Spotify A Product Story Summary – Part 1

11 Apr 2022

Spotify: A Product Story is a podcast about all the major decisions which shaped Spotify into the product it is today, discussed by the decision makers and other people who influenced the decisions.

This two part post will be a summary of the lessons in the podcast. Part 1 will cover Episodes 1 to 4 and the other 4 episodes will be covered in part 2.

Episode 1: How do you Steal from a Pirate?

This is the story behind Spotify’s first product, the desktop app. We explore how the first iteration of Spotify came to be, and identify the four biggest lessons learned through the process, lessons that still shape the company to this day, almost 15 years later.

Lesson 1: Convenience trumps everything

Sweden, where Spotify is from, had piracy ingrained in its culture, but they were betting on making it easy to play a song in Spotify than it is to pirate it.
They were betting that the speed of listening to a song and the availability of a vast catalog of songs would make people switch from piracy to Spotify.

Lesson 2: If you try to do something fundamentally groundbreaking, break existing standards and think full stack

Spotify decided to build a desktop app, as they were limited by the browsers of that time. This would allow them to have a much larger extent of control over the entire tech.
All Spotify clients would be in a P2P network where the recently streamed songs are in a local cache and is sent to other clients like torrents.
Anything time critical like the 1st 30 seconds of a song streamed from the servers, while other non-critical data would be fetched from the P2P network.
They established a persistent TCP connection to the server, which allowed them to counteract the TCP slow start problem
A heavily customised version of the open-source audio codec Vorbis was used instead of mp3 to improve latency.
They essentially built a full stack custom media distribution solution and streaming protocols.

Lesson 3: What attracts great talent is the level of ambition

People like Ludde Strigeus, the guy who created μTorrent, and many others joined the company because they saw the level of ambition and a clear long term thinking and planning from the executives.

Lesson 4: Every product needs a magic trick

A magic trick, in this context, can be interpreted as something that no one thought was possible
Spotify’s magic trick was songs instantly playing as soon as the user clicked the play button. It gave an impression of having all the world’s music in your device.

Episode 2: How do you Charge for Nothing?

This episode tackles how Spotify went from being a popular free desktop app that nobody paid for, to being a mobile app that people actually paid for, even though users already had all their pirated music on their phones for free.

Lesson 1: User research is great, for understanding what people think they will do, not what they will actually do

When people respond to questions, they’re predicting what their future self will do, not what they’ll actually do.
We never like to think of our future selves as lazy or in an unflattering light, so it is not really accurate.
If you have a strong hypothesis on why people would react differently than what they tell you, don’t be afraid to try it out.
When they conducted user surveys, almost all the users said they would never pay for a service like Spotify.

Lesson 2: Whether you realise it or not, you’re always optimising for something

Spotify’s engineers were initially optimising for reducing the play time, sacrificing bandwidth and power consumption in the process.
This turned out to be a problem in the context of mobile devices because they had limited power and bandwidth.
This is neither a good thing nor a bad thing, just make sure you’re aware of this fact.

Lesson 3: Really great product development almost always combines technological innovation with business model innovation

Spotify was a technically great product, but what set them apart was their focus on acquiring licenses from all the record labels and negotiating the terms to their advantage.
Spotify also got the pricing on their subscription right and had great user conversion. At that time Skype was leading the conversions game with around 7% while Spotify came in and blew it away by getting 20% of the users to pay for their service.

Episode 3: This Party is going to End

This episode tells the story of how Spotify went into negative growth and nearly didn’t survive to its 5th birthday as the world started adopting smartphones at a staggering pace. They also talk about how they had to reinvent their entire business model and find a new free tier that worked in a mobile first world.

Lesson 1: Never try to fight a macro wind, you will lose

When a macro, something much bigger than a product or an industry, changes, the best thing to do is to adapt to that change.
A good way to figure out if something is a macro or not, is to ask “How many of X are there?”, X being the thing in question. If that number is much bigger than your user base, then it is a macro and can affect your product in some way.
Mary Meeker’s 2010 internet trends report predicted that the smartphones will outship PC shipments within the next two years, and it will change the way the users access the internet.
This meant that Spotify had to adapt its entire product to a mobile first world.

Lesson 2: Test big or go home – but remember: Test as different as possible, not as much as possible

When Spotify was settling on the new free tier for the mobile app, they had a bunch of different hypotheses, like
- Limiting play time
- Giving a limited music library
- Restricting the amount of times a song can be played
- Having no content restrictions but only shuffle play
They started A/B testing all the hypotheses to a large portion of the users.
This led to the revelation that the lesser the restrictions on the free tier, the higher the conversion to paid users.
The goal is not to go deep on one idea, it is to cover as much area as one possibly can.

Lesson 3: Look at the competition, and then do something completely different

Spotify’s competition were releasing paid-only apps or launching free radio like services, Spotify looked at them and figured out both the approaches are not good.
Studying the competition isn’t a bad thing, as long as you’re using those insights to chart a different path.

Lesson 4: It’s better to be lower on a taller mountain, than higher on a smaller mountain

This means it’s better to farther behind on a path with greater potential than ahead on a path with less potential.
Spotify decided to go from a playlisting based mobile tier to a free tier where songs can be played on demand as long as it appears in one of your curated playlists.
This meant that it took years of massive engineering efforts to even be on par with the older model. This began raising some doubts among executives, but they saw the potential of this model and decided to stick with it, and it has paid massive returns in the long run.
Follow your convictions, check the data to back them up, and make sure that you have enough patience for the climb.

Episode 4: Human vs Machine

This episode is on what it really means to develop products in an AI/ML first world, how they invented the term “Algotorial” and how reinforcement learning applies to music.

Lesson 1: Build for yourself first. But don’t build for yourself only

The people who started Spotify were avid music fans, and they started developing the product based on their instincts, and it was a huge success among other music aficionados.
This was a great thing at the beginning, as they could understand the users’ need and build towards it.
But when casual listeners started making up more and more percentage of the total users, they realised they had to cater to them if they wanted to grow.
This led them to start showing interest in creating curated playlists for the user.
Designing with yourself in mind is a good place to start, but don’t limit your potential to people just like you.

Lesson 2: Throwing machine learning at data you don’t fully understand isn’t enough to give you a great product. You need to understand your data deeply

Spotify was using an approach called collaborative filtering to curate playlists.
This looked at all the existing playlists and tried to create new playlists.
It only took the appearance of two songs together in enough number of playlists to determine that the songs were related.
In reality, this is not how things worked, a rock song and classical ballet will appear in the same playlist because there are a lot of people who would enjoy them together, but that doesn’t mean that they’re good candidates to be in the same playlist.

Lesson 3: If you don’t have one side of an equation inside the company – Look outside for it

Spotify didn’t have the AI expertise needed to build a truly great recommendation engine in-house, so they acquired a bunch of companies that did it.
One thing to consider when making such acquisitions is how many years of time and effort will be saved when we go through with the acquisition rather than doing it in-house.
Also, what will be the impact of that acquisition in your product.

Lesson 4: In a machine learning world, you have to learn the product, not build it

In practice, this means traditional SDLC practices would not work when building a machine learning product.
While this is a well known fact now, it was something Spotify learnt the hard way.

Conclusion

The summary of the next 4 episodes will be covered in another blog post.