A reversible decision buys evidence before commitment.

Reversibility is a spectrum, not a switch

Almost nothing is perfectly reversible and almost nothing is perfectly permanent. A better question than “can I undo this?” is “what would undoing cost, and how long would it take?” Four kinds of cost matter most.

• Money: fees, deposits, notice payments, the price difference if you resell, and the cost of switching back.
• Time: how long it takes to undo, and whether there is a point after which it can no longer be undone.
• Trust: how far other people’s confidence in you, or your confidence in them, would suffer if you changed course.
• Optionality: which other options the decision closes, for example through exclusivity, a long term or a public announcement.

Reversibility is also not the same as low stakes. A decision can involve a lot of money and still be easy to undo, such as a refundable deposit. A decision can involve little money and be hard to reverse, such as a public statement. Judge the stakes and the reversibility separately.

How this method relates to other ideas

The two-way door metaphor tells you which kind of decision you face. This method goes one step further and turns that classification into a workflow: three questions that expose what could change your mind, and one small test with a measure, a success condition and a stop rule.

• A pilot or trial runs a small version of the real thing. This method tells you what the pilot must measure.
• A premortem imagines that the decision failed and asks why. The three questions do a similar job before you choose.
• An experiment, such as a field or online trial, compares alternatives under controlled conditions. A decision test can be much simpler; it only has to separate the options well enough to make the next step proportionate.

How to answer the three questions honestly

1. What becomes harder to recover?

Go through money, time, trust, access and options in turn and write one line for each. Note what you could recover fully, partly or not at all. The lines that say “not at all” are where the smallest test should aim.

2. Which assumption carries the most weight but has the least evidence?

List the assumptions behind the decision, for example “customers will pay this price” or “I will use it every week”. For each, mark how much depends on it and how much evidence you have. The riskiest assumption is the one with high dependence and low evidence. Test that one first, not the easiest one.

3. What smaller action would expose that assumption?

Look for a version of the decision that keeps the learning and drops most of the commitment: fewer people, a shorter time, a smaller amount, a narrower scope, or a borrowed or rented version instead of a purchase.

“I do not know yet” is a valid answer to any of the three. It usually points to what the test should find out.

The commitment ladder

Most decisions can be approached in steps, each of which costs a little more and teaches a little more. Stop climbing as soon as the evidence is clear.

1. A conversation: ask someone who has done it, or a person who would be affected.
2. A paper or simulated test: a budget, a schedule or a mock-up.
3. A trial or demo: use a real version briefly, often free or cheap.
4. A pilot: a small, time-boxed version with real stakes and an exit.
5. A limited commitment: a short contract, a first phase or a smaller quantity.
6. The full commitment, made with evidence in hand.

Anatomy of a good decision test

• A question: the single assumption the test will answer.
• A measure: something observable, ideally a number or a yes-or-no, not an impression.
• A success condition: what result would make you proceed, written before you start.
• A stop rule: what result, or what cost, would end the test early.
• A time limit and a cost cap: so the test cannot quietly grow into the decision itself.
• A decision rule: what you will do for each outcome, so a result cannot be reinterpreted afterwards.

Common mistakes

• Testing to confirm what you have already decided, rather than to find out.
• Making the test so small that it cannot tell the options apart.
• Making it so large that it is no longer reversible.
• Moving the success condition after seeing the result.
• Measuring effort instead of outcome, such as hours spent rather than customers reached.
• Testing under unrealistic conditions, such as only with friendly users or only in the best week.

Second example: a price increase

A small online shop wants to raise prices by ten per cent because costs have risen. The questions show that the weighty assumption is that regular customers will not leave, while the shop has no evidence on price sensitivity. Instead of changing every price, the owner raises the price of one popular product for two weeks, keeps the others unchanged, and compares conversion and repeat orders. The success condition and the stop rule are written down before the change goes live.

When not to run a test

• When the situation is urgent or involves safety, health or legal duties. Get appropriate professional help instead.
• When the test would cost as much as the decision, or would cause the same harm.
• When the decision is already easy to reverse. Just decide, and review it later.
• When the test would affect other people in ways they have not agreed to.