Inside the Mind of Jeff Bezos

These pizza teams have created some of the site's quirkiest and most popular features. They conceived of the Gold Box, a little animated icon of a treasure chest that gleams and wobbles at the top of Amazon's home page. A click to "open" the box reveals special offers that last for just an hour from the time of the click.

Another popular innovation is Bottom of the Page Deals, daily bargains on staples such as Clif Bars and Tampax. Bezos wanted Amazon customers to think of the site not just for occasional discretionary items -- books, CDs, DVDs -- but also for daily essentials. The conceit was that the prices (often 50% off on brand names) were so low that they sank to the bottom of Amazon's home page. Patty Stonesifer, a former Microsoft executive who's now an Amazon director, recalls Bezos coming into a board meeting with a cardboard box of sundries, such as toilet paper and packs of batteries, and spilling them onto the table. Stonesifer thought the idea was "innovative, playful, and fun for customers." Days later, the deals were running on the Web site.

The Web makes it easy to try out ideas like the Gold Box and Bottom of the Page Deals, since you can measure results quickly and precisely. The risks and costs are low. What has really distinguished Amazon have been the big gambles, such as Search Inside the Book. This feature lets customers search through the full texts of books and read several pages at a time for free as a way to get them to buy. Search Inside debuted with an astonishing 120,000-plus books. Each one had to be scanned digitally and indexed, a huge logistical challenge at a huge cost. The database took up 20 terabytes, which Bezos says is about 20 times larger than the biggest database that existed anywhere when Amazon was founded.

But a large-scale launch was the only way to see whether it would go over with Amazon's 43 million active customer accounts. "If we had tried it in a tentative way on a small number of books, say 1,000 or 2,000, it wouldn't have gotten the PR and the customers' perception," says Risher. "There's an X-factor: What will it look like in scale? It's a big investment, and a big opportunity cost. There's a leap of faith. Jeff is willing to take those gambles."

Bezos himself explains his modus operandi this way: "For every leader in the company, not just for me, there are decisions that can be made by analysis." He's setting up for the punch line: "These are the best kinds of decisions!" The booming laugh comes on cue. "They're fact-based decisions. The great thing about fact-based decisions is that they overrule the hierarchy. The most junior person in the company can win an argument with the most senior person with a fact-based decision. Unfortunately, there's this whole other set of decisions that you can't ultimately boil down to a math problem." For those judgments, Bezos relies on his most senior executives, whom he regularly recruits from larger companies.

With decisions he deems to be fact-based, Bezos will make an extraordinary effort to study the numbers rather than rely on his team's best instincts and judgments. Amazon conducted a test to see whether it was worthwhile to advertise the brand on television. It picked two markets, Portland and Minneapolis, which blended together to approximate the customer base for the entire country. Then it did no national advertising for 16 months while showing ads only in the two target cities. That was an unusual move, since such tests, which aren't often done by anyone, rarely last for more than three months. "We were unbelievably fixated on it," Bezos says. While the ads did lift Amazon's sales somewhat, Bezos figured it wasn't enough to justify the costs, and instead he decided to put the money into offering lower prices. "That was a long, expensive test," he says, "but we were really determined to understand this for our company once and for all."

"It's one thing to be a data junkie who just looks at history. Jeff takes risks, and he changes and changes."

"One of Jeff's most recurrent phrases when someone has a good idea is, 'We can measure that,' " says Stonesifer. But, she adds: "It's one thing to be a data junkie who just looks at history, but Jeff takes a prospective view. He takes risks and he changes and changes."

Sometimes, Bezos says, you can't rely on facts because it would be too hard to test an idea, or too costly, or you can't figure out how to do it. And "sometimes we measure things and see that in the short term they actually hurt sales, and we do it anyway," he says, because Amazon managers don't think the short term is a good predictor of the long term. For example, they found that their biggest customers had such large collections of stuff -- especially CDs -- that they accidentally ordered items they had already bought from Amazon years ago. So they decided to give people a warning whenever this was about to happen. Sure enough, the warnings slightly reduced Amazon's sales. But it's hard to study the feature's long-term effects. Would it reduce sales over a 10-year period? They didn't think so. They thought it would make customers happy and probably increase sales. "You have to use your judgment," Bezos says. "In cases like that, we say, 'Let's be simpleminded. We know this is a feature that's good for customers. Let's do it.' "