In 1994, a thirty-year-old investment banker in New York was trying to decide whether to leave a comfortable, well-paid career to start an internet bookstore from his garage. The internet was new. Online shopping was barely a concept. His employer told him he was mad.
The banker, Jeff Bezos, later described the mental exercise that got him through the decision. He called it a regret minimisation framework. He pictured himself at eighty years old, looking back at his life. Which version of himself would have the fewer regrets — the one who had tried to build the bookstore and failed, or the one who hadn’t tried at all? When he framed it that way, he said, the answer was instantly clear. The failure was recoverable. The non-attempt would live with him for fifty years.
The bookstore was Amazon.
Bezos’s reframing has since become one of the most influential decision-making tools in popular culture. Graduates quote it at one another. Start-up founders write it on walls. It even appears, in various forms, in financial advice columns and career coaching books. And there’s real research underneath it that’s worth understanding — but also real limits to it that the motivational version tends to skip.
What the research on regret actually shows
The most important finding in this area came from a pair of psychologists named Thomas Gilovich and Victoria Medvec, working at Cornell University in the mid-1990s. They asked a simple question: when people look back over their lives, what do they regret most? The results were striking and, once you’ve heard them, hard to forget.
Over short time horizons — days, weeks, months — people’s regrets tend to be about actions. The thing they said at the party. The job they took. The relationship they shouldn’t have pursued. These action-regrets are emotionally intense but tend to fade with time. The mind finds ways to make peace with things you did, because those things are done, and you’ve absorbed their consequences into who you’ve become.
Over long time horizons — years, decades, entire lives — the balance shifts sharply. Gilovich and Medvec found that long-term regret clusters, overwhelmingly, around inactions. The chances not taken. The things never tried. The people not spoken to. The direction not chosen. These regrets tend to intensify rather than fade, because the mind can’t finish processing them. There’s no outcome to absorb, only an open question that never closes.
Later work by Neal Roese, a psychologist at Northwestern, extended these findings and found the pattern held across many cultures and contexts. If you ask a ninety-year-old what they regret, they are unlikely to list specific things they did. They are much more likely to list things they didn’t.
This is the empirical foundation for Bezos’s framework. The reason “imagine yourself at eighty” works is that by projecting forward to the end of life, you’re consulting the version of yourself that weighs inactions more heavily. The present version of you, with anxiety about near-term consequences, is more likely to be action-averse. The future version, the one whose judgement actually matters for long-term life satisfaction, is more likely to be action-supportive.
Why the framework has power
The framework has a quiet elegance. It does something very specific to the brain’s natural decision-making.
In any significant decision, you’re usually weighing two kinds of risk. The visible risks of acting — money lost, time spent, reputation risked, relationships strained. And the invisible risks of not acting — opportunity foregone, path not taken, person not become. The visible risks are almost always easier to feel and count. The invisible ones are, by their nature, hypothetical.
Most people under-weight the invisible risks. Regret-minimisation, by explicitly asking you to imagine looking back over decades, brings the invisible risks into view. It makes you count the cost of not doing, not just the cost of doing. And for actions that have large upside potential and recoverable downsides — starting a business, writing a novel, asking someone out, telling someone how you feel — this reweighting tends to push you toward the right decision.
Where the framework goes wrong
But the framework has real limits, and they’re worth taking seriously before you apply it to a major decision.
The first and sharpest critique comes from Keith Stanovich, a cognitive scientist at the University of Toronto whose work on rational decision-making emphasises expected-value reasoning: multiplying each outcome’s probability by its magnitude, and choosing the option with the highest expected payoff.
Stanovich’s argument, roughly: regret-minimisation ignores probability. It tells you to act in a way the hypothetical older you won’t regret, but it doesn’t tell you which actions are likely to produce good outcomes. A young man deciding whether to take his life savings to a casino could use regret-minimisation to convince himself to try, on the grounds that his eighty-year-old self would regret not having the experience. This is not a good decision. The regret framework was pointed at the wrong question.
The regret framework works well for decisions with asymmetric upside — large possible benefit, modest possible cost, recoverable downside. It works badly for decisions with asymmetric downside — modest possible benefit, catastrophic possible cost. The surgeon who decides to try an experimental procedure because my future self would regret not trying is not using the framework correctly.
The second critique, sharper still, comes from research by George Loewenstein on what he calls the hot-cold empathy gap. Loewenstein’s finding: we are systematically bad at predicting how our future selves will feel. The thirty-year-old imagining how the eighty-year-old will feel about a decision is mostly imagining how the thirty-year-old thinks the eighty-year-old ought to feel. The actual eighty-year-old, shaped by fifty years of intervening life, will often feel differently.
This doesn’t destroy the framework, but it suggests humility. Your future self is not a perfect oracle. They’re just another version of you with different information.
Where research agrees with the framework
All of that said, the underlying research does support the core intuition for a specific class of decision. For choices that are:
Significant enough to be remembered decades later,
Likely to involve potential new directions, identities, or relationships,
Recoverable if they go wrong (you can return to something like your previous life),
About personal engagement rather than pure probability.
For these decisions, regret-minimisation is well-supported. The data on what people actually regret late in life consistently points toward inactions. The framework tells you, usefully, to weight inaction-regret more heavily than your immediate anxious mind would want to.
For other kinds of decisions — highly probabilistic, involving serious downside, about which you have real statistical information — expected-value reasoning is more reliable. Annie Duke, a former professional poker player who has written widely on decision science, suggests using expected-value thinking for decisions you make repeatedly (where the probabilities will express themselves over many trials) and something closer to regret-minimisation for decisions you make once (where the single outcome’s emotional meaning matters more than its probability).
Two tools for two kinds of decisions
The practical conclusion is worth stating cleanly. You need two different tools in your mind, and you need to know when to use each.
For one-shot, significant, recoverable decisions — whether to take a year abroad, pursue a relationship, apply for a competitive programme, start a side project, speak to someone who has mattered to you, make a major life move — regret-minimisation is probably the right frame. Bezos’s exercise is genuinely useful here. The long-term regret research supports it. The invisible risk of inaction is often larger than your anxious present self believes.
For repeated, probability-laden decisions — investments, business strategies, medical choices, negotiations, anything you’ll do many times — expected-value thinking is the right frame. You want the choice that, averaged across many similar decisions, produces the best outcomes. Emotional regret is less reliable here because the probabilities matter more than the vividness of any single outcome.
Most people use one of these tools for everything, and the tool they use depends on their temperament. Action-averse people use expected-value thinking for everything, which causes them to analyse themselves out of the big life-shaping decisions. Action-prone people use regret-minimisation for everything, which causes them to take reckless probabilistic bets they can’t recover from.
The skill, as ever, is knowing which tool the decision in front of you needs — and being willing to switch.
The question that remains
Bezos’s regret-minimisation story has become famous because it worked spectacularly in his case. But the framework is not a prescription to be more aggressive in all decisions. It’s a prescription to take the long-horizon perspective on the specific class of decisions where your near-term anxieties are probably misleading you.
The question to carry, especially before a decision that has you paralysed, is probably the one Gilovich and Medvec’s research really supports:
Is what you’re afraid of a real cost, or an imagined one — and in thirty years, will either have mattered? Sometimes the honest answer is that the cost is real and matters. Most of the time, for the decisions that keep us up at night, it isn’t, and it won’t.
Key research referenced: Thomas Gilovich and Victoria Medvec’s research on the temporal pattern of regrets (Journal of Personality and Social Psychology, 1994); Neal Roese’s extensive work on counterfactual thinking and regret; Keith Stanovich’s work on rational decision-making; George Loewenstein’s research on the hot-cold empathy gap; Annie Duke’s writing on probabilistic thinking in decisions.