Jeff Bezos, 2016

“Most decisions should probably be made with somewhere around 70% of the information you wish you had.”

Jeff Bezos wrote this line in Amazon’s 2016 shareholder letter while discussing what he called high-velocity decision making. He distinguished consequential, difficult-to-reverse choices from the many decisions that can be changed after new evidence arrives. Large organizations, he argued, often apply the heavy process required for the first kind to the second kind, slowing action throughout the company. The 70-percent figure was not presented as a statistical law. It was a practical threshold meant to expose a common cost: waiting for near-complete information can make a reversible decision late without necessarily making it much better.

The surrounding passage supplies the condition that short versions often omit. Teams must be good at recognizing and correcting bad decisions. Speed without feedback is merely haste. A two-way-door choice is suitable for early action because it can be reversed at acceptable cost, results can be observed, and the team retains the ability to change course. A one-way-door choice—such as a large capital commitment, an irreversible data migration, or a decision with serious safety consequences—requires greater care. The discipline lies in classifying the decision accurately, not in applying one percentage to everything.

This distinction is valuable in software, where some choices are naturally testable. A limited feature experiment, interface variation, or internal workflow can often be tried with monitoring and a rollback plan. Other choices are less forgiving. Security architecture, collection of sensitive data, deletion policies, and infrastructure contracts may create lasting exposure or dependencies. Calling a decision reversible does not make it so; teams need concrete evidence that data can be restored, users can be protected, and commitments can actually be unwound. The cost of reversal includes trust and operational disruption, not only engineering time.

The quote remains a useful response to analysis paralysis, but its deeper lesson is organizational design. Fast learning requires small scopes, clear owners, observable outcomes, and authority to reverse course. It also requires a culture in which acknowledging an error is rewarded earlier than defending it. The number 70 is less important than the question behind it: what additional information would materially change this decision, and what will delay cost? When the choice is reversible and the feedback loop is strong, acting before certainty can produce knowledge that further meetings cannot. When the stakes are irreversible, restraint is the faster path to avoiding a much larger failure. Writing down the classification and rollback conditions before acting makes it harder to relabel a risky commitment as an experiment after problems appear.