Gustav Söderström is the Co-President and Chief Product and Technology Officer at Spotify. He is responsible for Spotify’s global product and technology strategy, overseeing the product, design, data, and engineering teams. Prior to Spotify, he founded 13th Lab, a startup that was later acquired by Facebook’s Oculus. He also served as the Director of Product and Business Development for Yahoo Mobile and founded Kenet Works, a company focused on community software for mobile phones, which was acquired by Yahoo in 2006.
Find the episode transcript and Gustav’s references here.
The various roles Gustav has occupied at Spotify ▶️
Gustav joined Spotify in 2009 to lead mobile, because of his previous experience launching Yahoo for mobile. He eventually grew to take on responsibilities for the desktop and web apps as well, so all of product development given the rise of mobile. Immediately, he also took on the CTO role, which is not conventional. Currently, his role is co-president of Spotify and runs the company along with Alex Norstrom.
Why Gustav launched a podcast and what he learned ▶️
Gustav launched an internal podcast first about Spotify, which has proven to be a success in multiple ways. Not only has it made senior leadership more approachable, but it has also helped attract and retain employees. Many new hires have cited the podcast as one of the key reasons for joining the company, and it has even been instrumental in facilitating better communication between the company's leadership and its teams.
"It's very common that people come and tell me that, 'Oh, I listened to this podcast or this and the episode and it's at least one of the key reasons why I joined or sometimes the reason why I joined.'"
It also helped Gustav understand what podcasters go through to get an episode out and their pain points. This was the starting to point to then later make a public podcast about Spotify as a company.
How PMs and product teams should think about AI ▶️
The internet has gone through stages of curation, recommendation, and now, generation. As AI becomes more advanced and capable of generating content, product managers and teams must think about how this shift can be incorporated into their products.
For example, Spotify has experimented with an AI DJ feature, where a digitized AI version of a DJ talks to users about their music preferences and makes recommendations. This marks a significant departure from traditional recommendation algorithms and opens up new possibilities for creatively incorporating generative AI into product experiences. The genesis of AI DJ is when users wanted to listen to something but didn’t have an intent of what, which radio kind of did, but streaming didn’t yet.
Principles of designing products with AI
Design fault-tolerant user interfaces for AI products
When creating AI-driven products, it's essential to design user interfaces that are fault-tolerant and correspond to the current performance of your algorithms. With AI DJ, which aims to provide a seamless listening experience while accounting for the possibility of prediction errors. By understanding the performance of machine learning algorithms and ensuring an escape hatch for users, AI products can achieve better user satisfaction.
...have fault-tolerant user interface and a user interface that corresponds to the current performance of your algorithms.
Mid Journey, for eg, provides just four low res options which their GPU constraint as you wait for a minute, undrestanding both their fault tolerance and throughput of their algorithm. Once a user picks an option, the UX then goes to generate the high res version, keeping the UI still fast and intuitive.
Understand where the performance of generative AI was when they built the UI.
Avoid overusing AI technology and focus on user needs
When developing AI-driven products, it's crucial to resist the urge to show off the technology and instead prioritize user needs. For AI DJ, the objective was to do as little as possible and get out of the way, allowing users to enjoy the music without unnecessary interruptions.
The future of AI generated music ▶️
There are 2 aspects of AI generated music, the generation and legal / rights concern, and we are yet to see what is going to happen on the latter. Focussing on the former and going back in time to 2013, Avicii was initially not considered a real artist by the industry because he couldn't play instruments or sing, but today his music is still on the charts. This could happen for autogenerated music as well, as people's understanding and appreciation of music continue to evolve.
"I think now all of us consider \[Avicii\] very real music and that he had tremendous real musical talent." -
AI music might be seen as an instrument rather than "fake" music. Today artists are already using AI to fine tune their music, so the question is not if but how much will AI be used in creation. However, challenges like rights issues and defining what percentage of AI should exist in the music need to be addressed.
Will AI continue to be a magic trick for products?▶️
All great products need a magic trick, an element that makes users excited about the product and keeps them coming back. Discover Weekly, a personalized, automatically-generated playlist that recommends new music to users each week had a wow moment Spotify saw from user testing, correlating with both activation and engagement, in line with products like DallE and MidJourney when they launched.
How Spotify organizes product teams ▶️
Shifting from smaller squads to larger teams in organizational structure
Spotify initially focused on structuring their teams into small, full-stack and autonomous of about 7 people. However, as the company grew, this organizational structure created a lot of overhead. Nowadays, Spotify has moved towards larger teams of around 14 people, with a more traditional structure and less overhead roles.
How Spotify operationalized autonomy
While Spotify started with a highly autonomous, junior organization, where they’ve had 100s of squads, which meant 100s of product strategies running in different directions; they now places a greater focus on decision-making at the VP level, ensuring experienced individuals are making key decisions while still allowing a healthy degree of autonomy both at the leaves (squads) and at the top (executives)
"The team structure is more traditional, larger teams, less overhead. And we've been specifically working with where in the org do we put the autonomy."
Why Spotify uses a centralized model for structuring their organization ▶️
-
Decentralized teams: Like Amazon, where all teams have access to users and their behavior, any team can create new products directly, which has produced successful products like Kindle and Alexa. This structure creates competition between teams, where there is incentive to hide your code, so Gustav believes Jeff (Bezos) saw this early and mandated that each team expose their APIs to be used by other teams, giving birth to AWS.
-
Centralized teams: Like Apple, where you would never see disjointed user experiences even if two teams are working on the same page/app experience, because there is one decision maker, and often someone high up, that is a bottleneck, which has a drawback of speed.
Spotify started off being decentralized where you see multiple search boxes or multiple toasters on the playing view from different teams which is inconsistent; so they moved towards a centralized structure given they are a single application but different business models (rev share, royalties, book deals etc) to keep a consistent user experience.
The big bet Spotify took with redesigning its interface, and what they learned ▶️
Be prepared to face users' resistance to change
Spotify redesigned the primary feed of Spotify to give it a more modern, TikTok-like Reels feel. As expected, reactions were mixed, with some users loving it and others hating it.
"...you guys had this big launch event recently where you basically redesigned the whole primary feed... and you start hearing videos and music starts playing and some people loved it, some people did not."
Balance background music and podcast recommendations
So we is mainly a background application, and for a long time, we've been considered very good at background music and podcast recommendation. When the phone is in your pocket and you're listening to an EDM playlist or pop playlist or something, we're really good at inserting another EDM track (has a high hit ratio of 9/10) there or another pop track there or something like that in the background.
But, breaking out of the taste bubble can be challenging as it requires a low hit ratio. When recommending entirely new music genres, you need to listen to content with an open mind and expect that you may not like most of it, which lended itself to a feed type experience.
So you need a completely different paradigm. And you also need to be able to go through many candidates quickly because the hit rate is so low. You can't take three minutes per item. It's like, "Okay, I didn't like this," and it's still like two minutes left before the next one comes on. You need to quickly say, "No, no, no." So the obvious candidates for this are these feed-type experience, where you can go through lots of content, you're expecting the hit ratio to be much lower. And if you don't like it, the cost is very low, you just swipe.
Spotify's Home feature has tested features for music discovery but overemphasized discovery at the cost of recall.
While these feeds have worked well, Spotify mistakenly added too much emphasis on discovery to the Home feature, making it challenging for users to perform recall tasks, such as resuming an already in-progress session or finding a specific playlist. The newer version of Home being tested now combines subfeeds for easy music discovery with more effective recall tools.
"When we tested some of \[the subfeeds\] on Home, we switched it from 90/10 to 10/90. So 10% recall, 90% discovery. And while people want discovery, they probably don't want 90% discovery, instead of 90% recall."
How they tested their hypothesis before launch ▶️
The challenge with A/B testing can help validate hypotheses and gather constructive feedback. However, for the home feed, the biggest decisions was to decide to help users get out of their taste bubble, and as a consequence use latest technology (AI / ML) for it, which meant a high cost of development.
"You just have to be unemotional, believe in something fully and just look at the proof and the data. And then if you do that, you just move on and then you get to where you want to be, and you solve the same problem but you adapt. The biggest risk is a false negative. (This is like strong opinions, loosely held)
Gustav’s “10% planning time” methodology ▶️
The 10% planning rule suggests that you should spend about 10% of your time planning versus executing or building. For example, if you're working in six-month increments, try to spend around two weeks planning. This rule of thumb can help prevent over planning and ensure that there's enough time for execution and building.
How to bring energy and clarity to your work
Bring energy and clarity to the table: The ability to thoroughly explain your decisions can improve the overall understanding of the task at hand and foster a more inclusive work environment. Gustav goes around explaining his reasoning even if some of his counterpart don’t want to hear it, at times, just to test himself.
"...even if they \[employees\] don't agree, they should be entitled to understand why you're making the decision...if you can't explain it yourself, you probably don't really even understand it yourself."
How to systematize deep thinking ▶️
Condense explanations through rewriting: Gustav shares his explanations more concise by writing drafts and continually condensing them. He now finds it useful to have walk-and-talk sessions with peers, getting feedback in the process, and believes that thinking while walking is more effective.
I found that walking, talking, and thinking actually even if you're not in person, just over AirPods, it's super effective.
The peeing-in-your-pants analogy ▶️
The Swedish saying is along the lines of peeing in your pants in Swedish cold weather, that makes you feel warm but only for a a short amount a time; which is an anology of thinking short term, that he tries to do less of and uses this as a shared vocabulary term with his teams.
Thoughts on how the Swedish culture is portrayed in Succession ▶️
Swedish people tend to be serious, cautious, and this \[Alexander Skarsgård's\] guy's more of a player. So he's not the typical Swedish businessman from a negotiation tactic point of view (serious and cautios), I think.
What’s next for Spotify and Spotify Podcasting ▶️
Find podcaster’s audience: Spotify is currently working on enhancing music discovery and podcast discovery to help creators reach more listeners.
Monetization: Both free and paid options, to better support podcast creators.
We're investing a lot in is just the ubiquity and playback across different devices and in cars and all these things that we've done well for music. But I think the listening experience can get a lot more seamless. I think search can get better.
Lightning round ▶️
- 7 Powers by Hamilton Helmer
- The Complete Investor by Charlie Munger
- The Mystery of the Aleph
- Something Deeply Hidden: Quantum worlds and Spacetime by Sean Carroll
- Helgoland by Carlo Rovelli
- The Beginning of Infinity
- The Fabric of Reality by David Deutch
- The Case Against Reality by Donald Hoffman
- Incompleteness: The proof and paradox of Kurt
- The Demon in the Machine by Paul Davies
Favorite recent TV show: Succession is a recent favorite, he also recommends the show Halt and Catch Fire, which starts in the '80s Silicon Prairie and follows up to the present day.
Favorite products: Gustav enjoys playing around with ChatGPT GPT-4, creating bots and experimenting with different functions. He's also impressed with Duolingo, both as a product and as a tool used by his entire family for learning Spanish.
Favorite interview question: “What’s the meaning of it all”?
Favorite process to improve product impact: The Socratic debate.
"I'm trying to push a lot for what I call Socratic debate, where the idea is obviously that the best idea wins, not the most senior idea and so forth."
Favorite ritual: One fun ritual is with the four phases of product development: "Think it, build it, ship it, tweak it." By giving every stage a clear and catchy descriptor, team members can easily understand and communicate the current status of a product, which can helps streamline the development process.
Reach out to Gustav for feedback on Twitter: @GustavS
This is a human edited summary of the podcast episode with Gustav, by Gaurav Chandrashekar(@cggaurav, productscale.xyz). To listen to the full episode, go here.
