A Brief History Of Spotify: Gustav Söderström

[Music] hi welcomed thanks for coming it's super exciting to see a lot of you sorry there's a standing-room-only but yeah we're all comfy we're all friends my name is irani Ghazi I am a professor of the practice here at MIT and I teach music technology which I'm excited about and I see some of my students here hey guys but this is not about me this is this is about gustav Soderstrom who is here from Spotify Gustav is the chief research and development officer at Spotify and he's been with the company for almost 10 years I think 10 years this December so in his role he oversees all aspects of product development and engineering and design like he's got a lot of responsibilities under him an overall product strategy including new technology platforms specifically in areas of AI machine learning a lot of stuff that we're very excited about here at MIT - Gustav is also an entrepreneur so prior to starting to work at Spotify he was the co-founder and CEO of Connect works which was a company that basically created like a messaging platform back in that time of you know the early 2000s when when that was big um they got bought by Yahoo right away and so that worked out pretty well and then he started working at Spotify but then started another company while at Spotify which is kind of nuts but I like it that company was called 13 labs and they they created vision technology and then they were bought by oculus which was bought by Facebook so lots of cool entrepreneurship stuff going on as well of course as being chief R&D officer of Spotify which as you might know just earlier this year when public and to great success so a lot of lot of great things going on here I have a little personal story when I was at Harmonix as CTO I was talking to a friend of mine whose name is Tristan Jehan Tristan was at the Media Lab and it turns out Tristan and his friend Brian Whitman also from the Media Lab started a company was called the echo nest and the echo nest if you all might have heard of them they essentially had the idea of doing a computation computational analysis of every single song out there on the Internet so they started doing this this work and at Harmonix I was thinking you know it would be great to work with these guys so I started talking to Tristan could we make like some video games using this technology you made and then I get this cryptic email one day saying like actually Iran I can't talk to you about that anymore I'm like okay and then I hear three weeks later that the echo nest was actually bought by Spotify and Gustaf was the the one who arranged that so so that was cool getting but by Spotify so we're totally psyched to have Gustaf here with us today he's visiting MIT today and tomorrow there's actually a lot of great alignment between Gustav and Spotify and stuff we do at MIT Spotify is very research focused as is MIT of course in areas of AI and machine learning and cognition and linguistics just to name a few and also Gustav since he's an entrepreneur there's a lot of great object entrepreneurship happening at MIT as well you might have heard of the sandbox innovation fund which funds something like 300 companies a year and of course the martin trust Center for Entrepreneurship / its Sloan that sponsors a huge number of entrepreneurship programs so lots of research lots of startups it's a great combination and so I'm really excited to have Gustav up here to tell us about his startup story and what's been going on with Spotify for the you know the reason the pastor 10 to 12 years and also where they're going so please join me in welcoming Gustav thunderstorm to the stage [Applause] thank you hey everyone can hear me okay thank you for having me I'm very excited to be here it's an honor to be at this fantastic institution as you heard my role at Spotify is to head up product technology data and design and that means all the applications and consumer experiences that that you are using but it also means the services and backends and the machine learning that powers it now I've been with Spotify for almost ten years and I've had the chance to see it go from about maybe 30 people to something like 3,000 people a few hundred thousand ma use monthly active users to a few hundred million monthly active users and from a few hundred thousand dollars in revenue to a few billion in revenue now this has been a crazy journey and one that I'm very fortunate to have been a part of but as many of you are just about to embark and hopefully these kinds of journeys I'd like to try to share a little bit of what it actually feels like how exhilarating it can be and how scary it can be now I think that any good talk should make you feel like you learned some secrets when you leave it something that you thought you knew but it turns out you didn't actually know so I'm going to try to put a high bar here and you'll be the judge at the end of this presentation do you feel you learned some secrets or not but to take you on this journey I want to start by taking you back in time I want to take you all the way back to 2008 you remember it you were born then right so you know Timberland and one Republic one the charts would apologize yeah you all liked it it's one of those guilty pleasures I know cuz I was on the inside see it and 2008 was a pretty great year I mean the Large Hadron Collider was inaugurated at CERN in Geneva just look at this machine like a sci-fi author couldn't come up with this thing if they tried it's just amazing this guy happened as well just saying not taking any sides here I'm a I'm a sweet we didn't even take sides in World War two and that was a pretty clear answer but more importantly this happened right Spotify launched in Sweden so this is where our journey starts now I had just come back to Stockholm after having worked at Yahoo for a wireless was mentioned and I was trying to figure out what to do with the rest of my life and that's when I met this guy Daniel Nick the co-founder and CEO of Spotify now he had this crazy idea that he wanted to take this and I recognize that yeah it's Napster the original access model to music all the world's music for free with no advertising and offline sync think about that a lot of people mistakenly think that Spotify brought the access model to music we didn't it already existed so then you thought that this was actually a fantastic consumer proposition just to have this flat access to music we didn't have to pay a marginal cost to discover but being a part-time musician himself he wanted to do something that would actually also be good for creators something that would be legal so he wanted to take this free access to all the music ever made and he wanted to combine it with this this is iTunes at the time a much better user experience with great metadata beautiful cover art but more importantly legal we created actually made money so he showed me what he figured that this would look like I instantly convinced really this is actually the original design sketch of Spotify done by a guy named Rasmus and Daniel but in reality they actually had a working beta of the product as well that I got to try and I was instantly blown away away by this and I do in the company but the interesting question is exactly what was I blown away by well any really good product usually pulls off some sort of magic trick and this product pulled off a very distinct magic trick speed now back in 2008 streaming a file off of the Internet could take several seconds to start in fact even playing a file from your own old spinning physical hard drive could take seconds to start the Spotify team had come up with an end and media distribution solution both server and client side not using any CD amps and being able to ignore all the standards such as being able to hack the TCP protocol things like nagels algorithm talk to my sufficiency vs vs latency and so forth and using custom peer-to-peer technology the second thing that happened was that in Sweden we had government-funded high-bandwidth low-latency internet that by this time actually penetrated pretty large part of the population these two things together meant that when you clicked one of these tracks it started playing in about 250 milliseconds the perceptual limit of immediacy I was just floored by this I literally thought they had a huge hard drive in the cross that behind this like when they demoed it to me and the experience of this was the experience of having all of Napster downloaded to your hard drive that was the magic trick and that's why people switch to Spotify not because they didn't have access to music they had that and unfortunately not because they cared at all about creators it was the magic trick so in late 2008 I joined the company just as Spotify was launching and it was an instant hit by Christmas it had gone from about 2000 Emmys monthly active users to about 200,000 by March of 2009 it passed the million and by September of that year barely a year later it was 3.2 million ma years now maybe that doesn't sound so much these days but remember most of this growth was in Sweden a country of about 10 million people including babies and the elderly so we had penetrated something like half of the addressable market in in no time so Spotify on the desktop was a huge hit and was a success but we saw that Spotify was only a really small part of people's music use case of their music system what people actually did was they use Spotify to search for music to find it put it in playlists and his share it with friends and then when they had a great playlist they still went to Napster downloaded all the same files to iTunes to sync it to their iPod because that is where they actually consumed their music the vast majority of music consumption has always been on the go from Walkmans to disc man's to iPods to the car so Spotify vision was always to be your entire music system not just a part of the music system so when the iPhone launched in 2007 and then the app store opened up in 2008 and the iPod which was hardware at the time followed mark and Reese's prediction of software eating the world and turned from a piece of hardware that you had in your pocket to a piece of software on your phone Daniel figured that we could replace that piece of software with our piece of software and then we could be your entire music system so that's why you asked me to come on board to try to figure out what Spotify Mobile proposition would be and most people expected that Spotify Mobile would be a mobile version of Spotify desktop not an unreasonable expectation but remember the magic trick now when you were on a mobile phone in 2009 on 2g or edge now you try to stream something it could take 10 seconds to start then it would stutter and then you could get this SMS my operator saying you're out of data alright there was nothing magical about that experience at all so we realized that Spotify Mobile would at most be this fun thing that you could use when you happen to be on really good Wi-Fi but we didn't want to be a fun thing in companion with your iPod we wanted to replace your iPod so actually the requirements for Spotify mobile were pretty straightforward Spotify Mobile had to work wherever your iPod worked now your iPod would work in the subway when you were abroad with roaming charges and when you were on a flight so Spotify mobile simply had to work offline so we traveled to long-on we met with the labels and we asked them for a license which would allow us to store 10,000 tracks offline on your phone for up to a month before you had to reconnect and verify that you were still a paying customer there were lots of negotiations there were white papers on encryption sent back and forth but eventually they agreed at a price of $9.99 per month so now we have our license and we built the product for iOS and Android this was a truly seamless music experience as soon as you added a track to a playlist on desktop it would immediately and silently start synchronizing to your phone so that when you later wanted to play that song it was already there so it's a great experience but when we ask users basically the question how do you feel about starting to pay 100 20 dollars per year for something that you already have for free guess what none of them said they would mean zero because they think they figured the system they had work just fine fortunately for us we weren't good enough at user research so we launched it anyway see what happens the initial reactions were lukewarm at best ranging from good luck - no freaking way I'm gonna pay for this thing but then something really interesting happened we saw it in the data people started walking out the door realizing that they had forgot to manually synchronize one of their songs from Spotify to their Napster Plus iTunes playlist and they said like ah this is really inconvenient maybe 999 isn't that much it's only the price of two beers actually one beer in Sweden and they started converting first in single digit percentages then in double digit percentages and outperform any freemium model we had ever seen I mean the bar at the time was Skype and Skype was I think converting at something like seven percent we went from 7 percent past 10% past 20% past 30% and we had stumbled on this incredible freemium model where as a product we didn't only reach the people who already wanted to pay $9.99 for music which was a very small group at the time in Sweden because piracy was rampant and legal but we we had this proprietary engine or free product it basically created these users out of thin air we just sat and watch these no music lovers kind of become music lovers and start paying eventually so now we had not only a great product we actually had a business and our premium conversion curve started turning sharply upwards going into 2010 things were great we were growing at a great clip both in free and paid users the future looked pretty bright in June of 2010 Danny decided to have a live team off-site in this gazebo in the Swedish archipelago it's a very nice gazebo you know people were excited things were looking good but I had just read through merrymakers famous analyst at the time a classic state of the internet 2010 presentation and as we sat down and looked at this slide we realized that the party was going to end now the orange in blue bar here our desktop and laptops computers the green bar is smartphones now for the desktop and laptops we had our feature that perform this magic trick that eventually made people want to pay for that for the smartphone right but if mary meeker was right in 2012 we would start seeing users that were mobile only or at least mobile first they never had a desktop we had no free tour for that we had no magic trick so we realized that entire business model was doomed but you couldn't see it in the data yet and while we were it was this weird feeling of kind of panicking in slow motion imagine something coming up to you and screaming like oh my god you're gonna die you're like when when when and that oh no five years from now what do you do with that it's hard so we started thinking about this tricky problem like how do we give away the thing that is driving all of our revenue for free mobility that's what we were charging for mobility so while we were panicking about the desktop feature disappearing no one else saw it yet in fact the labels thought the desktop feature was too good and he came up with this bright idea that if we only made it much worse people would convert that's product development for you this was probably the as a product person this is probably the low point of my life I mean someone asked me to deliberately make my product as bad as possible as a strategy it didn't matter how much data we show them that the only correlation between between free and paid users was how much they engaged in the filter which is not an unreasonable hypothesis like you're gonna pay for a product that you use right and the more you use it the more valuable you're probably gonna think it is and at some point that value is gonna cross $9.99 you're going to become a premium user not unreasonable all the data supported it didn't bite so the labels forced us to introduce caps on the free desktop product so you could only listen for a few hours per week and then the entire product stopped working even worse than that you could only play a certain track a certain amount of times and then in grade up forever yeah that really happened we were actively kicking out our most engaged users of the product so the test went live and as predicted there was an insignificant one-time bump in premium conversion but no change to the slope of the curve and then user growth started tanking now it wasn't just flat we were actually losing users and as you may learn in life very few people very few companies get to stare negative growth in the face and live to talk about it it's very rare now we managed to later reverse these caps with the labels and and what might have been the dumbest a be test of the century an a/b test without a B group by the way but it didn't matter our product was tainted our brand was hurting consumers were really confused about what Spotify was and we kept declining in spite of this in July of this year we decided to launch in the u.s. now in the US people hadn't heard of this capped product so we weren't suffering from the same user confusion so we started growing a little bit now on the other hand people in the US hadn't heard of Spotify at all so we didn't grow very fast and we didn't know if this was actually a new beginning or just the last twitch of a dying corpse in September of that year we launched a partnership with Facebook when we try to make it as easy as possible to find share and listen to music with your friends it was kind of our Hail Mary fortunately for us it worked and the u.s. discovered Spotify we had escaped that Joseph's death for the first time so we were back to healthy growth and from the outside if you were an investor things were pretty good but we all had this all - Mary Meeker slide in the back of our minds of our impending doom and when we looked at the data in 2012 she sure seemed to be right smartphones were exploding while desktop was slowing down rapidly and in June of 2013 the shift to mobile finally showed up in our metrics and we were facing negative growth for the second time it was a summer vacation in Europe and our desktop usage always dipped during summer vacation because people left their clunky laptops and desktops at home when they were traveling use their phone instead but we had always seen desktop usage come back with a vengeance in July and August but this year it was as if they had put their computers in the closet when they went on vacation and he simply never took it out again this was the summer that Europe went mobile and it really happened all that was so now we really needed a mobile feature but our challenge was that it wasn't just enough to come up with what a mobile feature would look like that wouldn't cannibalize our premium users we also needed to convince the labels to license it to us and they had no intention of doing this as long as we were growing so it literally required negative growth to get them to the table now during the last few years we have been experimenting with different hypotheses for mobile feature for example what if mobile Spotify Mobile was free and Wi-Fi but paid on cellular and a bunch of other hypotheses but we had settled on an insight from a person in my team named Charlie Hellman who had looked at all the premium user data and realized that most of the premium users almost all of them they actually shuffled played their playlists about 50% of the time voluntarily even though they could play them on demand in order they chose to shop or play them so we started thinking this seemed like a pretty good use case to be able to shuffle player playlists so what if we would give away that use case for free that meant that we had a great product for free users but it also wasn't the entire use case for any premium user right it was at most 50 percent so the thinking was could we have our cake and eat it we give away one use case but not so much that the premium users would actually go back to free so during the fall we were working on this product that we call Spotify mobile free the idea was that it would be the only product in the app store were you as a user to go in create a playlist and listen to your own playlist for free there was really no competition for that in the app store the only thing that existed was here in the US which was Pandora with their stations so what we did was we took all the user data we reverse engineered and asked two labels for a shuffle license where we asked for exactly the artist and track diversity we needed for 99.9% of user playlist to be able to shuffle without us having to insert any artificial tracks so we used the data to look at exactly what kind of artist diversity we needed so during the fall we were working like crazy to get this product ready what the licensing team was trying to get it licensed and in late 2013 we got the licenses signed and we put the application up in the App Store the very same day and then we held our breath would what's the hypothesis right would we actually save the company again as soon as we hit the app store we started growing like crazy and we grew and we grew and then it stopped and it was Christmas what are you gonna do you can't move Christmas and we knew that there was always lower engagement during Christmas like there was during summer but it was really hard to know if this was a big one-time bump of pent up demand and we were out of growth again and we didn't have a company or if we were genius we was just standing on this rocket that had temporarily stopped to refuel so none of us relax very much this Christmas when we came back in January it turned out that we were on a rocket and we started growing and growing during 2014 we just try to hold onto our hats and make sure we didn't screw it up and by September 2014 we were 15 million any years and by February we had reached 63 million any use so we had escaped the jaws of death for a second time we now had a much more healthy and stable business model that actually worked so kind of for the first time in our history we got the chance to think about something else than how not to die we actually started thinking about how we can make Spotify better for consumers so Spotify was a really great experience if you were a music aficionado we gave you this really powerful search box and a music programming language called playlists and if you knew your back catalogue and you kept up to date with all the new releases you could soundtrack your life perfectly but the problem was that everyone isn't that good at music and we want to Spotify to be for everyone so in 2015 we saw this company called tune ago in Sweden who was using Spotify playlist as a platform and they were play listing along completely new dimensions things like work out party focus dinner with friends and these use cases seem to really resonate with users who kind of knew what they wanted to do and they sort of knew what it should sound like but they couldn't translate it into musical terms I mean is work out hip hop trip up big room or house it's not that easy to figure out now on the outside - Nico could really only see how many subscribers they had to these playlists but from the inside we can see the engagement we could see the streams and how many people were coming back and it was high so we decided to acquire them in a bet where we would give them access to all the data on the inside and try to build the most data-driven play listing system in history once on the inside these editors who by the way was this incredibly powerful team that was super diverse of musical experts representing all kinds of different genres they could take a playlist like songs to sing in the car for example and you could drop in a track and within minutes they could see place place completion skips saves they could remove add and we ranked in this inexorable loop that simply optimized the playlist for the use case so this made Spotify much more accessible for mainstream type of listening now while more mainstream listeners lauded these playlists and engagement was super high the music aficionados which turns out is a large part of the journalistic Corp they hated it they said that it was hopelessly mainstream which was of course the whole point and a complement to the team it turned out though that the music aficionados they seem to really like our recommendations and we had been lucky and already all the way from the start we had some people like Eric Barron Stone who actually understood machine learning I mean back in 2008 which is pretty early and we saw that these recommendations like similar artists for example they really resonated with the more advanced users and it gave them a great way to traverse the catalogue now the reason that we could have such good recommendations for rather obscure and narrow tastes was because we had by luck built a huge data advantage in our playlist now it turns out that when users are play listing and sound tracking their lives along different dimensions for their own benefit they're also essentially labeling the tracks for the rest of the system for all the rest of the audience but this is me not only grouping tracks that go well together but then classifying them as anthems and today we have over three billion of these classifications of the world's music catalog and this means that we can serve even very narrow tastes now as I said this was more luck by design but you have to get lucky there is this concept in machine learning called turning the dial that LJ aggarwal uses in his new book prediction machines essentially it means that when you turn up the dial of accuracy on your predictions or your recommendations when you start your product or your business is going to improve linearly right but at some level your entire product and maybe business model can switch now I didn't think about it as this concept back then because this the book is rather new but exactly this happened to me back in 2014 when I started using a product that was then actually not called Google photos it was called Google+ photos they applied exactly this concept right so while Apple was trying to solve the consumer problem of your iPhone producing way too many photos for you to organize by giving you an even better tool to manually organize them in AI photos Google took the completely different approach using machine learning with prediction accuracy or recognition accuracy was high enough they actually said you will never organize your photos again now this stuck in the back of our minds and we started thinking what if we could do something similar what if we could be so bold that instead of just using our recommendations to give you even better tools like similar artists and search to play this yourself we could say actually you're never going to play this again so with this in the back of our minds in 2014 we acquired the eminent company the echo nest out of Boston MIT now these people were by far the best and audio recommendations in the business it was a fantastic set of entrepreneurs but and they actually powered the recommendations of all of our competitors but none of them or us actually wanted to give them all the user data so they had to work on audio signals and externally publicly available data so we were thinking to ourselves what if we could take all of this talent and combine it with what was already done the biggest amount of music listening data in the world could we create something magical so we acquired them and about a year later actually pretty quickly this team figure out what this missing consumer proposition was how we could turn this thing on its head when we launched discovery weekly this is a musical assistant that uses your taste and tirelessly scours through the back catalogue of tens of millions of tracks to figure out if you might have missed a jam and then presented to you as a weekly playlist for the mainstream users the back catalogue wasn't that big of a problem you could actually mostly look at the charts it wanted to understand what was interesting in the 90s but for the music aficionados this was a godsend and for many of them it outperformed what they had been able to do with the help of search and with the help of their of their music friends this was our first self-driving playlist and since then we've created several more such as time capsule release radar daily mix and so forth but May 2016 we passed a hundred million mus there's Christmas by the way this time we didn't get it scared we recognized that and in 2017 something new and interested starting as the interesting started happening speakers now all the way back in 2013 we realized that if you went through all the trouble back then I'm hooking up your streaming solution to connected speaker system like so knows which is very expensive and exclusive and exclusive at the time it was a great experience and we had this realization that well pretty much everyone had a CD player in their home that they used to listen to a lot in the days of CD pretty much no one ever managed to hook up their files their naps turn I choose files to their stereo right it was just too complicated to do but when streaming happen it got super simple again you bought the speaker and it all the world's music built-in so we made a huge bet that people who start converting to connected speakers sooner or later and so what we did was we started creating a media streaming protocol like I said this was in 2013 long before cast or anything like that the only thing that existed was Apple AirPlay which streamed like from your phone to your speakers if you went too far away it started stuttering we wanted to build this media streaming protocol that was cloud centric and user centric right so all you sent was some metadata and the speaker started streaming from the club so you could get this this really magical experience of listening on your headphones you went up to the speaker and you press play and the stream would just continue from where you were in your headphones over there right because all the devices were aware of where what your account was playing right now this was a big bet we started building software SDKs for all kinds of embedded systems and and chips and after something like four years we had gotten pretty much the entire speaker industry to adopt this protocol it's actually running were actually running code instead of all this hardware except Apple of course so while we didn't predict voice being the thing that would unlock this upgrade of the of the home to the connected speaker we will right enough in our prediction that when this macro wind started blowing we were really well positioned to follow it this brings us to 2018 which was a huge year for Spotify this is the year when we went public yes now usually when you do an IPO which stands for initial public offering you actually want to raise money right so what you do is you do this Roadshow where you go and meet big investors and you pre negotiate a price for your stock which is usually a hefty discount or what it's supposed to go public at so that there is this pop as it's called when it goes public but support if I didn't need to raise any money we had plenty of cash the other problem we actually had to go public for regulatory reasons we're getting too big the other problem with the IPO is that because of lots of arcane rules which were originally aimed at protecting consumers and please ended up being locked up for months I couldn't participate in the potential upside so that was left only for investors and maybe some senior management and that seemed very unfair so our CFO Barry McCarthy he decided that he wanted to innovate on this process as well and he found this other way of going public which is called direct listing which no one does it's actually only used by companies that have gone bankrupt and want to relist and the interesting thing is they're very different rules and regulations for how that kind of listing happens instead of pre deciding the price of our months of negotiation with investors the price is actually set on the morning of the listing through true price discovery of supply and demand and the second thing is there are no lock ups for employees so we decided to do this now I remember sitting and listening on a conference phone to the floor of the New York Stock Exchange with people actually screaming prices and quotes now you don't do that anymore usually like I said it's all pre decided it's all computers but in this process what happens is before the stock goes public you have a bunch of people who want to sell and you have a bunch of people want to buy and you manually broker these deals and they can ask whatever price they want now hopefully the price interval starts to narrow and then it gets in some range where that is the price of the stock and it just floats and goes public now no one actually knew if there was enough supply and demand for this to work or it would be super volatile or go all over the place and the price discovery ended up taking hours and I remember a broker leavin unnamed who thought of us in on mute screaming like Jesus Christ will never get this F installed public but it turned out to be a ride eventually the stock went public and we were above the company and it was back to work now this brings us to today closing in on about 200 million users and the thing that we realized looking at this curve was that no single thing we ever did actually turned out to have exponential growth it all had linear effects right but the point is that the sum of many linear curves of many linear effects turns out to be exponential right so the job is to constantly apply linear improvements to your business and you will get exponential growth so with that I'm hoping that you feel like you learned some secrets and have a sense of what it might be like to be on one of these rides [Music] [Music]

Channel: MIT SHASS

Views: 31,544

Keywords:

Id: jTM7ZCKEUGM

Length: 42min 45sec (2565 seconds)

Published: Thu Jan 10 2019