To our shareowners:
Something strange and remarkable has happened over the last 20 years. Take a look at these numbers:
The percentages represent the share of physical gross merchandise sales sold on Amazon by independent thirdparty sellers – mostly small- and medium-sized businesses – as opposed to Amazon retail’s own first party sales. Third-party sales have grown from 3% of the total to 58%. To put it bluntly:
Third-party sellers are kicking our first party butt. Badly.
And it’s a high bar too because our first-party business has grown dramatically over that period, from $1.6 billion in 1999 to $117 billion this past year. The compound annual growth rate for our first-party business in that time period is 25%. But in that same time, third-party sales have grown from $0.1 billion to $160 billion – a compound annual growth rate of 52%. To provide an external benchmark, eBay’s gross merchandise sales in that period have grown at a compound rate of 20%, from $2.8 billion to $95 billion.
Why did independent sellers do so much better selling on Amazon than they did on eBay? And why were independent sellers able to grow so much faster than Amazon’s own highly organized first-party sales organization? There isn’t one answer, but we do know one extremely important part of the answer:
We helped independent sellers compete against our first-party business by investing in and offering them the very best selling tools we could imagine and build. There are many such tools, including tools that help sellers manage inventory, process payments, track shipments, create reports, and sell across borders – and we’re inventing more every year. But of great importance are Fulfillment by Amazon and the Prime membership program. In combination, these two programs meaningfully improved the customer experience of buying from independent sellers. With the success of these two programs now so well established, it’s difficult for most people to fully appreciate today just how radical those two offerings were at the time we launched them. We invested in both of these programs at significant financial risk and after much internal debate. We had to continue investing significantly over time as we experimented with different ideas and iterations. We could not foresee with certainty what those programs would eventually look like, let alone whether they would succeed, but they were pushed forward with intuition and heart, and nourished with optimism.
Intuition, curiosity, and the power of wandering
From very early on in Amazon’s life, we knew we wanted to create a culture of builders – people who are curious, explorers. They like to invent. Even when they’re experts, they are “fresh” with a beginner’s mind. They see the way we do things as just the way we do things now. A builder’s mentality helps us approach big, hard-to-solve opportunities with a humble conviction that success can come through iteration: invent, launch, reinvent, relaunch, start over, rinse, repeat, again and again. They know the path to success is anything but straight.
Sometimes (often actually) in business, you do know where you’re going, and when you do, you can be efficient. Put in place a plan and execute. In contrast, wandering in business is not efficient … but it’s also not random. It’s guided – by hunch, gut, intuition, curiosity, and powered by a deep conviction that the prize for customers is big enough that it’s worth being a little messy and tangential to find our way there. Wandering is an essential counter-balance to efficiency. You need to employ both. The outsized discoveries – the “non-linear” ones – are highly likely to require wandering.
AWS’s millions of customers range from startups to large enterprises, government entities to nonprofits, each looking to build better solutions for their end users. We spend a lot of time thinking about what those organizations want and what the people inside them – developers, dev managers, ops managers, CIOs, chief digital officers, chief information security officers, etc. – want.
Much of what we build at AWS is based on listening to customers. It’s critical to ask customers what they want, listen carefully to their answers, and figure out a plan to provide it thoughtfully and quickly (speed matters in business!). No business could thrive without that kind of customer obsession. But it’s also not enough. The biggest needle movers will be things that customers don’t know to ask for. We must invent on their behalf. We have to tap into our own inner imagination about what’s possible.
AWS itself – as a whole – is an example. No one asked for AWS. No one. Turns out the world was in fact ready and hungry for an offering like AWS but didn’t know it. We had a hunch, followed our curiosity, took the necessary financial risks, and began building – reworking, experimenting, and iterating countless times as we proceeded.
Within AWS, that same pattern has recurred many times. For example, we invented DynamoDB, a highly scalable, low latency key-value database now used by thousands of AWS customers. And on the listeningcarefully-to-customers side, we heard loudly that companies felt constrained by their commercial database options and had been unhappy with their database providers for decades – these offerings are expensive, proprietary, have high-lock-in and punitive licensing terms. We spent several years building our own database engine, Amazon Aurora, a fully-managed MySQL and PostgreSQL-compatible service with the same or better durability and availability as the commercial engines, but at one-tenth of the cost. We were not surprised when this worked.
But we’re also optimistic about specialized databases for specialized workloads. Over the past 20 to 30 years, companies ran most of their workloads using relational databases. The broad familiarity with relational databases among developers made this technology the go-to even when it wasn’t ideal. Though sub-optimal, the data set sizes were often small enough and the acceptable query latencies long enough that you could make it work. But today, many applications are storing very large amounts of data – terabytes and petabytes. And the requirements for apps have changed. Modern applications are driving the need for low latencies, real-time processing, and the ability to process millions of requests per second. It’s not just key-value stores like DynamoDB, but also in-memory databases like Amazon ElastiCache, time series databases like Amazon Timestream, and ledger solutions like Amazon Quantum Ledger Database – the right tool for the right job saves money and gets your product to market faster.
我们也对用于特殊工作的专用数据库的前景感到乐观。在过去的20到30年间，企业使用关系数据库来处理大部分业务需求。开发人员对关系数据库的选择，主要取决于他们是否了解该技术，而不是技术本身是否优异。尽管没那么好，但因为数据集的大小通常很小，所以查询等待时间还是在可接受的范畴。但是今天，许多应用程序的数据量级在TB和PB，而且需要满足低延迟、实时处理以及并发处理的需求。每秒处理数百万个请求是常见的需求。它需要是DynamoDB这样的键值数据库，而且也要是Amazon ElastiCache这样的内存数据库、Amazon Timestream这样的时间序列数据库、Amazon Quantum Ledger Database这样的分类账解决方案－用上正确的工具可以节省金钱，提升产品推向市场的速度。
We’re also plunging into helping companies harness Machine Learning. We’ve been working on this for a long time, and, as with other important advances, our initial attempts to externalize some of our early internal Machine Learning tools were failures. It took years of wandering – experimentation, iteration, and refinement, as well as valuable insights from our customers – to enable us to find SageMaker, which launched just 18 months ago. SageMaker removes the heavy lifting, complexity, and guesswork from each step of the machine learning process – democratizing AI. Today, thousands of customers are building machine learning models on top of AWS with SageMaker. We continue to enhance the service, including by adding new reinforcement learning capabilities. Reinforcement learning has a steep learning curve and many moving parts, which has largely put it out of reach of all but the most well-funded and technical organizations, until now. None of this would be possible without a culture of curiosity and a willingness to try totally new things on behalf of customers. And customers are responding to our customer-centric wandering and listening – AWS is now a $30 billion annual run rate business and growing fast.
Imagining the impossible
Amazon today remains a small player in global retail. We represent a low single-digit percentage of the retail market, and there are much larger retailers in every country where we operate. And that’s largely because nearly 90% of retail remains offline, in brick and mortar stores. For many years, we considered how we might serve customers in physical stores, but felt we needed first to invent something that would really delight customers in that environment. With Amazon Go, we had a clear vision. Get rid of the worst thing about physical retail: checkout lines. No one likes to wait in line. Instead, we imagined a store where you could walk in, pick up what you wanted, and leave.
Getting there was hard. Technically hard. It required the efforts of hundreds of smart, dedicated computer scientists and engineers around the world. We had to design and build our own proprietary cameras and shelves and invent new computer vision algorithms, including the ability to stitch together imagery from hundreds of cooperating cameras. And we had to do it in a way where the technology worked so well that it simply receded into the background, invisible. The reward has been the response from customers, who’ve described the experience of shopping at Amazon Go as “magical.” We now have 10 stores in Chicago, San Francisco, and Seattle, and are excited about the future.
要实现真的很难，技术上很难。需要全球数百名聪明敬业的计算器科学家和工程师的努力。必须设计和制造专用相机和架子，发明新的计算器视觉算法，合并数百个协作相机中的图像。而且必须以一种运行良好的方式实现，让技术退入背景，客户看不见也无须看见。成就感来自于客户的响应，他们称Amazon Go的购物经历为「神奇」。我们在芝加哥、旧金山和西雅图有10家Amazon Go，我们对未来感到兴奋。
Failure needs to scale too
As a company grows, everything needs to scale, including the size of your failed experiments. If the size of your failures isn’t growing, you’re not going to be inventing at a size that can actually move the needle. Amazon will be experimenting at the right scale for a company of our size if we occasionally have multibillion-dollar failures. Of course, we won’t undertake such experiments cavalierly. We will work hard to make them good bets, but not all good bets will ultimately pay out. This kind of large-scale risk taking is part of the service we as a large company can provide to our customers and to society. The good news for shareowners is that a single big winning bet can more than cover the cost of many losers.
Development of the Fire phone and Echo was started around the same time. While the Fire phone was a failure, we were able to take our learnings (as well as the developers) and accelerate our efforts building Echo and Alexa. The vision for Echo and Alexa was inspired by the Star Trek computer. The idea also had origins in two other arenas where we’d been building and wandering for years: machine learning and the cloud. From Amazon’s early days, machine learning was an essential part of our product recommendations, and AWS gave us a front row seat to the capabilities of the cloud. After many years of development, Echo debuted in 2014, powered by Alexa, who lives in the AWS cloud.
Fire Phone和Echo的开发始于同一时期。Fire Phone失败后，我们（包括开发人员）吸取教训，因此加快了Echo和Alexa的工作。Echo和Alexa的构想受到《Star Trek》中计算器的启发。这个想法起源于我们已经耕耘多年的另外两个领域：机器学习和云技术。 从Amazon成立之初，机器学习就成为产品推荐功能必不可少的一部分，而AWS使我们在云技术领域处于领先位置。经过多年的研发，Echo于2014年首次亮相，并可使用透过Echo使用Alexa服务。
No customer was asking for Echo. This was definitely us wandering. Market research doesn’t help. If you had gone to a customer in 2013 and said “Would you like a black, always-on cylinder in your kitchen about the size of a Pringles can that you can talk to and ask questions, that also turns on your lights and plays music?” I guarantee you they’d have looked at you strangely and said “No, thank you.”
Since that first-generation Echo, customers have purchased more than 100 million Alexa-enabled devices. Last year, we improved Alexa’s ability to understand requests and answer questions by more than 20%, while adding billions of facts to make Alexa more knowledgeable than ever. Developers doubled the number of Alexa skills to over 80,000, and customers spoke to Alexa tens of billions more times in 2018 compared to 2017. The number of devices with Alexa built-in more than doubled in 2018. There are now more than 150 different products available with Alexa built-in, from headphones and PCs to cars and smart home devices. Much more to come!
One last thing before closing. As I said in the first shareholder letter more than 20 years ago, our focus is on hiring and retaining versatile and talented employees who can think like owners. Achieving that requires investing in our employees, and, as with so many other things at Amazon, we use not just analysis but also intuition and heart to find our way forward.
Last year, we raised our minimum wage to $15-an-hour for all full-time, part-time, temporary, and seasonal employees across the U.S. This wage hike benefitted more than 250,000 Amazon employees, as well as over 100,000 seasonal employees who worked at Amazon sites across the country last holiday. We strongly believe that this will benefit our business as we invest in our employees. But that is not what drove the decision. We had always offered competitive wages. But we decided it was time to lead – to offer wages that went beyond competitive. We did it because it seemed like the right thing to do.
Today I challenge our top retail competitors (you know who you are!) to match our employee benefits and our $15 minimum wage. Do it! Better yet, go to $16 and throw the gauntlet back at us. It’s a kind of competition that will benefit everyone.
Many of the other programs we have introduced for our employees came as much from the heart as the head. I’ve mentioned before the Career Choice program, which pays up to 95% of tuition and fees towards a certificate or diploma in qualified fields of study, leading to in-demand careers for our associates, even if those careers take them away from Amazon. More than 16,000 employees have now taken advantage of the program, which continues to grow. Similarly, our Career Skills program trains hourly associates in critical job skills like resume writing, how to communicate effectively, and computer basics. In October of last year, in continuation of these commitments, we signed the President’s Pledge to America’s Workers and announced we will be upskilling 50,000 U.S. employees through our range of innovative training programs.
我们为员工推出的许多计划。我之前提到过职业选择计划，此计划预付最高95％的学杂费，让员工可以取得学习领域的证书或文凭，从而给他们尝试新职业的机会，即使这些职业会不在Amazon。现在已有超过1.6万名员工从此计划中受益且人数持续。同样的，我们的职业技能计划（Career Skills）对员工进行关键工作技能的短期培训，例如简历写作、有效沟通和计算器基础知识。去年10月，为兑现这些承诺，我们签署了《President’s Pledge to America’s Workers》，宣布将透过一系列培训计划，提高5万名美国员工的技能。
Our investments are not limited to our current employees or even to the present. To train tomorrow’s workforce, we have pledged $50 million, including through our recently announced Amazon Future Engineer program, to support STEM and CS education around the country for elementary, high school, and university students, with a particular focus on attracting more girls and minorities to these professions. We also continue to take advantage of the incredible talents of our veterans. We are well on our way to meeting our pledge to hire 25,000 veterans and military spouses by 2021. And through the Amazon Technical Veterans Apprenticeship program, we are providing veterans on-the-job training in fields like cloud computing.
我们的投资不仅限于现有员工，甚至不局限于当下。为了培训明天的劳动力，我们已承诺投资5000万美元，包括Amazon未来工程师计划，这是一项支持全国小学、高中和大学生STEM和CS教育的计划，将会特别着重于吸引更多女孩和少数民族。我们将继续招聘退伍军人，让他们发挥绝佳的才能。我们正在努力实现承诺，在2021年前雇用2.5万名退伍军人及其配偶。透过Amazon的退伍军人技术培训计划（Technical Veterans Apprenticeship），我们将在云计算等领域提供退伍军人在职培训。
A huge thank you to our customers for allowing us to serve you while always challenging us to do even better, to our shareowners for your continuing support, and to all our employees worldwide for your hard work and pioneering spirit. Teams all across Amazon are listening to customers and wandering on their behalf!
As always, I attach a copy of our original 1997 letter. It remains Day 1.
Jeffrey P. Bezos
Jeffrey P. Bezos
Founder and Chief Executive Officer
[…] 第二十二封信：非线性发明来自漫步 […]