Loading summary
Transcript79 lines
- [00:00]
Jessi Hempel
From LinkedIn news. I'm Jessi Hempel, host of the hello Monday Podcast. Start your week with the hello Monday Podcast. We'll navigate career pivots. We'll learn where happiness fits in. Listen to hello Monday with me, Jesse Hempel on the LinkedIn podcast network or wherever you get your podcasts. Struggling to keep up with customers with agentforce and Salesforce Data Cloud deploy AI agents that know your customers and act on their own. That's because Data Cloud brings all your data to AgentForce, no matter where it lives. Get started@salesforce.com data LinkedIn News.
- [00:41]
Tomer Cohen
I've always.
- [00:42]
Gustav Satterstrom
Been curious, always looking for patterns in society, how people navigate and find their way. You always need some level of conflict to initiate that dialogue. It's never a straight line.
- [00:53]
I'm Tomer Coyne, Chief product officer of LinkedIn, and this is building one.
- [01:00]
If I had known the complexities of the music industry, I would never have joined because statistically the chances of being successful with the music startup already back then were close to zero.
- [01:10]
That's Gustav Sadostrom, Spotify's co President and Chief Product Officer. He's sharing with me just how difficult it is to build in music and the novel solutions they found to stay.
- [01:20]
Tomer Cohen
Ahead of the game.
- [01:21]
Gustav Satterstrom
We're going to get into that and so much more, so stick around. Today's episode is a real treat. There are just a handful of apps I use frequently throughout the day. LinkedIn, of course, is one of them. But another special one is Spotify. As a music lover who used to spend my entire paycheck as a teenager buying music albums, the delight of having a great music experience always in my pocket is a very special one. That's why I'm excited to end the first season of Building One with a very special guest and a remarkable, remarkable builder, Gustav Satterstrom. Gustav first arrived at Spotify in 2009 to lead their transition to mobile, and he's been leading the company ever since. He is the Chief Product Officer at Spotify, but also the Chief Technology Officer and the co President. For Gustav, the labels don't matter. His entrepreneurial experience has meant he's long been adept at working and innovating across disciplines and domains. There's so much to learn from this episode, but I'll highlight a few items to get your appetite going. How patience and tenure are important for making big long term bets. Why Spotify will never quote move fast and break things. This is a good one. Why being able to explain yourself trumps having good intuition and how you can Scale the benefits of that, how Spotify built and developed its expertise in music discovery over time. And of course, how AI will reduce the cost of coding to the point where we can create truly personalized app experiences. Let's get into it.
- [02:58]
Tomer Cohen
So you have a background in technology. You started multiple companies before you joined Spotify in 2009, when Napstar was still a thing. How did your career turn out differently than what you expected?
- [03:09]
Gustav Satterstrom
As an engineer in electrical engineering, data science, my goal was to work in a big company. I did my master thesis at a Swedish company named Telia on mobile peer to peer networks and stuff. I worked for the research department and it was total nerds. I had a lot of fun. But then this was the tail end of the IT crash, the boom and the bust. Telia merged with the Finnish career named Sonra. They moved all the research to Finland and I just couldn't get a job. So I became an entrepreneur, but involuntarily it was the only thing I could do, because you couldn't get a job when you just had a fresh degree. And I had a lot of ideas and a lot of cool stuff that I wanted to do. The flip side of not being able to get a job was that neither could anyone else. So there was a lot of talent available. They had no opportunity income, so they were prepared to work for free. So I joined up with a bunch of friends and started a company around mobile data messaging, which back then was a new thing. This was before the iPhone. You could already start to send the text messages over data instead of over the SMS protocol. So I ended up becoming an entrepreneur and working as a founder and a CEO. And that was very much unexpected. That was never my dream. Growing up, I actually saw myself as a tech geek in a big company. And here I am, tech geek in a pretty big company a few years.
- [04:27]
Tomer Cohen
Later that you helped build from almost from the ground up.
- [04:30]
Gustav Satterstrom
It's like the notion of entrepreneurship as a necessity for you.
- [04:33]
Yeah, exactly. It wasn't a big thing in Sweden. Being an entrepreneur was a thing in Silicon Valley, you know, almost since the 90s. But not in Sweden. I actually didn't know what it was even so I had to figure out how you start a company, how you incorporate something, how you raise money, what a VC is like, all those things.
- [04:48]
Tomer Cohen
When you came back to Sweden, which I think is also remarkable for people who want to build a career in tech, you have this tremendous breadth and depth across multiple disciplines, whether it's technology or user experience or business models or go to market. And what do you think was the most important skill or quality that you developed over the years that actually allow you to have this like depth and breadth across.
- [05:13]
Gustav Satterstrom
Well, it's hard to know what is sort of nature and nurture, how much is an inclination to want to do full things. Like I said, I thought of myself more as a geek or researcher that would actually go quite deep and narrow on something. But maybe it was the fact that I had to become an entrepreneur. So I've been both founder and CEO of different companies. And I guess having that role even for a few years, this forces you to learn everything, have that sort of helicopter perspective and the ultimate responsibility for everything. Maybe that's where I learned it, or at least removed the notion that, you know, if you educate yourself as something, you can mistake yourself for being that. You know, I'm an engineer, so I could never be this, or I'm a business person, so I could never be this. But that's not true. That's just a. That's just a construction on top. So I guess I always try to have more of a sort of CEO perspective. It's about getting the job done. I was always very interested in business models, even though I've never officially had the role. My view was always that good products might leverage a technology innovation or UI innovation, but great products almost always incorporate a business model change. That's when you have big impact and I think it's just as fascinating as technology in itself.
- [06:23]
Tomer Cohen
Couple of great insights there from you. One, it sounds like your tendency is to go deep, but you had to go broad because of the roles you had. So then just allowing for that depth, I think you find out connective dots just going across multiple spectrums, technologies. You also mentioned being careful about being labeled. For me, like at heart, I love building stuff as an engineer, a product person, but I love taking them to market. Like there's so many aspects of what I enjoy in the process. So limiting yourself to a function doesn't feel like a great way to grow and to learn.
- [06:55]
Gustav Satterstrom
I totally agree. I guess I just don't enjoy the problem as much if I can't see the entire problem. It's the beauty of solving the entire problem. And that often incorporates some technology that enables something. But then you run straight into the user interface challenges of this new technology or how to use it. And then you run straight into the business model of how you get this into the hands of people. I mean, I've been labeled in that sense. Sometimes people say like, how can you be both The CPO and CTO and so forth. But that I think is actually a little bit false, at least to me. It doesn't feel like extra work. I was always interested and it's very much passion driven.
- [07:32]
Tomer Cohen
Could not agree more. You mentioned that great products also bring with them a great business model. Spotify is quite unique in that. Do you want to talk about that combination? Because I know the early days, what you've done in retrospect back then was not very clear for people.
- [07:46]
Gustav Satterstrom
No. And my first companies, they were not great successes as business models. Some of them had great technology and we sold them, but they were in between mostly acquihires. So we never sort of cracked a new business model. But I learned a lot about business models and also working at Yahoo, I learned about the search ads business model and so forth. But it was really at Spotify that I got to cut my teeth in combining business model and product. So when I joined Spotify, it existed as a free ad driven desktop product. Spotify was competing with piracy. Piracy was free. People were not interested in paying for music. The Spotify had to be free. And the idea was simply it could be like radio, but on demand. You can monetize it with ads just like radio does. And then when I came in, the iPhone has just come out. This was in 2008. The App Store hadn't yet come out, but it was in the cards. So Daniel asked me together with the team to figure out what Spotify mobile should be. And it wasn't just the product, it was the proposition. It was not that hard to develop an iOS client using some of the same code and so forth. But the thing was, if you were on WI fi, it was pretty straightforward to get the Spotify product to stream music. But the problem was that this product was not going to compete with your desktop computer that had, you know, low latency, high bandwidth broadband. It was supposed to compete with your ipod that had ubiquity, could play anywhere even without a connection. And so back then the mobile networks, even in Sweden, that was a pioneer, was terrible. It was edge networks at best. Not even 3G back then. You know, you could start streaming a song using sort of the same core engine, but instead of being instant, the Spotlight was famous for as fast as the MP3 that you would pirate from Pirate Bay. It did take 20 seconds to start, then it would stutter and then you'd be out of data for the entire month. So like the whole thing of Spotify being you don't have the files anymore, they're in the cloud, you don't pay. It didn't work. So we had to rethink the entire thing. And we actually came up with something that at the time was very contrarian, which was, forget about us telling you that you shouldn't have the files anymore. Actually, you should get the files. And we went and licensed a completely new model which was you can offline sync 10,000 files for up to a month using sort of local encryption to make sure that when the encryption key expires, you can't play your files anymore. That was very different. It was a sync paradigm and we had to negotiate with the labels to figure out what this cap should be convinced of the technology that the key would expire and you couldn't export those files and that users wanted to pay for this. And I thought it should cost like a few dollars. They thought it should cost $10 a month. Clearly it became $10 a month.
- [10:18]
Tomer Cohen
And back then, was it clear that mobile is going to be the device you're going to listen to music to?
- [10:24]
Gustav Satterstrom
We knew already then that listening on the go, you know, the car in your ipod was way bigger than listening at home. So that was already clear.
- [10:32]
Tomer Cohen
So it was clear that desktop was not supposed to be the main product. Desktop was just a way to start.
- [10:37]
Gustav Satterstrom
Yeah, it always had to be mobile, but it wasn't clear that it even could be. You know, we even synced files locally to an ipod for a while. We actually reverse engineered the ipod protocol and synced our files. It was crazy. But we didn't foresee the smartphone becoming as big as it was. We got very lucky in that sense. But it was clear that listening was going to be mobile. But for a while we thought it was going to be the ipod and we just didn't have access to the ipod to the same extent. So for us, it was fortunate that the smartphone wave happened.
- [11:08]
Tomer Cohen
Incredible. And you mentioned working with partners around figuring out the licensing model. You're in a very diverse and somewhat rigid ecosystem of partners. That can take a toll on the building the ultimate product experience you run. I could see that being a struggle around. I know what's the ideal user experience, but I can't do it because there's business oriented restrictions on the other side.
- [11:32]
Gustav Satterstrom
For sure. I've always felt a little bit jealous about my peers at Meta or Twitter. They can just think about something, use the research and say this is the right thing to do and then just build it. I would have to use the research, figure out this is the right thing to do and Then go and negotiate like a lowest common denominator between three, four major publishers and then hope that it still was an okay product. So very frustrating at times. Now I think if I had known the complexities of the music industry, I would never have joined because statistically the chances of being successful with a music startup already back then were close to zero. There are so many failed startups, so luckily for me, I was naive enough that I just didn't understand in the true sense of ignorance being bliss. So I was so impressed with the technology and with Daniel and his vision already. Back then he talked about taking over the world and go big or go home. And this was way back when it seemed silly. This was a small Swedish product, it was very cool in Sweden, but nowhere else. So I think I got enamored with the product and the team, so that's why I joined. And then I think I'm actually very patient. That's my key strength is probably not that I'm better at anything than anyone else, but I am very inclined to being bored for a very long time and not give up. I just accept a lot of. Not necessarily pain, just boredom, like things going slowly, you know, it's funny, you.
- [12:59]
Tomer Cohen
Were saying not many music startups or companies lasted and I was just going to the app store to look at them and write like it's right now you have Spotify, Apple Music, there's, you know, YouTube music, but there's so many lists of like music company startups that existed for a year, year and a half took off, but then kind of died almost at the same pace they took off. Like when you build a product for the long game, is it a lot of deliberation, a lot of principle thinking, like, demystify this a little bit.
- [13:28]
Gustav Satterstrom
It's easy to sort of post hoc reconstruct greatness. I don't want to do that too much. A lot of it is luck. But on the other hand, you don't get Only lucky for 15 years in a row. So there's something systematic to it. I think a lot of it goes back to Daniel and, and his sort of tenacity. So Daniel has this incredible skill where he is definitely back then, like a nobody from Sweden who goes to meet these huge personalities of the world's biggest music labels and somehow manages to get them to agree to what he wants to do. I still don't quite understand how that happens, but I've noticed about Daniel that he manages to get everyone to like him. He's a really nice guy, so I understand why they do. But it's interesting, even people that don't like each other, they usually agree that they like him. He has this uncanny ability to also think with a lot of patience, just get things done that are hard to do with a lot of great people around him in terms of lawyers and so forth. But I think he has that ability to inspire patience and he has a clear vision for what he wants to do. So I think many of us have sort of adopted that around him. Having done this for a while, I do think it's actually built a skill which is Spotify is very good at this licensing business, working with these behemoths where you negotiate contracts that are multi year and you have to set requirements and so forth. And I think what's interesting about that that is maybe a little bit different than other companies is that the cost of being wrong is quite high when that feature set is written into a deal that expires only in four or five years and you have to pay huge amounts of money. So the decisions are very consequential. So like these quick AB tests of everything are not really possible to do. So you both have to try to test your way into the things that you're going to bet on to the extent you can. But you don't have the licenses to test everything. So you have to be quite deliberate and strategic. So we've tended to discuss a lot of strategy in my teams about what ifs and game theory scenario and what happens if they do this and we do this and what is the landscape going to be like in two years and so forth? I think that's built a certain kind of skill that I haven't necessarily seen in other companies. I don't think it's even right in other companies because why would you spend all that time strategizing when you can actually test? Yeah, so that is maybe a little bit different, but we've realized that over the years and tried to lean into it. And instead of doing less of that because it's very painful, we're now saying let's do more of that because it's very painful. That's probably the area that other companies are not going to follow because we're really good at it.
- [15:58]
Tomer Cohen
Is there something to building away from Silicon Valley? Is it being in Sweden to an extent kind of allows you freedom to think differently, you know, by inherently being a different company. I'm curious if I'm making a leap there or if there's something there.
- [16:12]
Gustav Satterstrom
No, I think there's definitely something to that There are a couple of advantages, obviously, disadvantages as well with being in Sweden. You're further away from where the core of the tech culture is, the latest discussions and findings and so forth. So it's required a lot of traveling, remote relationship building, trying to stay up to date. So there are definitely downsides with being this far away from like the center of technology, but there are some real benefits. One is that pretty early on, you know, Spotify in Sweden was kind of like Google in Silicon Valley. When Google was at its peak as an employer, you know, we were the biggest in our market. So you had, quote, unquote, unfair access to the best talent. You know, we had an office in Silicon Valley for a while, but it's not great having your back office where your competition has their main office. So there were just geographic advantages to being in Sweden, but it was also cultural because Daniel certainly brought a lot of the Silicon Valley culture. I brought some of that and those findings, like challenging ourselves. But also we were far enough away that we always figured like, maybe we should do this a little bit our way. And we had this Swedish culture of more consensus and so forth, which at times has slowed us down, but at times has also helped us tremendously. So I think you're right that being that far away let us develop a bit of our own culture. And certainly I think our self confidence in that has changed over the years. You know, while we try to take the best of these other companies, we developed our own cultural skills and that started paying off. We realized that we wanted to be as different as possible, in a sense, from other companies.
- [17:42]
Tomer Cohen
You've seen both cultures closely. Like, what stands out for you around those advantages that you've built in?
- [17:48]
Gustav Satterstrom
It's hard to know what part of it is the culture versus geography, et cetera. We just have very, very high retention. I've worked there for 15 years, but my co president, Alex Nordstrom, he has worked there for 14 years. And many of the people who work in my team, they worked for me for 12 years or 10 years. We have very long tenure. That was a clear advantage versus being in Silicon Valley. If your horizon is one and a half, at the most three years, then as a leader, you're just not going to attempt to build a product that's going to take four years. Makes no sense for you because you're not going to even be around, right? So I think that 10 year meant that many people made bets that were quite long term. They may have thought they would take three years to play out, but they took Five years to play out. And that has benefited us, I think, versus some other companies. And then we try to lean into that. So there was tremendous pressure to use the metaphrase, move fast and break things. There was this pressure to just don't talk. Code decides, arguments, just move, move, move. But as I said, that didn't really make any sense in our world because the cost of being wrong was so high. So we tried to counter that. And I tried to develop counter phrases that would annoy people, like, you know, talk is cheap, so we should do more of it because it's much cheaper than writing code and certainly much more cheap than shipping the wrong thing and rolling it back for six months. So we try to build into the culture these things that were a little bit counter to the mainstream culture.
- [19:17]
Tomer Cohen
Yeah. Internally, we at LinkedIn deliberate a lot because we're trying to build something which is differentiated from the pack. Like, we're not building an entertainment product, we're building a productivity product, which is very different. We get lumped into the social network ones. I'm like, you don't get us. Ultimately, you come to LinkedIn to check in, not to check out.
- [19:33]
Gustav Satterstrom
Exactly.
- [19:34]
Tomer Cohen
And that comes with a lot of deliberation. We have, like, the Socratic method of really discussing film through it. I would love to test 10x more things, but not at the expense of talking them through. And that's healthy tension to have.
- [19:46]
Gustav Satterstrom
Yeah. This kind of goes into how you think as a leader. One of my big heroes is the physicist David Dodge, and he talks a lot about, why are explanations so valuable? So I think that in a company, people don't necessarily need to agree with you. That's too much to ask. That's actually a problem in the Swedish consensus culture that everyone needs to agree. But I think everyone deserves an explanation. And you, as a leader, need to be able to explain yourself why you're doing something. And then people can say, like, well, I don't agree with the premise of this and that and those assumptions, but I understand why you do it, and I think that's fine. What I don't like is when people say, I can't explain this to you. It's like, above your pay grade, you're not smart enough. It usually means that the person saying it doesn't really understand. So I'm trying to force people to explain themselves and I'm trying to force myself to explain myself. And I think a great way of doing that is to use models. So explanations, models are dangerous in a sense, because they're not the truth model is this implication of the truth. And if it's too low dimensional, you can miss an important dimension. But if you have either a fairly complicated model or at least a few different models with different dimensions, you can triangulate something and you can get to a pretty decent prediction. And the benefit of a model versus an opinion is that it scales. If you explain that model, if you write it down, if you teach it to people, it just spreads over. The organization has a life of its own. That's the benefit of explanations. They have this fantastic power. And I want to try to push for explanations. Now that's dangerous because there is such a thing as pattern recognition and instinct. And if you've worked for 20 years, you've seen a ton of stuff, you may have really good pattern recognition that you can't fully explain. And that's valuable. It's called seniority. So I don't want to discount it, but I want to try to push.
- [21:33]
Tomer Cohen
For explanations because ultimately people are trying to learn together. It's worth trying to say, like, hey, this is what I'm basing this on. I might be wrong, but that's what my intuition tells me. And I've seen this through. And I've seen this through. Maybe an overgeneralizing, but that's what I believe needs to happen. Now, the other side, they might disagree, they might challenge you, but it's not like you're leaving stuff off the table and they don't know if they're seeing the complete picture.
- [21:57]
Gustav Satterstrom
Yeah.
- [21:57]
Tomer Cohen
And I think that just build a better organization.
- [21:59]
Gustav Satterstrom
An instinct of great pattern recognition is very valuable. I want that in my senior leaders. It's kind of second price. It's like silver. But first price is if you can explain why you have that intuition. Because then people can take it. They can run with it. They can answer 15 other questions using that model that they didn't even ask you. It's not always possible and it can slow you down. But that would be my ideal state.
- [22:20]
Tomer Cohen
Exactly. Let's shift to the Spotify experience. I love my Spotify account. My whole family. I think I've joked in the past that I'm saying I'm almost like an hourly user to an extent.
- [22:31]
Gustav Satterstrom
Those are the best.
- [22:31]
Tomer Cohen
Even at work, I would put stuff for myself so I can focus.
- [22:34]
Gustav Satterstrom
Hey, Jus. Hourly active users.
- [22:37]
Tomer Cohen
Yeah. When it comes to music, I'm generally an hourly user. And I was thinking a lot about like this dichotomy. It feels like there's an explicit and implicit mode, at least for me. The explicit mode is I have a song running in my head. I just want to play it. I don't want to have any friction in my way. I just want to play it through. And it's almost like Satisfaction, right? Something is playing. I need it. I want to play it through. At least for me, then the implicit side is more of the discovery, delight experience. This is where I'm not expecting you to give it to me, but you gave it to me, and that just elevated my experience as a whole. And for me, with music, like, literally, you can feel elevated by the discovery of it. And I was wondering if that construct is correct, and if so, do you.
- [23:21]
Gustav Satterstrom
Think about building for those separately?
- [23:23]
Tomer Cohen
One is just get the fundamentals right. Don't write a song.
- [23:27]
Gustav Satterstrom
Give it to him as quickly as possible.
- [23:29]
Tomer Cohen
And the other one was when he is open. You know, I just pick one song, and I didn't pick the song after. Go for it. Try to bring something in.
- [23:36]
Gustav Satterstrom
Yeah, I think it's a great observation, and it's also like a journey over time. There was a playlisting tool, so, like, you could search for the tracks and you could build your own experiences using playlist. It was a lot of work on you. You had to know the entire back catalog in the back of your head. It's like, I want this song for that moment. You had to keep up with all the new releases that came every day. But if you were good at music, you could build great experiences. And then people started building experiences for each other. People started listening to other playlists. So we saw that some people had very high intent, know what they wanted to do. Many of them sort of playlisted in traditional genres, but more and more, they playlisted sort of almost for use cases like workout or background or studying or dinner table. And so we actually got a lot of ideas for what the different modes of music listening were from the playlist data. And then we kind of said, okay, it's great if people can find that socially or if they happen to have a friend that can send that list. But most people don't have that music friend. We realized, you know, maybe we could create that friend for everyone. So we had editors sort of identifying the use cases like songs to sing in the shower or songs to sing in the car. You can almost think of them as product managers. Like the work. The product manager is to understand the use case. What mode is Tomir in? What is he trying to do? Is he relaxing with friends in the kitchen? Is it driving? What is going on? And then they create that use case, literally the image and the label. Just fast forward like a few thousand songs to sort of. You can think of it almost as telling the system what this use case sounds like. And now the machine learning algorithms can understand, oh, that's what songs this thing in the car is like. And they can pick out another few hundred thousand out of the, you know, 100 million catalog. And then when we deliver it, we can still personalize it to you. So users were actually doing the use cases for each other, and they covered these different modes that you're talking about. And as they started playlisting for others, we got more users. And then we realized if we started playlisting for them, we brought the friction down even more. And eventually we went into AI and we started personalizing and combining them with editorial. And that's a big reason why the product scale and what's so beautiful about music is exactly that, that it's used in many, many different contexts. So we try to reflect that in the ui. When you come in, we have a pretty dense layout and we try to cover like the different use cases that you may have. There is certainly your liked songs, which a lot of people listen a lot to. For example, you have your playlist, but then you have something like the aidj, for example, which is this drastically just I don't even know what I want. Just click that button and people go between these use cases. Sometimes they're very leaned in and come in with an opinion, and sometimes they have no opinion. They want to be entertained. And that's a little bit different from other products. I think if you open maybe Instagram, it is always the intent. I just want to be entertained. I don't know who's going to be there. I don't know what it is. Just entertain me with Spotify. It's a little bit tricky because sometimes you come in and say, entertain me. Sometimes you come in and know exactly what you want to do. And our UI just gets in the way. We have lots of recommendations for other stuff that you didn't want to do. So this is actually something we've been struggling with, finding what the Spotify UI is because you are in different modes within music. But now, to make it more complicated, you also may be in different sessions, you may be in two different podcasts and in an audiobook, and also listen to music. And so we're trying to keep all those states for you so you can quickly go back. But we also can't just guess on one of them. It doesn't Work we're not good enough at predicting, so we need to kind of show a mix of those. It's a challenge, actually.
- [27:02]
We're going to take a quick break, but don't go anywhere. When we come back, Gustav is going to share with us how he believes AI will completely transform music and how we build products.
- [27:11]
Some people say my Spotify and sort of as a friend, but it was always an analogy. I think it's entirely possible that some of these products actually become your friend.
- [27:22]
Jessi Hempel
Struggling to meet the increasing demands of your customers. With AgentForce and Salesforce Data Cloud, you can deploy AI agents that free up your team's time to focus more on building customer relationships and less on repetitive low value tasks. That's because Data cloud brings all your customer data to AgentForce no matter where it lives, resulting in agents that deeply understand your customer and act without assistance. This is what AI was meant to be. Get started@salesforce.com data.
- [27:53]
Gustav Satterstrom
All right, I'm back with Gustav Satastrom, the co president of Spotify.
- [27:57]
Tomer Cohen
The AI component is where I usually get very giddy and excited and I. There's kind of, there's understanding the music side of it, right? Just understanding the song or the, or even the artist to an extent. And then there's understanding the user, right. So for example, if you know, I'm in a meeting right now, like I'm like at work, then probably it's not like, you know, rock music, but it's probably like more focused music, right? Maybe it's like me imagining, but I'm feeling like the app is starting to get this from me. Like I do my meditation in the morning, it shows up first kind of thing. And so I'm wondering like, was there any, like breakthrough features that really changed how the matching is being done?
- [28:37]
Gustav Satterstrom
So there are like three phases. The first was we got all this playlisting data and what happens when people group things together and give it a name. It was just a grouping. So we had, you know, pretty early on, billions of examples. You have the cluster exactly.
- [28:50]
Tomer Cohen
In a way you have unsupervised learning done by others for you.
- [28:53]
Gustav Satterstrom
It's exactly that. So I might put together a bunch of songs and say, this is Edmund. We might not even know it's edm. But it is actually the perfect training data saying these songs are close to each other. So we built similarity vectors between songs and playlists, which we got a ton of mileage out of and that's really how we understood the music. But it's interesting because we didn't understand anything about what was in the music. It was just IDs of music. Just, you know, these numbers are close to these numbers in a vector space based on playlisting.
- [29:21]
Tomer Cohen
Incredible.
- [29:22]
Gustav Satterstrom
So that gave us some things that no one else had. It gave us these really esoteric discovery experiences where if you're into this minimalist music, it turns out if you're millions of people, there are going to be at least ten thousands of playlists from minimalist. So we found these songs that people were like, how could you possibly know? So that was great. But we had another problem because we didn't understand anything about the music. We could also play you some heavy metal and then Justin Bieber, because statistically, for one reason or another, this happened in the data. And we had no understanding that these were different genres. There were no genre understanding, just similarity vectors. Right. So we actually made this acquisition of a company called the Econest many years ago, which was sort of a big bet on AI, but they were actually doing a lot of classification and heuristics around genres and understanding the actual music. So it was really, when we made that acquisition, we had all the similarity data and then we bought this company that could say that, yes, but this is actually also Christmas music, or this is actually also Swedish pop. So then we bought ourselves a later on the outstanding of semantic understanding. So we had like, mathematical understanding of similarity. We lacked the semantic understanding. Once we had them, then we could start building great sessions where we could actually promise you that this is going to be Swedish pop. Not just these songs are similar in a vector space. And that combination was really powerful. So I would say those two things were important.
- [30:42]
Tomer Cohen
And you can look at music discovery today and it's like, it's really good. Like, I couldn't see where else we can go. You can look at it and saying, wow, we were just at the early innings because, like, imagine understanding the user better. So I'll give my example of just thinking about it. Like, you know, I work out in the morning. Usually my workout starts slow and I want some, like, calm music when I do my initial stretching and basic drills. But then it becomes very intense and I don't want to create a playlist for that. I would love you to be able to kind of read my heart rate on my wrist and knowing the music I want and then just become natural. And then I think about the spirit elevation you would get from that. As somebody who loves music, that feels incredible. And that's just one thing.
- [31:21]
Gustav Satterstrom
Yeah, we actually built at one point, like, A running experience that did something like that, a message or cadence, and so it would adapt a single song. We actually recorded many versions of the same songs with Tiesto and a bunch of classical music. It would actually adapt your cadence and it was a beautiful experience. But it was a business model learning as well, which was that we needed a big supply of content for that experience, to fit everyone's taste. And because it was so expensive to record these things in studios, the backend model of it didn't scale. We just couldn't produce enough content. But the experience was amazing. I think now with AI, it's very straightforward how you could just make the song change. Back then, you needed to record different versions. So I'm very excited about looking at those things again now that we have the technology that makes it affordable.
- [32:05]
Tomer Cohen
When I looked at Spotify's mission statement, it's not just music, it's the whole creative space that you guys are looking at. What kind of excites you when you look at it, you're like, I don't know exactly how I do it, but I know it's possible and we're going to go for it.
- [32:17]
Gustav Satterstrom
Like you, I think this is one of the most exciting times and product because you have this new technology of AI, and specifically maybe generative AI, and so I think it's different things in different categories. For music, the goal was always to build that perfect session, as if you had a personal DJ that knew you as a user, could talk to you and say, like, hey, Tamer, I've been up all night looking through all the new releases and like, look at what I found for you. But it was an analogy for a long, long time because it was actually an editor that, you know, playlisted for 2 million people. It wasn't personalized, but it's starting to become true. That's the promise. You know, we think we can actually build that perfect session with that DJ that understands you, understands what happened in the world of music, have read the news in the morning, can talk to you about your music, tell you anecdotes about the bands you didn't know, play mix, like all of these things. It's now entirely possible. We're not quite there yet, but this is my dream and my passion to build that perfect session. We try to imagine this product. We say, like, what if we could afford to hire three human editors per user working in shift? You know, they're just working 24 hours a day, sitting there, going through all the new releases, all the back catalog, thinking about Tamir Tamir, Tamir. Like, they interview you and they just work for you all the time. You could simulate that experience if you had a lot of money and have experts work for you. That's the experience we're trying to go for, and I think it's going to be that. So that's what I'm excited about. In music, in podcasts, it's different. I don't think it's going to be a voice that replaces creators. I think it's about someone helping you to find creators. There are different problems to be solved there. There are very exciting things about translating creators into many languages so they can have bigger audiences, helping them summarize, helping them get discovered and so forth. If I would say, sort of macro. The thing, I think it's possible. Now back to analogies is that I think products might actually become truly personal. So we've talked about Spotify being personal for a long time. Some people say my Spotify and sort of as a friend, but it was always an analogy. I think it's entirely possible that some of these products actually become your friend. I don't think it's very controversial. Even after OpenAI, these things can become friends.
- [34:30]
Tomer Cohen
What you just described for me sounds remarkable. I was thinking about the last few decades, and both you and I came from kind of communication engineering, where you learn how to talk to machines. Like how everything you've been taught in school was, how do I talk to a machine and get it to do what I want it to do? And now machines are learning how to talk to you in a way like it's shifting.
- [34:51]
Gustav Satterstrom
Exactly.
- [34:52]
Tomer Cohen
And, you know, you were talking about that kind of person, like, almost like a personal trainer for you, for music that can actually see. Like, am I breathing heavily? Like, am I working out? Like, the voice, the tone, the facial expression, if you can actually have a camera kind of looking at you. Like, there's so many things you can pick up that you just can't pick up from text or just when I open a session, almost like the app is in the background. So, like, imagine a world where Spotify, the app, is not the main thing. Spotify, your trainer, is the main thing.
- [35:17]
Gustav Satterstrom
That's exactly what I would imagine could happen. You know, a dream scenario would be you walk down the street in Manhattan, you hear a voice, you know, and you're like, oh, that's Spotify. Because Spotify has turned from like, this utility and this app into actually, like a friend or a voice that you would recognize. And that's how you think of it. That's at least a dream of mine. We're not quite there yet, but it's very clear how that could happen now. And I think it's very exciting because it kind of challenges everything to a point. It's sort of much more dynamic. It questions a bunch of things. It's both scary, obviously, but also very exciting, I think. Yeah.
- [35:51]
Tomer Cohen
I have no doubt that one of the biggest relationships you'll have in the future is with your AI. You know, in fact, there's a word for the anxiety of being away from your mobile device, which is called nomophobia. For most people, the phone is always on them. It's like there's actually an anxiety of leaving it at home or going to work without it, or even going to the grocery store without it. Then I think of how intimate AI is going to be to you. Homophobia is nothing versus what you're going to see in future.
- [36:17]
Gustav Satterstrom
Yeah, for sure. And they're obviously both very interesting and scary things about that. But I think what I like about the music, podcast and audiobooks world, it's not unambiguously good, but almost unambiguously good. At least music most people consider being good, and then most people agree books are good for the world. It'd be terrible if they disappeared. So I feel like I'm in a space where AI can help make humanity better, more relaxed through music, more educated through books, more up to speed on science and education through podcasts and so forth. So I'm very excited about being one of the few people who get to work with applying AI to that space. It's probably the most exciting area since mobile phones, because back then it was also like, oh, everything is going to change now. Nothing is true anymore. Have to rethink everything. Scary, but very exciting. Yeah.
- [37:07]
You know, Gustav, you recently went through quite a big reorg where you centralized a lot. I'm curious how your product shaped your organization. What sort of changes have you made along the way?
- [37:17]
So one of the things that get a lot of questions on is doing all of these things in a single app. You know, did you ever think about maybe doing separate apps? Because, you know, a dedicated podcast app, you could have, like, all the podcast features you want. Dedicated music app, dedicated audiobooks app could just optimize more for the user. And this was actually a strategic decision we made. I actually agree with that. We could do three better apps than the single one if you optimized only for the amount of features you could have with uploading the ui. But it was a strategic decision because unlike many of our competitors like Apple or Google, that can pre install themselves on all the iPhones in the world, we don't have that kind of distribution. So if we manage to get ourselves installed on your phone, it is very valuable for us to be able to double down on that existing distribution. And we didn't come up with this. You've seen the Chinese super apps do this, but in the west it was not very common, so we kind of adopted that pattern. The second thinking we had that was more futuristic and not so much just for strategy, but for where we think things are going that I think is interesting in the age of AI is that as a user, what are you doing when you're switching between the apps? I mean, it's all code, it's all software. So why couldn't just the software adapt the UI instead of the user adapting the software? You know, think about, you have three different apps. You could technically merge those three code bases and just switch the ui. So we try to think about that. And so we chose to bet on adaptive UI and we chose to bet on our existing distribution. And that's been working very well for us. The drawback of that back to the organization is you obviously risk complicating the user experience when you're trying to do all of these things. So we kind of had to design the Org literally after the product. I usually take the examples of Amazon where they try to build small teams with hard APIs that have no dependencies, so that everyone can run fast, in parallel, all the way to the end consumer. You can see that sometimes in the consumer experience having multiple search boxes on the same page, whereas in Apple they've done it differently because they're shipping hardware once per year. You can't let teams run fast and do their own thing. You have to synchronize. So we've adopted much more of that approach. We have a single consumer experience organization that owns the application across all interfaces and is instantiated in a single person. And then in the same vein, we're recommending audiobooks, podcasts and music to a single user. You know, for our competitors who have three different apps, it looks like three different users, but it's actually the same user on the phone just switching apps. And we're trying to not fool ourselves. We're saying like, no, it has to optimize globally. So when we recommend something, we have to figure out like Tamir right now, you know, should we show this Beyonce song, or should we maybe show this Peter Attia podcast? Or maybe this Audiobook that he's into right now. And those have different cost implications, but also different retention profiles and so forth. So we've had to make that a central function as well. So we have one that we call the PCM personalization organization under single leader that tries to balance the user recommendations globally. And then behind that, then we have three different teams. So we have a music business team, a podcast business team, and an audiobooks business. And they kind of live as if they're different businesses, but they go through this central point. So we have very much shaped the organization after the product that I think is crucial. Otherwise our product wouldn't work. It would just collapse into complexity if everyone shipped their own stuff. So we kind of said like, okay, we're going to have to eat all that complexity. So we built this organization and it's very hard work for us. It's super painful, but hopefully we don't ship the pain to the user.
- [40:53]
Tomer Cohen
What gets me excited is right now with AI, you can actually remove that complexity. In fact, if you're a complex app right now, I think you have a massive opportunity to reduce it to an incredible experience, which is simple just having that AI agent playing that for you.
- [41:08]
Gustav Satterstrom
That's exactly what I think too, with AI now, now you can start personalizing the product. And yes, we've personalized which songs you get, and LinkedIn has personalized which posts you see for a long time. But the clear promise is you could personalize more than that. You know, it could be almost only a podcast experience for people who only listen to that. The UI could be much more dynamic. You know, I think one of the most interesting analogies I've heard about AI is from this economist many years ago named Ajay Agrawal. So he says, like, what happens when the cost of doing something goes through the floor, right? If you have coffee and the price of coffee goes through the floor, well, its substitutes are going to suffer like tea, but its complements are going to benefit greatly, like sugar and cream. And I think that's a really interesting way, way of thinking about it, try to figure out what the complements are. Everyone's talking about the substitutes, and that's scary. But what are the complements of like, very, very cheap computation or personalization or prediction, as he calls it. But I think the point on this, like, the cost of computation going to zero, you know, I think what he didn't see back then because the LLMs weren't here, he said the cost of prediction goes to zero. But now that these LLMs can write code. One way to think about it is the cost of writing code goes to zero. What does that mean? It probably means that it makes sense for you to write code for things that made no sense before. You would only write codes for things that you would repeat many times, because the fixed cost of getting an engineer to write code and test it was very high. And this is probably the reason why our UIs look like they do, even though we don't think about it. It's because the cost of writing code is so high. But what if the cost has now gone to zero? Wouldn't it make sense to do not just lots of algorithms, but you could also do custom ui? I think this is what people underestimate all the time. It just seems like you could write entire programs on the fly for a very small problem at almost no cost, including the ui. So I would imagine that we see these products getting very dynamic, maybe starting with search pages where you're asking for something specific and different every time. But maybe overall in the product there's a counter side to that. Maybe you don't want your app to change UI every time you open it because it could be pretty frustrating to remember how you get to your library. But there's something interesting there.
- [43:16]
Tomer Cohen
I tried that, by the way. There is a lot about just muscle memory and knowing where to go without doing so. But I tested it and it didn't work out. It failed miserably. But I think the hypothesis changes right now because my assumption was the app was the thing you came for. So the navigation has to be stable because you need to know where to go without thinking about it.
- [43:36]
Gustav Satterstrom
I think that's a good insight. If you're navigating, you want muscle memory and you want predictability. But if someone is presenting to you, you should tailor the message to what you're presenting. And it's easy to get over your skis. There's lots of over promise and on the delivery happening as well right now.
- [43:52]
What a great interview. I'm excited to jump into my key takeaways with you. As always. There were so many in this conversation, so I'll share my top ones first. Nature versus nurture. Gustav's tendency. His desire has always been to go deep. It's a great quality, but it's just not what fate had in mind for him. When he entered the job market as a young professional. During a bust, circumstances required that he take on an entrepreneurial role, making his own opportunities. And that meant teaching himself a broad range of skills on the go. That willingness to explore helped him along his career, and he's been able to pair that with his natural desire to go deep. That combination made him a remarkable builder who is able to build differentiated and unique products. Second, similarly, Gustav, wearer of many hats, resists fitting himself into labels. Just because you're an engineer doesn't mean you can't extend yourself and innovate with business models, licensing deals or go to market efforts. This highly resonates with me. Allowing yourself to be labeled limits your thinking and prevents you from growing regardless of how your job is defined. Venturing into adjacent domains, from engineering to product to marketing, will make you a superb builder. Third, in a world where mobility between jobs, especially in tech, happens at a frantic pace, it's helpful to remember that it's hard to play the long game without building depth and expertise. Whether it's because of cultural, geographical or other reasons, Spotify has been able to take advantage of its employees long tenure that helped them see their bets through with patience and stability, especially if a three year bet turns out to actually take five years to realize. Fourth, move fast and break things has been a philosophy that has been adopted in tech for quite a while. The goal was meant to encourage experimentation and speed, even at the expense of damaging the experience in the short term. With all of the contractual and licensing obligations Spotify is locked into, Gustav knows that the consequences of a misstep could be harsh. Breaking things could mean breaking the company. He also notes that talk is cheap and that's why Spotify has fostered a culture of debate and deliberation, gaming out consequential decisions. It might seem slower at first, but done well, the final outcome would probably be better and maybe even faster. Fifth, somewhat related Gustave is a big believer in the value of explanations. He'll even err on the side of over explaining himself because it forces him to lay out his rationale on the table. That way, even if your co workers disagree with your decision, they know what it's based on and can either inject a useful counter argument or at least get aligned with you. After all, as Gustav says, everyone deserves an explanation, and you as the leader, have to explain yourself. For me, this goes back to principled thinking. But it's not enough to base your thinking on principles. You need to share those principles with your team and 6 these explanations are not just instructive and helpful for debate, they're also helpful to create broader understanding across the company. If you lay out all the dimensions of an explanation, write it down and teach it to your colleagues, it will spread easily throughout the organization, and people can apply your reasoning to current and future work easily. Lastly, depending on the context in which your product operates, you may need to lean into distribution advantage over creating a design advantage. Let me explain that when Spotify got into podcasts, there was a strong argument to be made that they could have made a different app with a simplified user interface optimized for a podcast experience, since most people do not listen to music and podcasts in the same way. But Spotify realized it wasn't competing with the quality of different podcast players, it was actually competing with Apple's own podcast app, which had a distribution advantage of coming pre installed on over 2 billion devices. So by creating a separate app, Spotify would have had to reignite again the distribution reach it already built for its main music app. Do you have an idea of how that might influence your thinking as a product builder? Let me know on LinkedIn I'm Tomer Cohen. Thank you for listening. I learned a lot from this conversation and I hope you did as well.
- [48:08]
Tomer Cohen
If you liked this episode, don't forget to rate and review us on Apple Podcasts. It'll help people discover the show. We're hard at work on bringing you Season two, but if you can't wait till then, then tune in next week where we'll have a bit more wisdom from Gustav as we conclude Season one.
- [48:25]
Gustav Satterstrom
I try to understand competitors and what they're good at. Sometimes to borrow from them, I think you should borrow with pride.
- [48:31]
Tomer Cohen
Building one is a LinkedIn editorial production. Our host is Tomer Cohen, LinkedIn's chief product officer. This episode was produced by Max Miller. Our associate producers are Lolia Briggs and Rachel Karp. This episode was mixed by John Partham and engineered by asaf Gadrone at LinkedIn. Sarah Storm is senior producer and Enrique Montalvo is our executive producer. Dave Pond is head of news production, Courtney Koop is head of original programming and Dan Roth is the editor in Chief of LinkedIn. Thanks to Alicia Mann, Haley Saltzman, Mary Wilson, Sarah Scully, Ayanna Deldridge, Kamen Rojas, Mikayla Greer, Kyle Ranson Walsh, and Maya Pope Chappelle. If you know of a product leader we can all learn from, send us a line@pitchinkedin.