Two-Way Door Decisions

The problem it solves

Most organisations run a single decision-making process, and it's calibrated for the worst thing that could plausibly happen, so the same document, the same review and the same six people turn up whether the question is which analytics tool to trial or whether to close an office.

That has two costs, of which one is visible. The visible one is speed, because everything takes as long as the most dangerous thing takes. The invisible one is worse, because a process that runs on everything gets treated as a formality, so it becomes the thing you route around or the thing you complete after deciding, and by the time a genuinely irreversible choice arrives the process meant to catch it has already been trained to wave things through.

The instinct is to sort decisions by size, or by cost, or by how senior the person asking is, and none of those predicts how much care a decision deserves.

The idea

Ask a different question first. What would it cost to walk this back?

Some decisions are consequential and irreversible or nearly irreversible — one-way doors

The framing is Bezos's, and what makes it useful is that it isn't about magnitude, because a large decision that can be undone in a fortnight has earned less scrutiny than a small one that can't be undone at all. The economics predates the metaphor by forty years, since Arrow and Fisher showed in 1974 that irreversibility carries a cost of its own by destroying the option to decide later with more information.

The letter presents this as two types, and that's the version worth improving on, because real decisions don't come in two kinds.

Enlarge: The question is what reversal would cost, not how large the decision looks.

The question is what reversal would cost, not how large the decision looks.Close

The question is what reversal would cost, not how large the decision looks.

Almost everything interesting sits in the middle of the axis, and the middle is where the misplacements happen. A senior hire is the clearest case, because it's reversible on paper at a cost most organisations won't pay, so it gets processed like the reversible decision it technically is and behaves like the irreversible one it actually is.

How to use it

Ask the reversal question before comparing the options rather than after, because once you're weighing option A against option B the framing has already been set by whoever put the options on the table.

Estimate the cost concretely rather than adjectivally, so not "fairly reversible", but what we'd actually have to do, who we'd have to tell, what we'd already have spent, and what people outside the company would have seen by then. Written down like that, most decisions place themselves.

Then match the process to the position on the axis rather than to the size of the number or the seniority of the person asking. At the reversible end the payoff isn't really speed, but rather that the decision can move to whoever is closest to the information, because the cost of them getting it wrong is bounded and legible. At the irreversible end, spend the time the other end gave back.

And spend some of it on the move that matters most, which is looking for the version of the same decision that stays reversible for longer. Lease rather than buy, run one market rather than the lot, and keep the old tables while deferring the drop. Where a decision sits on this axis is often a design choice rather than a fact about the decision, and buying reversal time is usually cheaper than the review the irreversible version would deserve.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

Whoever wants to move is the one doing the classifying. "It's a two-way door" is now a sentence people say to end a conversation, and there's no referee. The heuristic has no defence against being asserted, and a rule that can be invoked unilaterally by the party it benefits will be.

Reversible doesn't mean free. You can switch a feature off, though you can't un-show it to the customers who used it and you can't recover the two months. The door swings back and the room on the other side has changed.

A door nobody walks through isn't a door. This is the serious objection, because reversing a decision requires somebody to be seen to have been wrong, usually the person who made the call, and the escalation-of-commitment literature is consistent about what happens next, which is that people put more behind their failing choices rather than less. Unless the reversal is scheduled and owned, "we can always undo it" describes a permission nobody will use.

It doesn't travel across balance sheets. Amazon in 2015 could absorb a reversible bet that would have ended a thirty-person company, because reversal cost is relative to what you have, in cash, in attention, and in the number of times your team will watch you change direction before they stop believing the next one.

People are not config values. Hiring, restructuring, promotion and redundancy are the decisions the heuristic handles worst, because they're formally reversible and humanly not. Undoing a hire costs that person a year of their life and costs you the team's belief that you know what you're doing, and neither of those appears in a reversal-cost estimate unless somebody puts it there deliberately.

The evidence is uneven. The economics of irreversibility is well established and the escalation-of-commitment findings are among the more replicated results in organisational psychology, and the management heuristic built on top of them is neither. It's one company's practice, described in three paragraphs by its founder, and adopted at a scale wildly out of proportion to the evidence supporting it.

In one line

Ask what walking it back would cost before deciding how carefully to walk forward, and be honest that most organisations never walk back.