How Bezos keeps a big company fast

A large company can look healthy for years while it gets slower. Jeff Bezos used his 2016 letter to shareholders to describe that slide, and the habits he wanted Amazon to keep so a big company could still move.

The letter starts from a question at an all-hands. "Jeff, what does Day 2 look like?" He had been telling people it was Day 1 for a couple of decades, and he worked in a building of that name. When he moved buildings, he took the name with him. He answers with a sequence: Day 2 is stasis, followed by irrelevance, then an excruciating, painful decline, then death. He adds the part a board can miss. The decline happens in extreme slow motion. An established company might harvest Day 2 for decades, and the end still comes.

He then asks how you keep the vitality of a first day inside a company that is already large. He offers four habits: customer obsession, a skeptical view of proxies, eager adoption of external trends, and high-velocity decision making. Each one is specific enough to use in a meeting this week.

Stay with the customer who says they are happy

Bezos lists the other centres a company can choose: competitors, the product, the technology, the business model. In his view, an obsessive focus on the customer is the habit that best protects Day 1 energy. Customers, he writes, are always beautifully, wonderfully dissatisfied, even when they report being happy and the business is great. Even when they do not know it yet, they want something better, and a company that wants to delight them will invent on their behalf.

Prime is the example. No customer asked Amazon to create the membership. They wanted it once it existed. A team that waits to be asked will keep polishing the current offer. He is explicit about the day-to-day cost of the other path: patient experiments, room for failures, protection for the small bets, and a willingness to double down when customers are clearly delighted.

If you own an offer, write a name and a date before the next review. Who on the team last sat with a customer, and what did that person want that the current product leaves unmet? Put the satisfaction score under that note. Prime is there to remind you that the want may be one nobody has asked the customer to design.

Doubt the number standing in for the customer

As companies get larger, Bezos says, they manage to proxies. He calls it dangerous, subtle, and very Day 2. Process is the case he spends the most time on. A good process exists so you can serve customers. Stop watching, and the process becomes the thing. People check that the steps were followed and skip the outcome. He describes a junior leader defending a bad result with "Well, we followed the process." A more experienced leader treats that sentence as a reason to investigate and improve the process. The question he wants asked is a direct one. Do we own the process, or does the process own us?

When a bad outcome shows up, listen for that sentence. If it closes the discussion, ask which step produced the outcome and change that step.

Surveys are his second proxy, and he is wary of them when a team is inventing. He pictures the update in which 55% of beta testers are satisfied with a feature, up from 47%, and he says that movement is hard to interpret and can mislead. Good inventors, in his account, develop an intuition for the customer. They study anecdotes and live with the design. He still wants beta tests and surveys. He wants them in the hands of an owner who already understands the customer, has a point of view, and loves the offer, so the research can find a blind spot. A remarkable customer experience, he writes, starts with heart, intuition, curiosity, play, guts and taste. "You won't find any of it in a survey."

Take the outside trend while everyone can already see it

The outside world, he argues, will push a company into Day 2 if it refuses a powerful trend. Fight the trend and you are probably fighting the future. Adopt it and you have a tailwind. Big trends are easy to spot, because people talk and write about them constantly, and strangely hard for large organisations to embrace.

In 2016 the trend he names is machine learning. For decades, computers had automated tasks a programmer could describe with clear rules. Machine learning was starting to handle tasks where those rules are much harder to write. He separates the visible work from the quiet work. At Amazon the visible pieces were Prime Air delivery drones, the Amazon Go store, which used machine vision to remove the checkout line, and Alexa. Under that, machine learning was already driving demand forecasting, search ranking, recommendations, merchandising, fraud detection and translations. He expected much of the impact to look like those quiet improvements to core operations.

Name the trend your industry is already discussing in public, then name the operation it should sit inside this quarter. He pairs drones and Alexa with forecasting, search and fraud. If your version of the trend only appears in a lab presentation, it is still at the spotting stage.

Decide while you still wish you knew more

Day 2 companies, he writes, make high-quality decisions, and they make them slowly. Day 1 means high-quality decisions at high speed. He says that comes easily to a startup and with real difficulty to a large organisation. The senior team at Amazon was determined to keep the velocity up. Speed matters, he writes, and a fast decision environment is more fun.

He offers four tactics.

First, stop running every decision through one process. Many decisions are reversible, two-way doors, and those can use a light process. If the call is wrong, you come back through the door. Label the decision in the pack. If you can undo it, name the person allowed to walk it back, and let them do it without waiting for the monthly forum.

Second, make most decisions with somewhere around 70% of the information you wish you had. Wait for 90% and, in most cases, you are being slow. The duty that makes 70% safe is the correction. You have to notice a bad call quickly and change course. If you can do that, a wrong call can cost less than a late one. Being slow, he says, is going to be expensive for sure. Write three things into the approval: the date you will look again, the figure that would make you stop, and the name of the person allowed to stop it.

Third, use "disagree and commit." If you have conviction and the room has no consensus, say so, ask the others to gamble with you, and take the commitment. By then nobody knows the answer for sure, and you will often get a quick yes. Bosses have to do it too. He says he disagrees and commits all the time.

The example is an Amazon Studios original. He told the team the show was debatable as entertainment, looked complicated to produce, had poor business terms, and was competing with plenty of other opportunities. They wanted to proceed. He wrote back the same day: "I disagree and commit and hope it becomes the most watched thing we've ever made." Forcing the team to convince him would have slowed the cycle. He is precise about the note. It was a genuine disagreement, said in the open, with a chance for the team to weigh his view, and then a sincere commitment to go their way. That team had 11 Emmys, 6 Golden Globes and 3 Oscars, and he was glad they let him in the room.

Fourth, separate a disagreement you can commit through from a true misalignment. Sometimes two groups want different things, and further meetings leave the gap where it was. If nobody escalates, the resolver is exhaustion. Whoever can endure more meetings carries the decision. "You've worn me down" is, in his words, an awful way to decide, slow and de-energizing. He wants the escalation to be quick.

His example is the choice to let third-party sellers compete directly against Amazon on Amazon's own product pages. Capable, well-intentioned people were deeply against it. The big decision created hundreds of smaller ones, and many of those had to reach the senior team. They escalated those smaller calls to the senior team, which is what he recommends once stamina is the alternative on offer.

What to put in the next pack

Bezos closes on the standard he wants. A company can have the scope and capabilities of a large one and the spirit and heart of a small one. It has to choose that. The choice shows up in who may decide, how complete the file must be, and whether a reversible bet can start before the forum sits again.

If you chair the meeting where new offers wait, the next pack can carry a shorter list drawn from the letter.

1. Name the customer and the last time someone on this team sat with them. Write what they wanted that the current offer leaves unmet. Prime is the reminder that they may never have asked for the thing you should build.
2. If the case rests on a completed process or a satisfaction score, say so at the top. A move from 47% to 55% is the kind of figure he thinks is easy to misread. Put an anecdote beside it.
3. Mark the call as a two-way door or a one-way door. For a two-way door, name the person who can reverse it and the date they will look.
4. State how much of the picture you have. If you are near 70% of what you wish you knew, and the missing piece would leave the call unchanged, decide in the meeting.
5. If you disagree, put the disagreement and the commitment in the same note, and release what the team needs. The Studios note is the model: his doubts, then the commitment, the same day.
6. If two functions are paid for opposite outcomes, escalate this week and name the senior owner who picks.

Our lean startup and product sprint workshop is where a team takes one live idea through a thin test and a decision, with the owner in the room. Use it on a two-way door that is waiting for a forum. The letter's standard is a reversible call decided in the week it shows up.