Startup Annual Planning: Why Amazon Writes Six Pages Before It Funds Anything

Startup annual planning at Amazon starts around June and often isn't finished until February.

Six months to write a document that will be wrong by March. At a company with more than a million employees, part of that is bureaucracy. The shape of it is still worth stealing, because the process is built to survive being wrong.

Every team writes a narrative. Six pages, maximum. Appendices are allowed, and you can't cheat the page count with tiny fonts or a wider margin.

It opens with a Purpose and a Background, so a reader from any other function can follow it cold. Every claim is supposed to sit on research instead of the word "should".

Then the room goes quiet. The first fifteen minutes of the review are silent reading with a pen. No one presents.

Colin Bryar remembers the switch. Jeff Bezos stopped PowerPoint at S-team meetings, and the memos started at four pages. Six turned out to be right for an hour meeting, because a narrative carries roughly ten times the information, and it can hold a multi-causal argument that a deck flattens into bullets.

Bill Carr wrote down what the executives actually asked each plan. Are these the right input and output metrics. Does the team have too many or too few resources. Is the team thinking big enough. Where are the risks. Is the central assumption true, and the way to test that is to hunt for cases where it isn't.

By 2010 Bezos and the S-team were setting aside more than two weeks for those reviews, and spending two to four hours on each plan.

Here is the part most founders skip. Dave Anderson, who went through planning as a senior manager, describes the plan being revised almost continuously through finance reviews, VP reviews, top-down changes and S-team reviews. At some point it becomes OP2. It often isn't final until February.

Six months of revision on a plan that will still be wrong. The revision is the product. The plan gets rewritten because the reasoning sat on a page where a VP could point at it and ask whether it had stopped being true.

Adam Fishman argues from the other side, and he is worth reading before you copy any of this. He names two failure modes. The conviction fallacy: a long horizon pushes a team to overstate conviction, and once a belief is locked into the plan, changing course means admitting the plan was wrong. The financial fallacy: a top-down number built from bets whose probability is too thin for the discounting to mean anything.

His prescription for an unproven bet: about three months of forward-looking plan aimed at learning, kept out of the annual financial forecast.

So some good companies dropped the ritual. Linear runs about 80 people, profitable for more than three years, with no OKRs at all, keeping a few strategic goals as themes for the roadmap. PostHog required OKRs in 2022 and walked it back after engineers agonised over picking the right metrics. Quarterly planning there now runs 60 to 90 minutes, and teams change a goal themselves when circumstances change.

Falk Gottlob calls the annual strategy deck a fossil on arrival: too expensive to replan, so teams execute against assumptions everyone knows are stale. What he runs instead is one page. Five to ten beliefs, each tagged with a confidence level and a date last reviewed, each paired with the specific signal that would falsify it, reviewed every Monday in a 20-minute solo session.

Both approaches agree on more than they disagree on. The companies that dropped the annual plan kept the beliefs in writing and moved the review to weekly. Amazon kept the beliefs in a six-page document and made the whole chain of command read and argue with them.

The teams that fail are the ones with neither. A plan no one can interrogate, because the assumptions died in someone's head the week after the offsite.

Six pages is a forcing function. Short enough that one person can hold the argument, long enough that the argument has to exist somewhere outside it.

A plan's useful half is the record underneath it. What you believed in November. What you funded and what you cut. What you said would tell you it had stopped working. With that on paper, January is an afternoon of comparison. Without it, the replan comes out of memory, and memory keeps the outcome you wanted while quietly dropping the assumption that made it look reasonable.

Six pages, written in June, read in silence, argued over until February. The record outlives it.

A plan you can't re-read is a plan you can't revise.