Lessons from scaling Spotify: The science of product, taking risky bets, and how AI is already impacting the future of music | Gustav Söderström (Co-President, CPO, and CTO at Spotify)

May 21, 2023 Episode Page ↗
Overview

Lenny interviews Gustav Soderstrom, Co-President, Chief Product, and CTO at Spotify, about taking big bets, team structure evolution, AI's impact on music, and the "magic trick" of great products, offering insights into product strategy and organizational design.

At a Glance
15 Insights
1h 24m Duration
15 Topics
6 Concepts

Deep Dive Analysis

Evolution of the Internet: Curation, Recommendation, Generation

Gustav Söderström's Journey and Roles at Spotify

Lessons from Launching Spotify's Product Story Podcast

Product Team Strategy for AI and Generative Models

The Future of AI-Generated Music and Artist Impact

The 'Magic Trick' Principle in Product Design

Spotify's Evolution from Squads to Centralized Organization

Optimizing Organizational Structure: Centralized vs. Decentralized

Spotify's Big Bet: Redesigning the Primary Feed and Learnings

Navigating User Feedback and Data in Product Redesigns

The 10% Planning Time Methodology

Bringing Energy and Clarity as a Leader

The 'Peeing in Your Pants' Analogy for Short-Term Thinking

Swedish Culture and Business in the Show Succession

Future of Spotify and Podcasting

Internet Evolution Stages

The internet has evolved through three main stages: curation (users digitize and organize content, e.g., early Facebook/Spotify), recommendation (algorithms suggest content, e.g., Discover Weekly), and generation (AI creates new content, e.g., AI DJ). Each shift requires rethinking user experience and business models.

Fault-Tolerant User Interfaces

This principle suggests designing user interfaces that account for the imperfect performance of underlying machine learning algorithms. Instead of a single 'play' button for a potentially flawed recommendation, a fault-tolerant UI might display multiple options or provide an easy 'escape hatch' for users to correct errors, like Midjourney showing four images for one prompt.

Product Magic Trick

Truly great products often incorporate a 'magic trick' that initially makes users wonder how it's done, even if it's just data and statistics. This sense of magic correlates with products going viral and taking off, often achieved by reaching a certain performance level and carefully scoping down features.

Organizational Spectrum

Companies can be organized along a spectrum from decentralized (like Amazon's two-pizza teams, optimizing for speed and parallel work) to centralized (like Apple's functional org, optimizing for user experience and coherence). Both extremes can lead to success, but the choice depends on the company's core strategy and what it prioritizes.

Socratic Debate

This approach emphasizes that the best idea should win, not the most senior. Leaders should explain their decisions thoroughly, even if employees don't agree, to foster understanding and ensure the decision-maker truly comprehends the rationale. This also helps share knowledge and makes the organization more effective.

Peeing in Your Pants Analogy

This analogy describes short-term thinking or actions that provide immediate, satisfying results but lead to regret or negative consequences later. It's used to caution against prioritizing immediate gratification over long-term benefits, particularly in product development.

?
Why did Gustav Söderström launch a podcast despite his demanding role at Spotify?

He launched it to fulfill a personal 'secret creator dream,' empathize with podcast creators by experiencing their challenges firsthand, and build internal culture by making leaders more approachable and sharing Spotify's product strategy in an engaging way.

?
How should product teams approach AI in the generative era?

Teams should view the generative era as a fundamentally new shift, not just an iteration of machine learning. They should focus on building 'fault-tolerant user interfaces' that account for AI's imperfect performance and avoid simply showing off technology, instead prioritizing user needs and getting out of the way.

?
How will AI impact music artists and the music industry?

AI-generated music could be seen as a new, powerful instrument, enabling new types of creators and potentially leading to entirely new music genres, similar to how DAWs led to EDM. While it makes generic music easier to create, true originality will remain challenging, and a new business model is needed to ensure creators and rights holders benefit, akin to the shift from piracy to streaming.

?
Why did Spotify move away from its famous 'squads' model?

The squad model became inefficient at scale due to high overhead with small, autonomous teams. Spotify found that placing high autonomy at the VP level, rather than at the leaves of the organization, provided a better balance for freedom of thought and effective execution with a more senior, pattern-recognizing leadership.

?
How does Spotify's organizational structure compare to other tech giants like Amazon and Apple?

Spotify leans towards a more centralized model, similar to Apple, prioritizing a simple and coherent single-application user experience across diverse content types (music, podcasts, audiobooks). This contrasts with Amazon's decentralized 'two-pizza teams' model, which optimizes for speed and parallel development at the cost of potential user experience complexity.

?
How does Spotify handle negative user feedback on major product redesigns?

Spotify anticipates negative feedback with redesigns because users' habits are disrupted, even if the change is ultimately beneficial. They separate 'upset due to change' from 'upset due to bad design' by analyzing quantitative data, conducting user research, and comparing new vs. old user cohorts to scientifically prove or disprove hypotheses, remaining unemotional and adapting as needed.

?
What is the '10% planning time' methodology?

This rule of thumb suggests that teams should spend no more than 10% of their time on planning relative to their execution period. For example, in a 10-week quarter, one week should be for planning; for six-month increments, two weeks. If more time is spent planning, either planning is excessive or the execution period is too short.

?
How can leaders bring energy and clarity to their work?

Bringing energy often comes from genuine excitement and belief in one's work, especially when innovating. Clarity is fostered by consistently explaining ideas to others, as this process helps the explainer understand the concept better. Leaders should also commit to explaining decisions to their teams, rather than just dictating them, to maximize organizational effectiveness.

?
What's next for Spotify's podcasting platform?

Spotify is heavily focused on helping podcast creators find more audience and improving monetization options (both free and paid) as the industry matures. For listeners, they are investing in enhancing the user experience, improving ubiquity and playback across devices (like cars), and leveraging AI and generative technology for better search and overall experience.

1. Design Fault-Tolerant UIs for ML

When designing user interfaces for machine learning-driven products, ensure they are “fault-tolerant” by showing multiple options and providing clear “escape hatches” for users. This accommodates the reality that ML models are rarely 100% accurate, preventing user frustration.

2. Prioritize User Need Over Tech Show-off

When building generative AI products, avoid the urge to solely “show off the technology.” Instead, ensure the AI serves the core user need and gets out of the way, enhancing rather than overshadowing the primary experience.

3. Optimize Autonomy at VP Level

For large, scaling organizations, place significant autonomy at the VP level (many tens to hundreds of senior leaders) rather than at the very top or the very bottom. This balances freedom of thought and diverse thinking with sufficient seniority and pattern recognition, avoiding bottlenecks and misaligned efforts.

4. Match Org Structure to Strategy

Deliberately choose an organizational model (e.g., highly centralized for seamless UX, or decentralized for speed) that directly aligns with your core product strategy. Different structures optimize for different outcomes, and the chosen model should support the company’s strategic goals.

5. Embrace “Strong Opinions, Loosely Held”

Be prepared to change your mind and adapt your approach based on data and new learnings, even after investing significant time and effort into a particular direction. This scientific, unemotional approach allows for continuous improvement and prevents being “precious” about initial ideas.

6. Practice Socratic Debate & Explanation

Foster a culture where leaders and team members are expected to clearly explain their reasoning and decisions, rather than relying on seniority or vague “feelings.” This ensures the best ideas win, promotes shared understanding, and forces thorough self-reflection.

7. Separate Feedback on Change vs. Flawed Design

When redesigning a core product, anticipate and differentiate between two types of negative feedback: users upset by any change (broken habits) and users upset by actual flaws in the new design. This distinction is crucial for making informed decisions about whether to revert or iterate.

8. Empathize with Creators by Becoming One

To better understand and solve problems for your product’s creators (e.g., podcasters, musicians), try to become a creator yourself. This hands-on experience reveals the real challenges and pain points creators face, enabling more effective product solutions.

9. Use Internal Podcasts for Culture & Strategy

Create internal podcasts featuring senior leaders discussing company culture, mistakes, successes, and product strategy. This makes leaders more approachable, reduces distance within the organization, and shares strategic thinking in an engaging way.

10. Leverage External Podcasts for Recruitment

Extend internal podcasts to a public format, sharing insights into company culture, product strategy, and leadership. This can serve as an effective recruitment tool, allowing potential employees to feel like they know people at the company and understand its thinking before joining.

11. Limit Planning Time to 10% of Execution

Aim to spend no more than 10% of your time on planning for a given execution period (e.g., 1 week for a 10-week quarter, 2 weeks for a 6-month increment). This rule of thumb helps prevent over-planning and ensures a healthy balance between strategizing and actual building.

12. Utilize Walk-and-Talks for Creative Thinking

Engage in “distributed walk and talks” (e.g., over AirPods) with peers or direct reports to discuss ideas and strategy. Walking can enhance thinking and creativity, and collaborative discussion helps refine ideas.

13. Avoid Short-Term Thinking (“Pee in Pants”)

Be wary of actions or decisions that provide immediate, satisfying results but lead to negative consequences later. This analogy serves as a reminder to consider long-term impacts and avoid short-sighted solutions.

14. Focus on Discovery & Monetization for Creators

For creators on your platform, prioritize efforts to help them find more audience and improve monetization opportunities (both free and paid). These are two of the biggest needs for creators, and addressing them effectively can significantly grow the creator ecosystem.

15. Continuously Improve Listening Experience Ubiquity

Invest in making the listening experience seamless and ubiquitous across all devices, including cars. This enhances user convenience and ensures the product is accessible and enjoyable in various contexts, mirroring successful efforts for music.

I think what we're, we're entering now is we're going from your curation to recommendation to generation. And I suspect it will be as big of a shift that you will eventually have to rethink your products.

Gustav Söderström

If you're hiring smart people, one way to think about it is you're rain, you're renting brain power. So if you're renting all of this expensive brain power, and then you give them no room to think for themselves, that doesn't sound smart.

Gustav Söderström

You have to believe in things a hundred percent until the data says no. And then you believe in something else a hundred percent. That's, that sounds easy. It's very hard to do it.

Gustav Söderström

I think we use the word art and magic. We have historically used the word art and magic for anything that we couldn't yet explain.

Gustav Söderström

Evolution doesn't optimize for seeing the truth just for fitness.

Gustav Söderström
Under 30
Average age at Spotify when Gustav joined (14 years ago) Gustav was the oldest at the time.
7 people
Typical size of a Spotify squad (old model) Comprised of front-end, back-end, mobile QA, agile coaches, etc.
2-3 times that of old squads
Typical size of current Spotify teams Approximately 14 people per manager, with less overhead roles.
Almost 90%
Percentage of Spotify homepage usage for recall (pre-redesign) Users primarily used it to access existing sessions or known playlists/use cases.
10%
Percentage of Spotify homepage usage for discovery (pre-redesign) True discovery, where users didn't know what they wanted.
10% recall, 90% discovery
Shift in homepage usage with the redesign The redesign inadvertently flipped the primary use case, leading to user frustration.