Do it the Amazon Way!
DR. WERNER VOGELS, CHIEF TECHNOLOGY OFFICER & VICE PRESIDENT, AMAZON and the driving force behind the technologies that have powered some of the biggest business ideas of the last two decades e-commerce, e-books, the AWS Cloud and many others opens up about leadership, innovation and the trends that will matter.
In exclusive conversation with WOOL
WOOL:Going by your blog -"All Things Distributed", your weekend reading has a lot of fundamental research packed into it. Is that the academic in you?
WERNER VOGELS:At the size and scale of Amazon’s operations, a lot of theoretical assumptions of underlying algorithms become irrelevant. Whether we are talking about log based file systems or our Amazon Simple Storage Service (S3) we launched in 2006, real systems can fail in many ways theoretical models do not account for.
When Amazon engineers try and tackle these problems on scale, they need to use both practical approaches and theory. At the size and scale of our operations, with trillions of storage transactions processed in a day, the practical solutions can reach their limit.
But a lot of academic work has limitations too - they are written with a lot of assumptions. Unlike academics - real life systems don’t fail by coming to a stop, they fail by spitting out a lot of garbage. On Amazon scale, even the statistically improbable failures will happen once. And finally, failures are never uncorrelated in real life.
I was hired partly to bring an academic rigor to the problem solving, and going back to basics peels away the mass of assumptions that academics pile on to the original research. Looking at the problem that the original algorithm was designed to solve clarifies the clutter and often suggests a way forward for large scale problems. So my weekend reading often helps me solve work problems by going back to basics.
WOOL:Experimenting is core to Amazon’s DNA. But unlike any other firm, you are remarkable at scaling what works and moving on from your failures. What’s your secret sauce?
WV:Jeff Bezos, the CEO of Amazon, established the philosophy with the famous shareholder letter of 1997. I think the second principle there is about experiment, measure and learn. Amazon’s business philosophy has been built on experimenting continuously and learning relentlessly. And then scaling and making big bets on the ideas that succeed. So we celebrate experiments and failures because learning is valuable for the business.
There are two things that help us with this. The first is the technology infrastructure that runs on Amazon Web Services (AWS) - it helps us deliver new ideas quickly and painlessly, that lowers the cost of experimentation and failure.
The second is how the organization treats people involved in failures - we make sure they don’t have diminished career opportunities or a bad reputation. They don’t feel they have to keep extending failed experiments to prevent their careers from going down the drain.
Amazon’s business philosophy has been built on experimenting continuously and learning relentlessly. And then scaling and making big bets on the ideas that succeed. So we celebrate experiments and failures because learning is valuable for the business.
And now this has inspired one of our customers - ENEL, a utility based in Italy. They have started an internal TV station with a program called My Biggest Failure where engineers and project managers come and talk about things that went wrong.
At Amazon, we have innovation and experimentation that goes on at all levels, even in smaller teams. For instance, the recommendation engine works differently for different categories. A recommendation engine for books will not work for shoes. Because with a recommendation engine you want to create more opportunities for the customer to shop but minimize avoidable returns. In shoes - that can happen because a Jimmy Choo’s 8.5 size may be very different from a Valentino’s 9. But the returns on books have a different motivation. Each category’s team will use different data sources and use customer feedback in different ways to fine tune the recommendation engine.
And then you have the larger innovations, the bets which require significant capital and significant change in direction. As Jeff says, "The bigger the company is, the bigger your failure should be and also you should take the bigger risks simply because you can afford to do that."
When we started selling e-books in early 2000s, there were few takers for it. It was because reading e-books on laptops weren't a very good experience and tablets were new and expensive. But we believed there was a big business in e-books, so we built the Kindle and we sold it at cost, to prime the market. The original Kindle is museum piece now and the Gen 9 we have today is much more sophisticated, but I am very proud of what we did then, it was a big bet for us - we had no experience in hardware design.
Then we did the phone - not our proudest moment, but a lot of learning came from it. Alexa and Echo have benefited by our learnings in building that phone and integrating the assets we built. So the larger the company, it is inevitable that there will be bigger experiments and the bigger failures - because if you already know the outcome, it isn't an experiment. Real experiments will fail sometimes.
WOOL:How do you keep the employees motivated to build a culture of incessant innovation?
WV:First we need to hire the right people who are familiar with this culture. What we do at Amazon is build small teams of 10-12 people which are relatively independent of each other and of the organization. And in a team that size - they don't need to have meetings to find out what the others are doing. And teams this size don't need a hierarchy or constant direction.
So we have to hire people who will succeed in this environment. Our hiring process looks for evidence of our 14 leadership principles. Each interview aims to uncover evidence of those 14 leadership principles.
We have a process called "The Institutional Yes". How does that work? In most meetings, there is a tendency to kill off new ideas. The natural reaction when someone presents a new idea - someone already knows why it won't work. Sometimes it's just a dismissive hand wave that kills the idea.
Are you a missionary or a mercenary - we think people should be motivated by something other than just money. Do you have a bias for action - do you love to just talk about ideas or get around to doing stuff. You can be a great manager and a leader but if you can't deep dive, say if your team is writing code in SCALA which you don't know - how can you earn their respect. If you see a problem - do you fix it yourself or find someone who needs to fix it and make them do it - there is no other definition for "ownership". And of course you have to deliver results. (See the full list of the 14 principle here).
WOOL:Is there a tension between being obsessively customer-centric and innovating for tomorrow? Does it mean having to deal with conflicting priorities all the time?
WV:There is no conflict at all - all our innovation is for today and with keeping the customer first. Because if we stop innovating - we won't have a business in 10-15 years. There will always be someone who will figure out how to sell diapers better, another will figure out how to sell shoes better. It will be death by a thousand cuts. Innovation in small and big steps is part of our culture. We have a process called "The Institutional Yes". How does that work? In most meetings you must have been, there is a tendency to kill off new ideas. The natural reaction when someone presents a new idea - someone already knows why it won't work. Sometimes it's just a dismissive hand wave that kills the idea.
The Institutional Yes works differently, if someone has an idea that is good for customers, it's not easy to dismiss it with a hand wave. You have to write a four to six page justification for rejecting the idea before you are taken seriously, you can't wave a hand and dismiss a new idea. When someone takes the trouble to say in four pages why an idea is bad for customers or the company - then we sit up and take notice.
So that means we do things that are more adventurous. We once built something called "Your Digital Soulmate". It worked like this - we matched customers with others ( anonymously of course) who had identical purchasing patterns. We thought it was brilliant - you could predict what you will need to buy if your digital soulmate just bought it. Customers hated it from the bottom of their hearts. We are all unique in our ways - there is no one exactly like us.
In hindsight, we could have predicted that - by hiring a whole group of psychologists and waiting for all of them to agree about it. But that's the other thing Jeff wrote in his shareholder's letter. You can't wait for 100 percent of information and analysis to decide - you will be too late.
We think 70 percent of the information is good to get started if you have an idea - we move fast and get it into the hands of the customers. And it's Pareto efficient -because 80 percent of the detailed execution happens in the last 20 percent of the plan. So you have the ability to tweak your execution at the end - but you have the advantage of starting early based on part information.
Of course, ultimately your customers can tell you if it's a bad idea. Here is another one - we thought customers would love to buy more books if we extracted the unique phrases that appeared and recommended them on that basis. We called it the statistically improbable phrase. Did customers hate it? No - they found it cute. Did it sell many books? No! When we removed it after a couple of years - no one complained.
WOOL:Tell us about your strategy on Voice? With Alexa doing well and a whole lot of investments in voice related skill set - where are we going next with Voice?
WV:The interfaces we have today are driven by capabilities of our digital systems. Screens, keyboards and even the big innovations - the mouse and the touch screen are not the natural way we interact and have conversations. We can be fuzzy in our natural conversations and not very precise, but it works. Conversations and voice is the natural way we interact. As digital natives - we know exactly how to manipulate the question to the search engine get the exact answers we need. But not by asking a fuzzy question which is what we do in natural conversations and using voice. So while smartphones which provide voice enabled features started off as hands free interfaces for drivers, it was still only because we couldn't type while driving. But not really meant for continuous interactions, not designed for the natural fuzzy way we have conversations.
Alexa which is one of the first entries in the market and other advanced systems being built are letting people just talk, and you don't have to be that perfect in the way you say things but we can get the answer anyway, without using your hands.
We observed two groups of people who immediately flocked to Alexa. One is the youngsters, the others are elderly - say someone who is trying to communicate with their grandkids. They have tablets - but they still had to type which was not natural. Now you can talk to it and it does what you ask it to do.
When I am driving back home, I can tell Alexa to open the garage door, switch on the porch lights, play Red Hot Chili Peppers, set the temperature or even interact with WebMD to figure out what to do with a minor rash, all while having a conversation.
But still, even now, despite the interaction being natural - the back end is still hooked up to page based, digital systems that work on a linear approach. So how do you build the back-end systems very differently that support the conversational approach - that opens up exciting new possibilities.
Another driver is that a large part of the population in the developing countries, those outside the middle class cannot access the information they need easily, even if they have access to the internet - and giving them a smartphone is not a solution, but a voice interface might be.
Eventually we will have ubiquitous computing everything connected to the Internet. Using it effectively will make all the difference.
One of our customers - the International Rice Research Institute in Philippines, have information on 70,000 different strands of rice DNA. Since much of the rice is grown by smaller farmers -they have built a system for farmers who want to know which rice to grow, how much fertilizer to use but it would be useless with a web front end, as most of those smaller farmers don’t have a laptop or a smart phone. But they do have access to a village phone. So they built a voice interface for it - for Indonesian farmers who can go up to it and describe their land. The system uses the information, applies machine learning and gives them the answers. They can actually reduce the fertilizer amount by 90 percent and yet double the yield of rice. It's all completely automated, and completely on voice. It works on fuzzy information, can have a back and forth conversation till it understands.
You can see the potential of such systems - they might not change everything in the next two to three years but in the long run, they can have a big impact. Once you go beyond the constraints of the digital system and instead design for the environment, there are lots of possibilities. Why only voice? Johnson & Johnson brought out an Internet enabled toothbrush that tells you if you are brushing your teeth right. Rather than look at your smartphone, it can glow differently, or even vibrate. If you want to monitor energy consumption at home, looking at graphs on your phone may not change behavior - but if the clock in your home starts going red, that might be more effective in changing behavior.
We have a customer in New Zealand - an energy company which was trying to get their customers to pay their bills on time. Putting them on pre-paid doesn't work, sending them reminders by SMS is not effective - it was a difficult problem to solve. Then they put a lamp in the house which starts to glow brighter and brighter if you are overdue. That was effective - because nobody wants to be embarrassed.
Eventually we will have ubiquitous computing - everything connected to the Internet. Using it effectively will make all the difference.
WOOL:When you talk about voice, Alexa is currently only in German and English. What are your plans to expand into other languages to overcome barriers?
WV:Of course we are going to do that. I think there are 9606 languages in this world. Of which in the western world there are 203. And of these, text to speech systems supports 23 languages, 47 voices to speak. We may not get to all languages - certainly not all the 850 which are in New Guinea alone. But if we can move from human validation to automated validation by systems, then having the systems learn new languages could be much faster. That will also depend on whether we have enough recordings of these languages.
WOOL:What according to you are the next top technology trends?
WV: Security, Data and Artificial Intelligence.
SECURITY trends are dramatically changing. We have seen high profile hacking cases with respect to politicians or state sponsored activities. Also, the bad actors are aggressively progressing because of the incentives - their software has to only work for five minutes and they can hire more PhDs and pay more and they are not bound by international boundaries.
One of AWS’s strengths is that it is continuously evolving to find new ways to protect customers from operational point of view as well as innovating and building the new tools by which customers can protect themselves. Because without security - neither we nor our customers have a business. Take the Indian Government Initiative of UID - which provides a Unique Biometric ID to all citizens and a lot of services can be provided by businesses based on that. But we must ensure that the biometric information is secure and all businesses using that information have implemented the multiple levels of security for that service.
DATA - The Cloud has levelled the playing field for small and large companies with respect to computing capabilities and analytics tools. The difference will be how you use the tools. The quality of your data, the new data streams you can integrate will become the key. Tata Motors is instrumenting all their trucks on the road with data streams that go beyond location of the trucks. The data about the health of those trucks can be used for the preventive maintenance on those vehicles. Instrumenting your supply chain, instrumenting your factory floors - all of these things are generating new data streams and this is where service providers like Wipro can play an important role in coming up with quality data. Our customers cannot do these things by themselves but their partners can help with quality data which they can trust for making decisions related to efficiency or for new products being built or the adjustments to make for making factory floors safer.
ANALYTICS or Artificial Intelligence both make use of data from the past to make predictions about the future. So whether it is traditional machine learning, the stuff that we have been doing like making recommendations, inventory levels, price setting, detecting fraudulent orders or counterfeit goods all of these can be done based on data from the past. AI is moving into new things like voice recognition, image processing - all of these will be driven by advance software platforms as well as the hardware like the new GPGPU boards.
The Cloud has levelled the playing field for small and large companies with respect to computing capabilities and analytics tools. The difference will be how you use the tools. The quality of your data, the new data streams you can integrate will become the key.
WOOL:What is the best part in being a CTO of Amazon.com?
WV: You mean besides getting to travel to India? Being able to do things that nobody else is able to do! We are taking customer centric innovation to a new level - not being constrained by conservative thinking. Our shareholders understand that we are in it for the long haul and that’s how Jeff has set expectations. If we are successful in our big bets - look at AWS for example - they have a major impact on our balance sheet.