Jeff Bezos’ Regret Minimization Framework for Decisions You Can’t Undo

You have a decision sitting open that will not reverse once you make it. The Jeff Bezos regret minimization framework is the test most people reach for in that spot. Picture yourself at 80, look back at the choice in front of you, and ask which option you would regret less. Bezos used it in 1994 to leave a senior job on Wall Street and start Amazon.

Thirty years on it gets quoted at people weighing a resignation, a relocation, a proposal, a second degree. Some of those decisions it fits. Several it does not, and used on the wrong decision it produces a confident answer to a question nobody should have asked.

This piece sets out what Bezos described, in his own words, the distinction from his later writing that decides whether the framework applies, two further places it misleads, and the finding from the research on regret that makes it safer to use.

What the Jeff Bezos Regret Minimization Framework Actually Says

The fullest account is in an interview Bezos gave the Academy of Achievement in San Antonio in May 2001. In the spring of 1994 he was a senior vice president at D. E. Shaw, a New York investment firm, writing software that traded stocks, when he came across a figure he described as a wake-up call: web usage “was growing at 2,300 percent a year.” He decided to sell books online.

He told his boss. The boss took him for a two-hour walk in Central Park, said the idea sounded good but would be a better idea for someone who did not already have a good job, and asked him to think about it for 48 hours. It was the middle of the year. “When you do that,” Bezos said of leaving then, “you walk away from your annual bonus.”

Then the sentence the framework is built on. He said that what made the decision “incredibly easy” was something that, in his words, only a nerd would call a regret minimization framework. He projected himself forward to age 80 and asked what that man would regret. “I knew that when I was 80 I was not going to regret having tried this.” And the other side: “I knew that if I failed I wouldn’t regret that, but I knew the one thing I might regret is not ever having tried.” He said that not trying “would haunt me every day.”

He told the same story nine years later at Princeton’s baccalaureate on 30 May 2010, in a speech he called “We are What We Choose”. The wording there is plainer. “I decided I had to give it a shot.” “I didn’t think I’d regret trying and failing.” The speech ends with the line “In the end, we are our choices.”

Read both accounts together and the framework has three working parts. A long horizon, which strips out the discomfort of the next six months. A focus on the person deciding rather than on the outcome, since the 80-year-old is asked about regret rather than returns. And an asymmetry Bezos states outright, that failing after trying is survivable and never trying is not. That asymmetry is the engine, and the thing to examine before you borrow it.

One-Way Doors and Two-Way Doors

Bezos supplied the test himself. In Amazon’s 2015 letter to shareholders he split decisions into two kinds. Some are “consequential and irreversible or nearly irreversible”, which he called “one-way doors” and labelled Type 1, to be made slowly and with care. Most decisions, he wrote, “are changeable, reversible”, and those are the “two-way doors”. If a Type 2 decision turns out wrong, “You can reopen the door and go back through.” He wanted those made fast, by individuals or small groups, and he warned that treating two-way doors as one-way doors produces “slowness, unthoughtful risk aversion, failure to experiment sufficiently”.

Leaving a senior post in mid-1994 to start an internet bookshop was a one-way door. The bonus was gone, the seat would be filled, and the window on the web was open for a limited time. The regret minimization framework is a Type 1 instrument, a long and heavy view of a choice, which is right when the choice is heavy and cannot be undone.

The trouble starts when people point it at two-way doors. Should I apply for the internal role? Should I ask for more money? Should I take the evening course? Each of those can be reversed or repeated within a year. Summoning your 80-year-old self to adjudicate them adds weight the decision cannot carry, and weight is what makes people freeze.

For a two-way door the useful question is cheaper. What does it cost to try, and how fast will I know? A salary negotiation email costs twenty minutes and answers itself within a fortnight. The conversation an introvert has been avoiding at an industry dinner is covered in our networking tips for introverts, and the worst case is an awkward ten minutes. Neither needs a deathbed.

Which Decisions the Framework Fits

This is the sorting our newsroom uses when someone brings the framework to a decision.

Decision	Door type	Does the framework help?	The better check
Leaving a secure job to start a company	One-way	Yes, this is what it was built for	Add the runway test below: can you state the loss and live with it?
Relocating a family to another country	One-way	Partly, it ignores the other people	Run the framework once per person affected, then compare
Asking for a raise or a title change	Two-way	No, it inflates a cheap test	Cost to try, time to an answer
Taking a course or learning a new skill	Two-way	No, you can stop or restart	Smallest version you can begin this month
Turning down a one-time offer with a deadline	One-way, in both directions	Yes, because declining is also irreversible	Write out the regret of both branches, dated
Staying put for another year	Usually two-way, sometimes not	Only if the window is closing	Name the date after which staying becomes a one-way door

The last row is the one most people misread. Staying feels like the absence of a decision, and the framework’s lean toward action makes it feel like the regret-heavy option by default. Often it is reversible for a long time, and then at some point it stops being so. The task is to find that date, which is a calmer exercise than imagining yourself at 80. Our piece on career change at 40 is largely about locating that date honestly.

The Safety Net Problem

Go back to the asymmetry. “If I failed I wouldn’t regret that.” That sentence is only true for a particular kind of person in a particular position, and Bezos was candid about his. He had a good job, which is why his boss made the remark about someone who did not. He had a skill set that Wall Street was paying for and would pay for again. He had discussed the move with his wife and, as he told the Academy, she was supportive. The failed version of his plan would have returned him, a year or two later, to roughly where he started, minus a bonus.

For someone in that position the framework is right to tilt toward action, because the downside of acting is bounded and the downside of not acting is open-ended. For someone without that cushion the tilt is a bias. An operations manager with two children, a mortgage at the top of her range and four months of savings cannot say that failing would leave no regret. Failing might cost the house, and her 80-year-old self would remember the house.

So the framework needs a gate in front of it. Before asking what you would regret, write down what the failed version of the action costs, in months of income and in specific things lost, and decide whether you can carry that. Our net worth by age piece gives the benchmarks people compare themselves against; for this purpose the comparison is beside the point and the number that matters is runway. If the loss is survivable, the framework applies as Bezos used it. If it is not, the right move is usually to shrink the action until the loss is survivable, then run the framework on the smaller version.

That is also the honest reading of his story. He had an idea with a number attached, a specific product to sell, and 48 hours of forced reflection imposed by a boss who liked him. The framework was the last step of a decision, after the analysis, and the analysis is the part that gets left out of the retelling.

The People the Decision Affects

The second gap is quieter. The framework has one witness, and that witness is you. The 80-year-old looking back is asked about his own regret. Nobody else is in the room.

Most one-way doors in adult life open onto other people. A move takes a partner out of her job and children out of their school. A founder’s first two years are paid for, in part, by whoever shares the household. Bezos had the conversation with his wife before the one with his boss, by his own account, so the people were consulted before the framework ran. The framework itself contains no step for that. Borrow it without the preceding conversation and you have borrowed half the method.

The repair is mechanical. Run the question once for each person the decision materially affects. Where the answers point the same way, you have something stronger than one person’s projection. Where they diverge, you have found the real decision, which is about how to share a cost rather than whether to act. The same logic is why we argue, in our piece on decision-making in business, for naming who bears the downside before anyone votes on the upside.

What the Regret Research Actually Shows

Thomas Gilovich and Victoria Husted Medvec published “The Temporal Pattern to the Experience of Regret” in the Journal of Personality and Social Psychology in 1994. When they asked people to look back over their lives, regrettable failures to act outnumbered regrettable actions by nearly two to one, a split of 63 percent to 37 percent in their second study. Their summary line is the asymmetry in one sentence: “Actions cause more pain in the short-term, but inactions are regretted more in the long run.” The following year they set out why in “The experience of regret: What, when, and why” in Psychological Review.

The pattern has held up, with a wrinkle. In 2023 Gilovich and Jerry Richardson re-ran the study in the field at a museum of psychological science in Chicago and reported that the interaction between action and time frame replicated, while “the precise pattern of that interaction diverged from that reported earlier.” The direction survives. The exact shape is less tidy than the 1994 numbers.

One caveat follows for anyone using the framework. If long-run regret clusters around inactions, then an 80-year-old consulted about regret will tend to vote for action whatever the decision is. That is partly wisdom and partly an artefact. A failed action closes; you know what happened, you tell a story about it, the pain fades. An inaction stays open, because you never find out, so there is nothing to make peace with. The framework is sampling a judge with a known lean. That is fine when the inaction is itself a one-way door, a window that will close, because then the lean points at something real. It is misleading when the inaction can be reversed next year, because the imagined regret is a feature of how memory treats open questions rather than a fact about the decision.

One practical consequence. Predicted regret and remembered regret are different things. Write the prediction down, with the date, the cost you expect to bear and the regret you expect to avoid. Memory techniques, including the ones in our guide to how to remember things quickly, are good for names and figures and useless against the quiet rewriting a decision undergoes after the fact.

How to Use It This Month

Take the decision you have been circling and classify it first. Is the door one-way or two-way, and if two-way, what is the cost to try and the time to an answer? If it is a two-way door, stop reading this and go and try it. If the thing you keep not doing is a skill you have never built, the smallest version is a learning plan rather than a resignation, and our piece on how to learn faster is where to start it.

If the door is one-way, run the gate. Write the cost of the failed version in months of runway and specific losses, and decide whether you can carry it. If you cannot, shrink the action until you can. Then name every person the decision lands on and run the 80-year-old question for each of them, out loud, with them in the room where that is possible.

Only then ask Bezos’s question of yourself. Give it the time he was given: two hours away from the desk and 48 hours before anything is signed. The longer form of that discipline, a whole week set aside for one set of questions, is in our piece on Bill Gates’ think week. When in the day to do that thinking is a separate matter, and Bezos has said a fair amount about it, which our piece on Jeff Bezos’ morning routine covers.

Once the one-way decision is made, put the framework away. What follows is a year of execution, which runs on a productivity plan rather than on projected regret. For the two-way decisions that fill the rest of the calendar, our guide to a decision-making framework you will actually use is built for that, and structured decision-making covers the team version.

Our position, having watched the Jeff Bezos regret minimization framework applied to a great many decisions that were nothing like leaving a Wall Street firm in 1994, is that it is a good instrument with a narrow range. One-way doors, survivable downside, other people already consulted. Inside that range it cuts through months of dithering in an afternoon. Outside it, it mostly manufactures drama.

Frequently Asked Questions

What is the Jeff Bezos regret minimization framework in one sentence?

Project yourself to age 80, look back at the decision you are facing, and choose the option that leaves the least regret. Bezos described it in a 2001 Academy of Achievement interview as the test that made leaving Wall Street to start Amazon in 1994 feel easy.

When should you not use the regret minimization framework?

On reversible decisions, which Bezos called two-way doors in Amazon’s 2015 shareholder letter. Asking for a raise or taking a course can be undone or repeated, and treating them as deathbed questions makes them heavier than they are. It also needs care when failure would be ruinous, and when the decision lands on people who have not been asked.

Does the research on regret support the framework?

Broadly, with a caveat. Gilovich and Medvec found in 1994 that long-run regrets are mostly about inactions, around 63 percent to 37 percent, and a 2023 replication found the same direction with a different shape. That lean means an imagined 80-year-old will usually vote for action, which is useful when the inaction is itself irreversible and misleading when it is not.

The framework earns its reputation on the decision it was invented for and loses it on the ones it gets borrowed for. Bezos had the growth figure, a job to fall back on and the conversation at home settled before he asked his 80-year-old self anything, and the question only works in that order.