 Well, good afternoon, and it's great to be here at Slush with Verna. Verna, the first question I have to ask is, what's this t-shirt saying? For those of you who don't have great eyesight, it says... Encrypt everything. Yeah, I think if you look at sort of all the events that have happened in the past years around data breaches and things like that, I think everybody, especially when you're a young company, when you start up, mostly from day one, have to start thinking about how to protect your customers. You know, without protecting your customers and your business, you don't have a business. And so I think something that sort of most startups have sort of delayed till later on in the process is thinking about security, and thinking about privacy, thinking about what are the kind of data that you are actually collecting about your customers. And I think you have an obligation to reach your customers to protect them. And I think the one and only tool you have to really ensure that you're the only one who has access to it is encryption. So at Amazon and AWS, the Cloud Computing Division of Amazon, we're very strong believers that you should encrypt everything, all the data that you have, so you can never make a mistake. You never have to think about is this... Is this person identifiable information or not? If you encrypt everything, you're the only one who has access to it. So take this at heart. If you're using the Amazon Cloud, AWS, all services have encryption enabled in them, and we give you also the capabilities to manage your own keys so that you're the only one who has access to your data. Not everyone clearly shares that view, particularly in government. They want access to certain amounts of information. Where are you in the whole end-to-end encryption debate? Yeah, I'm a strong believer. At Amazon, my strong belief is that backdoors will eventually be discovered by anyone, and so you might as well open things up anyway. So we're strong believers that backdoors should not be put into encryption. It's really a conversation between the persons who own the data and whoever would like to have access to it, maybe for very valid legal reasons, but we prefer not to be in the middle of that. And so if you encrypt your data, you determine who has access to your data and who not, whether it's governments or law enforcement or your competitors. Now, your chief technology officer is one of the most fascinating companies in the world. To what extent is your job just making sure things don't go wrong, and to what extent is it trying to figure out what the next technology is that you can absorb and use at Amazon? Well, in that sense, making sure that nothing goes wrong I think is everybody's job at Amazon, not just mine. I think at the scale that Amazon appropriates at, it is clear that we hire the absolute best engineers in the world and in a search, I don't need to oversee them. They really are in charge of their services themselves and whether it is the thousands of services that we have in Amazon retail or all the cloud computing services that we have, it is everybody's job to do that. And I've always repioneered something that now has become much more popular in terms of the DevOps world, where basically the developers also do operations. And the history of that is that if you would have an operations department that runs your software, developers see depth as their customers. And we always had the idea that developers should be in direct contact with their customers. So we had this principle that you build it, you run it. And in a search, everyone is responsible for the interaction with their customers in a search also for the reliability. OK. Now, I wondered if you could also talk a bit about some of the specific technologies that you're using at Amazon. Let's start with voice. Clearly that's a big deal for Amazon at the moment and the whole interface that you have with Alexa. Where is this voice revolution going to go? If we roll the film forward five, ten years, how are we going to be interacting with voice? Well, I think first of all, I think that this doesn't only is not an Amazon approach to things. I think what we've been looking for in the past years is if we can find way more natural interfaces to our digital systems. If you look at how things are built now, it's really driven by the capabilities of the machinery itself. Screen, keyboard, mouse, finger, all of these things are driven by how this machine works, not how we as humans like to communicate. And when sort of the hardware advances and the software advances in our machine learning in the past years really took off, we certainly could be able to do sort of voice processing both in terms of automatic speech recognition and natural language understanding and text-to-speech in a real-time manner, because that's important around voice. If you use voice, it needs to have a natural interface to it. That means that responses need to come back within a second, otherwise it doesn't sound like that. So being able to have the inference engines run at such speed that you can now do a natural conversation, certainly the whole world changed, because it's such a natural approach to doing things. This is not a Slack channel or a WhatsApp group that we're all sitting in here, we're talking. And so I truly believe that voice and I would say human-centric interfaces will be the access to digital systems in the future. And mostly because it also unlocks our digital systems for much larger groups of the population that are now not capable of using it. And whether that is elderly people, or whether it's young people, or whether it's people in developing countries that don't have access to smartphones or internet or things like that, much of a very large part of the world will, their first interaction with digital systems will not look like digital systems at all. They will just look like voice, speech, and as such nobody will know that those are digital systems. And to what extent are they going to become interoperable so that you're using your Alexa at home that that personality, as it were, transfers to the car or to your office? Are we just going to be surrounded by our virtual Alexa? We already see quite a bit of that. We literally have hundreds of manufacturers that have really integrated Alexa into their system. You have to remember that the Echo devices that we first launched as sort of the digital voice assistant are only one of the versions that actually you can interact with. I think really you have solar speakers at home that have Alexa in them. GE has their fridges with Alexa in them. You can ask your fridge what's still inside. And so, or you have a few microwaves saying this is a cup of coffee. I want to hear it to this degree. Twenty of the top car manufacturers have integrated Alexa into their cars. So voice will become ubiquitous. But it will be very different from how we used to have it in our cars. In that setting it was much more command and control. Do this. Call that person. And it's not at all having a fuzzy conversation with multiple layers that you can have with systems like Alexa. I'd like to move on to cloud computing, which is obviously a very big thing for Amazon. To what extent has it changed the rules of business? I mean, I was talking to an AI company here earlier. They were making the point that this is perhaps one of the vast unappreciated revolutions in business history. Would you go that far? Yeah, absolutely. I think so. No, the good point there is, especially if you think about younger businesses, there's been a study and a report by Harvard Business Review around sort of the impact that AWS has had, which is the cloud computing division, on the funding circuit. And they claim the launch of AWS has been directly responsible for a two to three X increase in the number of companies that are being funded at the angel level. So basically a massive increase in young businesses being created. The A and B circuits still has remained the same. So basically many more companies get a chance to prove that they have a viable approach in their business. And then you still see the normal funding circuits happening after that. But it's clear that the rise of startups in the world has been completely driven by cloud computing. And if you ask any of the, I mean, I'll consider it to be sort of home names in every place, whether it's Airbnb or Dropbox or Spotify or Uber or Supercell or Klarna or whatever. All of these household brand names today, all grown up from the cloud. And if you would ask them, they would never have been able to get off the ground. Because one of the parts that actually increasing the number of companies that receive angel funding also meant that many more high risk investments have been done. And those would have been companies that have wild ideas and where in the past nobody would be willing to put five or 10 million into just into an idea. But $50,000 or $100,000 is something that's easily get off the ground. And so many more high risk operations have been able to sort of get off the ground and prove their right to existence that would never have received funding in the past. And as you're saying, this is enabled a whole new slew of businesses to emerge. I mean, some of them like Netflix and Spotify have been built on the cloud, which raises an interesting dynamic because they obviously compete with you or Amazon Prime. Yeah. How is this new kind of ecosystem going to develop? Well, yeah. And it's not just in, let's say, the Netflix is of the world. But take, for example, Zalando in Western Europe being one of the bigger e-commerce companies run on exactly the same platform as that Amazon runs on. And I think for most of these companies, their thinking is that they should not be spending time and effort on things that are not their core expertise. And so I think, again, there's a great paper a long time ago by Nick Carr, also in half a business group, it's called IT Doesn't Matter. And he didn't mean that IT Doesn't Matter. He meant that if everybody has to do exactly the same thing, it is not a competitive differentiator. And any amount of effort and intellectual, whether it's financial capital or intellectual capital that you need to pour into there is a wasted effort because your competitors do exactly the same thing. And so you might as well start using platforms, in this case, from a competitor, that really allow you to focus on building the better products. And by the way, the competition between Netflix and Amazon Video is not on who has the best infrastructure, it is who is the best programming. And so any effort you can put into that makes you more competitive. And so the same goes for whether it's e-commerce or any other area that Amazon is active into where many of our competitors will be using the AWS platform because it actually allows them to focus better on their customers. Right. I'd like to move on to machine learning. In what ways does Amazon use machine learning and how is it making your business better? Well, Amazon has been using machine learning for the past 25 years. We just never told you about it. And I think this is something that anyone that actually has significant amount of data whether it's about your customers, whether it's about your operations, whether it's about sort of the safety and or the factory floors. With these large data sets, you can discover patterns that you would not be able to do anywhere else. And so if you ever visited the Amazon front page, you'll notice that everyone gets their own personalized environment. There is no one Amazon page. If you recognize you, you get a different page. And so as such, the whole personalization, whether there is recommendations or similarities and those kind of things, are all areas that Amazon has pioneered. And so these things didn't exist 25 years ago. It's all things that Amazon has come up with over time. And those are just the customer-facing things where we help customers make better buying decisions. But if you think about sort of all the things you do under the covers or counterfeit good detection, for example, or fraud detection, we literally sit on billions of orders. We know which ones of those are fraudulent. And we know which patterns of fraud changed over time. So you can build machine learning models where you can ask a question to if you get a new order in. What is the likelihood that this is a fraudulent order? And so if it then reaches a threshold, you still kick it off to a human to investigate, but it tremendously helps having a lot of data from the past to be able to make smarter decisions right now. And so what has happened for, let's say, about five, six years ago at Amazon is that we were a little bit bummed that we weren't making faster progress with machine learning. And mostly that was because machine learning was done by data scientists. And there's just not enough data scientists in the world. And most of the processes are actually pretty bothersome. There's a lot of heavy lifting around it. Yet there is quite a very large number of machine learning problems or data sets that could be processed if it could be just engineers doing it. And so we decided to build services and platforms first internally to make sure that every engineer can become a data scientist or can become a machine learning engineer. Because in the end, for most of these cases, this picking algorithm and do the training and build a model merge more than figuring out the algorithm by itself. And so we're giving support now. We have a multi-layer support there where a monitor of data scientists still creating new algorithms. But then there's something called SageMaker which allows every engineer in the world to become a data scientist basically. And then we have another set of services on top of that which we have a core AI services which are really pre-built models out of the machine learning world that helps you do image recognition and video processing and automatic translations. By the way, translate is available for finish as well if you want that. So comprehend understanding text. For example, we just launched Comprehend Medical that is specifically focused on all the handwritten stuff that you see in hospitals and things like that. To turn that into coherent medical texts. So the messages stay out of the way of the engineers at Amazon? Well, every engineer in the world because we want SageMaker now for everyone to use and it is quickly becoming the platform of choice for anyone that wants to do machine learning. There is so much heavy lifting around machine learning. Basically, some of the machine learning is actually really brute force, almost dumb. Basically, you do training over and over and over again with a different set of parameters until you get an accuracy that you like. Basically, you try to predict the past. And so that's really brute force. So most of this stuff will be hidden around the corner under the covers. You have your data set. You make sure it's a label right. You pick an algorithm. You do the brute force training. You wait till it's finished. The accuracy shows up. And then with one click of a button, you can create a model and deploy it. So much of the heavy lifting is taken away not only for the Amazon engineers, but for every other engineer in the world. So, anecdotally, whenever I use Amazon and I look at the recommendations, a lot of them, I think, are kind of way off beam. Tell me what the data is. I mean, how effective are your recommendation engines? I like to believe we're pretty effective. We, however, show you lots of recommendations. So some of those will have higher accuracy and some of them will be off. And there's always something more to improve. But we're also very conscious, very conscious about the privacy of our customers, about what kind of data we collect. And we really try to keep that to a minimum and only collect that data that helps us make you better buying decisions. And definitely not reusing your data to help other people make buying decisions. So, as such, data sets don't get as large as that you would like to, but we still are pretty effective. By the way, if you think about recommendations, we just launched Amazon Personalized. One of the new cloud services that we built that comes out of the Amazon retail world where anybody can now integrate machine learning-based recommendation engines into their products. We've heard some great talks on this stage about blockchain and its impact. Where's that going to go at Amazon? What do you think of that? So, I look at it mostly from a cloud computing division. And to most of the customers that we've been talking to, they still see blockchain as incredible technology still in looking for an incredible problem to solve that you can only solve with blockchain. And this is just sort of beyond cryptocurrency. And I'm a distributed systems researcher by nature, so I love this decentralized stuff. It is just that our current engineering practices favor other techniques that we have very well understood. And so, when we're talking to our customers and asking them what they were looking for at a blockchain, there are actually two different answers that came back. One of them is we want an immutable ledger with cryptographic proof that this is immutable. Sort of a centralized trust. But the much smaller part we're actually looking for a decentralized solution and there aren't that many of them that are actually looking for that. So, we just launched QLDB, a ledger database which you can build. It's a cryptographic proof in an immutable ledger database. And I think that will actually address most of the cases that many of our customers that were originally interested in blockchain are actually looking for it. And then also we have a managed blockchain service that supports those who really want to do decentralized operations. One of the things that really fascinates me about Amazon is this culture of experimentation and innovation. And I mean, Jeff Bezos has always said that Amazon has a day one mentality and he never wants to have a day two mentality. I mean, that's fine if you're some of the entrepreneurs in the audience here who really are on day one. But how on earth do you keep that mentality when you get to be so vast as Amazon? Well, the realization at Amazon comes that if we stop innovating, we'll be out of business in 10 to 15 years. And not because there's another large Amazon stepping up. It's really death by a thousand cuts. Now someone will be doing diapers better. Someone will be doing shoes better. Someone will be doing electronics better. And customers will have access to perfect information. They know exactly where they get the best deal and whether the best deal is finding the right products that you're looking for, whether it is the lowest price because no one will say, oh, I love Amazon, I wish them were more expensive. Or just having convenience, having good guarantees on shipping. So customers know exactly where to get the absolute best deal. And so if we stop innovating, there's a strong belief that this company will go downhill pretty quickly. How does that play out on a day-to-day basis? Because we started our conversation by you saying you're the guy who's responsible if things get wrong. So how do you balance things not going wrong on the one hand and innovating very fast on the other? Well, if you do innovation at the high neck-breaking speed, some things will go wrong. And we have a culture around what we call experimentation. And remember, an experiment is not an experiment if you really know the outcome. There needs to be a certain level of uncertainty that goes with doing experiments. So we want to make sure at all times that we have a pipeline of experiments lined up. And we do this by having actually a really decentralized organization internally. Let's say in retail, maybe about a thousand services out of which the whole retail experience is being built out of. And each of those teams that manage these services have a challenge to innovate each year, to build their own innovation area. And so they don't need to go up in the chain in sort of responsibility to be able to execute this. So we have a continuous pipeline to make sure that we have these experiments. And we also remove every possible barrier for these experiments to go into this pipeline. For example, we have something which is called the institutional yes. This is a process within Amazon where if you've ever sat in a room with someone proposing something new, there's always someone in that room that says, oh, that won't work. Or our customers won't like it. As someone who proposes something new, you have to do all the work. We turn that around. If you want to block something, you have to do the work. It takes care of 95% of all objections. If someone is objecting, you better pay attention to them. So it does mean we have a pipeline of experiments always filled. It does mean that quite a few of those will fail because customers don't like it or there's no viable business for it. But that's a good thing. However, with a culture around fast failure, also comes if, first of all, you have to make it cheap from a technology point of view. Now, we have AWS, so if something isn't really popular or not becoming really the good path we want, probably haven't spent much money on it either. But you also have to make sure that teams that actually work on programs or new products that fail, that they don't actually get hampered in their career opportunities later on. I mean, many companies out there, traditional companies, have really shunned away from failure because, you know, you basically kill your career. And I mean, in Amazon, it becomes a badge of honor because there's something actually that goes together with experimentation is that learning. Experiment, measure, learn. And so even if things fail, if you have learnings that come out of that, that's a great approach. And sometimes it's easier to do these experiments than actually sort of sit in a room and talk about it for a long time. There's one of the things that another principle within innovation in Amazon is that in many of these experimental areas, it's hard to gather all the information you need before you can get started. And so if you wait till you have 100% of the information, you're probably too late. You might as well start when you do it 70%. You know, quite a few decisions are two-way doors. Basically, you can always back out. And so if you are conscious of that and you continuously look at what new information is flowing in and know where you have to maybe take a step back to take two steps forward again, that's a process that really works well within Amazon. All right. Just between us, Werner, what's the worst mistake you've ever made at Amazon and what did you learn from it? Well, okay. Probably mistakes are always mistakes of omission. If I look at some of the things that we've been doing recently in AWS, for example, I wish we'd done them five years ago. You know, these were things that our customers were asking for five years ago that we should have acted on that at that particular moment. So not moving fast enough or just not doing it is something that I think something always holds us. And next to that, we've made mistakes. I mean, in the earlier days of AWS, we did something where basically your account was also your identity. And it turns out that these two things are very separate. An account is something that you build to and identity is a security construct. It took us a long time to actually start sort of pulling these things apart. And so it was not necessarily an omission thing there. It's much more that in the earlier days we had no idea how this would evolve and that we need to sort of really iterate and make sort of smaller building blocks that's easier to tease things apart. One of the things Amazon is that you have a real kind of global view on what's going on in the technology industry. You're originally from the Netherlands. And so I'd be fascinated to get your take on how do you look at Europe in the kind of global tech world? What does Europe do well? What does it do badly? I like to separate the younger business so that the startup internet scale. Just look at the Nordics. What is it? Skype, Spotify, Klarna, Rovio, Supercell, and a whole range of other unicorns all coming out of Northern Europe. By the way, it has just been explained to me that's because if you only have one hour of daylight you'd have a lot of time to spend on programming. I don't know whether that's true or not, but it's clear that the Nordics are a real hotbed for innovation and younger businesses. And the same goes for Berlin and Amsterdam and London and Lisbon these days. We see tremendous innovation happening here. I think there's always been a little bit of complaints around sort of the funding circuit around it that this is not as mature or as high-risk taking as that the funding circuit in the US is. I don't know. I see some of these companies being successful regardless of those challenges. I do think that European entrepreneurs are actually much more suitable for building worldwide businesses. I think in Europe you only have to travel 200 kilometers east or west and you have a different language, a different culture, a different tax system. Even if you have an EU there's all these challenges that are continuously around it. If you're an American entrepreneur everything, everyone speaks English. Everything that gets published in the New York Times gets read throughout the whole country. Here if an article gets published in the Frankfurt-Ralgemeine nobody in the UK cares about that. And so I think as Europeans we're much better suited for building worldwide international operations than many of the Americans are. So that's a great message that Europe can succeed in innovation because we're messy. There's a downside to this. That's the younger businesses. I think if we look at enterprises the investments in R&D in the US are two to 50 times as high as that many of the European enterprises are doing. And if you ask yourself why all of these massive internet powerhouses all come out of the US it is because the investment in R&D. And if you do not make these investments you do not create a competitive proposition. And I think that's a message that should be heard loud and clear that many of these the more you invest in a IT-driven R&D the better your competitive position will be as an enterprise at the worldwide scale especially now that all of these enterprises want to go through digital transformation phase. I know that there's some dispute about the classification of these numbers but in stark terms Amazon spends what $23 billion on R&D every year which is about half in what Britain spends in total both public and private sector. So these numbers are staggering but there is a necessity from our point to really be able to drive new innovation and come up with all these new products and whether that is new types of stores being built versus the access of this world versus the cloud computing division imagine all of this cloud world would not exist if Amazon would not have been making these investments and even if many European companies or international companies are now copying this it is these more that's a US-based companies that are willing to take this high risk and make these massive IT investments where if you're not doing that you lose your competitive position. Alright, well clearly Verna has a lot of money to spend on innovation and technology so the audience please note that we must end it there very much Verna for a fascinating conversation. Thank you. All this R&D money also means we have a lot of money to spend on new employees. So, if you're an engineer we just launched a thousand new open positions in Dublin so you know who to contact. Thank you.