide, not how to sort, and sorting comes first. Test any candidate question by saying it aloud: if a call center supervisor can’t say it in a planning meeting without sounding like she’s reading from a book, it isn’t the question.
- “Can I personally undo this without asking anyone’s permission?”
- “Can the other person undo their part as easily?”
- “If I have to walk this back, who talks about it?”
Between them they sort two costs: the operational cost of undoing the thing, and the reputational cost of having been seen to undo it. Question one is the self-test everybody already runs. Question two is the short one, and usually the answer is an obvious yes. Question three catches what the first two miss, because a decision can be cheap to undo and still leave a mark.
Bain’s research on decision effectiveness puts its top group ahead by 3.2x on revenue growth, 3.6x on EBT growth and 5.7x on shareholder return, averaged over five years. Sorting is the free part of that.
“We can turn it off on Thursday if the numbers look bad.” Mark runs growth at a scheduling-software company, and he said it eleven seconds into a twenty-minute meeting about a landing page test. He’d run the three questions while the last person was still talking. The test is his to start and his to kill, so yes. The designer can put the old page back, so yes. If it flops, it’s a line in the weekly update. It shipped Tuesday. (Mark and the people who follow are illustrative composites, not real individuals.)
What to do when question one isn’t a clean yes
Sometimes you can only undo it inside a window: before the contract countersigns, before the all-hands, before Friday’s payroll run. Say the window out loud and put an hour on it, then write it into your decision log.
The individual contributor’s dilemma: act or escalate
All of that assumes you’re the one who decides. Plenty of people aren’t, and for them the question is whether the decision is theirs at all.
Rachel handles support tickets at a similar company. The canned reply for refund requests opened with two sentences of policy before any apology, so she rewrote it, used it on the next forty tickets, and mentioned it in her Monday update. She could have put the old template back in ninety seconds; waiting a week for permission would have cost forty customers a worse email.
Daniel, a data analyst, had something that looked identical. He was about to change the default date filter on a shared dashboard from ninety days to thirty. Four teams open that dashboard on Monday morning and read whatever is on it as the quarter’s number, so undoing it wouldn’t undo what they’d already reported upward. He escalated in one sentence: “I want to switch the ops dashboard default to thirty days, which I can undo, but people are quoting that number in reviews, so I’d rather you make the call than have me break someone’s deck.”
Bain’s RAPID framework names who decides and who recommends. Daniel worked out which of those he was in about ten seconds.
Reversible for the company, irreversible for the person
Every decision has at least two undo buttons on it. There’s the one you press, and there’s the one belonging to whoever the decision lands on. Almost nobody checks the second, partly because it isn’t on the same screen, and partly because the first one is genuinely, verifiably, working.
This is where the door model quietly stops being about doors. A door is a fixture, and it’s the same door for everyone who walks through it. What you’re sorting here isn’t a fixture. It costs a different amount for each person standing at it, and the person with the authority to do the sorting is usually the person with the cheapest way out.
The list of decisions shaped like this is longer than it looks, and most of it sits at the administrative end of managing a team: a shift change, a desk move, a project reassignment, who runs briefings for a quarter. None of them get announced as consequential. Every one of them costs the company a few minutes to reverse. What it costs the person is set by what they’ve already done in response, and that is information the manager doesn’t hold and rarely goes looking for.
The schedule that already left the building
Laura supervises a support floor of nineteen agents. In March she moved four people from the 8am to 4pm shift onto 10am to 6pm, to cover an evening queue that was backing up. She checked the obvious things first. The rota tool takes a change in about four minutes. No contract specifies hours. HR had no objection, and payroll was unaffected. By every measure available to her the change was a two-way door, and she was right about that. Eleven days later, once the evening spike turned out to be fallout from a botched release, she moved everyone back. It took four minutes again.
What she hadn’t asked was whether the four people could move back as easily as the rota could. One of them had spent that week and a half rebuilding an after-school arrangement around the new hours: a neighbor collecting her son on Tuesdays and Thursdays, a daycare place she’d taken off a waiting list and paid a deposit on, and a swap with her husband’s own shift pattern that he’d had to request from his employer in writing. The rota reverted in four minutes. The deposit was already spent, and the husband’s shift request had already been approved by an employer who wasn’t going to withdraw it.
Laura isn’t careless. She asked more questions than most supervisors would have, and every answer she got back was accurate. The question she didn’t ask was whether the undo was hers alone to perform.
The version of question two that would have caught this takes nine words and about four seconds. Asked in March, before the rota changed, it produces the word “daycare,” and Laura either has a different conversation or the same one with two weeks’ notice attached to it.
The agent said one thing about it, two weeks after the reversal, in the corridor: “I don’t mind the shift. I mind that I’m the one who paid for it being temporary.”
The reporting line that reads as a verdict
Tom is a senior engineer who, in a reorg, stopped reporting to the VP of engineering and started reporting to a newly hired director one level below her. Twenty-six people moved that week. Tom’s move was one line in a spreadsheet, and on the org chart it’s as reversible as any other line, since the next reorg can put it back and reorgs at that company come round roughly every eighteen months.
The VP’s reasoning was sound and she explained it clearly. She’d grown to twenty-six directs, she couldn’t give any of them real attention, and the director had been hired precisely so that a group of senior engineers would get a manager with time for them. Every word of that is true, and Tom believed it.
It didn’t matter, because two things had already happened by the time the email went out. The first is that Tom had spent the previous year telling people he reported to the VP, since at that company reporting to the VP is what a staff-level promotion case gets built on. The second is that his skip-level, the meeting where the VP had been forming her own opinion of him for two years, became a meeting that happens if there’s time.
A reorg can put the line back, but it can’t re-run the eighteen months in which a director’s secondhand impressions stand in for a VP’s firsthand ones, and it can’t reach the peer at another company who asks him, next spring, who he reports to now.
When he raised it, four weeks in, he didn’t say any of that. He said: “I just want to make sure this isn’t a signal I’m supposed to be reading.” The VP told him honestly that it wasn’t, which was true, and which didn’t answer the question he was actually asking, because the signal was never hers to control.
Question two has an answer here, and it’s an uncomfortable one. The VP could have asked Tom what the change would cost him to undo, and Tom, four weeks in, couldn’t have told her, because the cost wasn’t visible to him yet either. Some of what lands on a person takes a year to surface. The question is still worth asking. It just isn’t a guarantee.
Tom still reports to the director. The next reorg is due in the spring. He’s stopped bringing it up.
Some reversible decisions cost you anyway
Run the script honestly and you’ll still hit calls that pass questions one and two and cost you anyway. There’s nobody else in this section: you and two columns.
| What your reversal costs the system | What your reversal costs the record |
|---|---|
| The setting flips back inside the hour | People remember you pushed it |
| The schedule reverts and payroll balances by Friday | Your next proposal picks up an extra week of review |
| The announcement gets a correction | The correction is the version people forward |
| By month end there’s nothing in the ledger to find | By month end it’s been said about you twice |
You already know the left column clears; that’s what reversible means. The right column is a tally nobody publishes and everybody keeps. It clears, if at all, in a planning meeting you’re not in.
There’s no fix here. Two ways people dodge it, both costlier: deciding more slowly than the decision deserves, which is decision fatigue by another road, or deciding fast and then refusing to reverse, which is sunk cost and one of the cognitive biases that outlasts knowing its name.
Neither stops the tally.
Whose door is it, really?
Go back to the three questions and notice which of them you can answer sitting still. Question one you answer out of what you already know about your own authority. Question three you answer out of your own read of how this company talks about people. Both are answerable from inside your own head, which is exactly why they feel like real questions.
Question two isn’t answerable that way. What it costs the other person to undo their part is a fact only they have. To get it you have to ask them, in the room, which means saying out loud that you don’t know what this will cost the person you’re deciding about. The question doesn’t get skipped because it’s minor. It gets skipped because it’s the only one of the three you can’t answer alone.
You ranked it third yourself, back when it arrived as the short one between two longer ones. Nearly everyone does.
So next time you reach it, don’t answer it. Ask it: “Before I do this, what would it cost you to put it back?” Then stop talking until they’ve answered. If you want to rehearse that before the real conversation, Merlin will run it with you.