Jeff Bezos

The pattern they actually represent

In the 2009 letter to shareholders, Amazon audited its own goal list in public. There were 452 goals for the year ahead, each with an owner, a deliverable and a target completion date. 360 of them touched customer experience directly. The word revenue appeared eight times and free cash flow four. The terms net income, gross profit and operating profit appeared zero times.

A public document, saying that the three numbers most businesses run on are not what anyone gets asked about in a review. The letter gives the reason. Energy spent on the controllable inputs is what maximises the outputs later.

Read the mechanism, not the slogan. An input is something a named person can act on this week: a delivery time, an out-of-stock rate, a page that loads slowly, a price. An output is what happens to you if enough inputs move. Amazon put owners and dates on the first category and let the second arrive.

The habit predates the audit by twelve years. The 1997 letter names the measures he expects to be judged on: customer and revenue growth, the degree to which customers keep buying on a repeat basis, and the strength of the brand. Repeat purchase rate is a leading indicator. It moves before revenue does, and it moves because somebody did something.

By 2001 the loop is written out for shareholders in plain language. Cost improvement makes lower prices affordable. Lower prices drive growth. Growth spreads fixed costs across more sales, which makes further price cuts possible. Then he tells shareholders to expect the loop repeated.

The same discipline paid for the infrastructure. Amazon launched S3 in March 2006, selling outside developers the storage the shop already ran on, and did not report AWS separately until the first-quarter 2015 results. Nine years inside a retailer's P&L, carried by customers who came to buy books and shoes.

The blockage it speaks to

Read this one if you open the same dashboard every morning and close it without changing anything.

Revenue. Followers. Signups. The number is down, or flat, or up for a reason you cannot name, and the next hour of your day is identical either way. That is a category error, not a discipline problem. You are watching an output, and an output is not actionable. Nothing you do today shows up in it today, so the number becomes weather. You check it for a mood reading and you get one.

The subtler version is worse, because from the inside it looks like measurement is happening. The 2016 letter names it: process as proxy, survey as proxy. Fifty-five per cent of beta testers report being satisfied, up from forty-seven. Bezos calls that hard to interpret and capable of misleading you. A rising proxy is a false all-clear. The obstacle is not that you failed to measure. It is that the measure has quietly replaced the thing it stood for.

Then you start switching. New channel, new product, new plan, because switching is the only move that feels like new information once the dashboard has stopped supplying any.

Bezos is the right study for this, and not for the reason he is usually cited. The interesting part is not that he waited. It is what was on the dashboard during the wait. Patience pointed at the wrong gauge is a longer stretch of not knowing.

Three moves you can steal

Count the output words in your own plan. Take whatever passes for a plan this quarter and mark every line naming a result you cannot act on directly: revenue, followers, funding, ranking. Then mark the lines with an owner and a date. Amazon's list ran to 452 items, reviewed several times a year, with goals added, modified and removed as it went. A written list somebody re-reads is also what stops scope drifting. It is hard to quietly acquire a sixth priority when the first five have owners.

Under each outcome you want, name the one input that moves it. Not a proxy for it. A thing a person does, countable, available weekly. If the outcome is revenue, the input might be conversations booked, or the days between a signup and first use. Pick one, put a target on it, and check that instead.

Write the reasoning down instead of presenting it. Amazon does not use slide presentations. The 2017 letter describes narratively structured six-page memos, read silently at the start of the meeting. A deck lets a weak argument pass as a bullet point. Prose has to hold together. Before your next decision, write one page: the input you are moving, why you believe it moves the outcome, and what you will see if it does.

Where the pattern breaks

The operating system generalises further than it should. In December 2024 OSHA and Amazon entered a corporate-wide settlement resolving ten ergonomics cases, the agency's first major multi-site investigation in over a decade, with an agreement requiring Amazon to assess ergonomic risk across its facilities and to implement controls. A system that measures what it is pointed at will move what it measures. Point it only at throughput and the people inside the loop become an output, the one category this discipline exists in order not to manage by. Note what the remedy was: another input metric.

What to do this week

Open the dashboard you actually check and write down what is on it. For each number ask one question: what did I do last week that could have moved this? If the honest answer is nothing, that number is weather, and it belongs on a monthly schedule rather than in your morning.

Then take one outcome you want this quarter and write a single input underneath it. Something a person does, countable, available every week. Conversations booked. Days between signup and first use. Orders shipped inside two days. One input, one target, one owner, and the owner is you.

Put it where you will see it on Monday. By Friday you will know whether you moved it, which is more than the old number told you all year.