For much of music’s recent history, people listened to the radio, jammed along to jukeboxes, or roamed record stores in search of their favorite artist on vinyl. In the early 2000s, people would load mix sticks, transfer compact discs onto laptops and iPods, and purchase $1.29 tracks from the iTunes store.
But by the early 2010s, everything began to change. Spotify launched in 2008, and began disrupting the music industry — mostly, how users accessed their music. Now, hundreds of millions of people use streaming services daily — and Stockholm-based Spotify is the dominant player, controlling about one-third of the global market for streaming.
For over 430 million users and nearly 190 million subscribers, the service offers a uniquely personalized experience — and over 81 percent of those users cite personalization and discovery as what they love most about Spotify.
“We think of our role as trying to bring a personalized soundtrack to your life,” said Ziad Sultan, Vice President of Personalization at the audio giant, who works out of the company’s Boston office.
The company’s personalization offerings range from the well-known Discover Weekly playlists, featuring customized recommendations for each user; to Spotify Wrapped, which allows users to review their music preferences and data for the year; to the Blend line, which allows for collaborative family and friend playlists that Spotify creates, fusing together mutual interests and recommendations.
And the offerings keep growing. Spotify most recently introduced audiobooks to its platform.
Sultan said Spotify has three major goals: for the platform to be free for users who don’t want to purchase a subscription; to personalize content for users; and to be ubiquitous.
The Goal of Personalization
At Spotify, personalization aims to match users with content creators — so the platform aims to recommend content that will cater to a user’s interests.
“It’s not just about recommending your current favorite, which is already great. It’s also about helping you find your next favorite, and that’s a pretty magical moment,” Sultan said.
But personalization also affects creators whose music and content appears on the platform, he said.
“What’s important to keep in mind is that the mirror image is for every user who finds [or] discovers an artist or piece of content, what happens on the other side is you have a creator who just got one step closer to their dream, and to growing an audience, and to making a living off their art,” Sultan explained.
He said the synergy between users and creators adds value for each stakeholder group, and for the company.
It’s not just about recommending your current favorite, which is already great. It’s also about helping you find your next favorite, and that’s a pretty magical moment.
“We match content and users in a relevant, unique, magical, beautiful way. That’s our goal,” he said. “Think of 400 million plus users on one side, and an infinite number of possibilities on the other. The value is realized when those two things match.”
The number of matches has grown significantly in the past several years. When a user finds a new artist, track, or other content, Sultan calls that a “discovery.”
“Four years ago, we had 10 billion discoveries every month on Spotify. We are now up to 22 billion discoveries every month,” Sultan said.
Increasing Investment in R&D
A careful mix of artificial intelligence, machine learning, and reinforcement learning technologies help to lead the way.
2021 Spotify research concluded that reinforcement learning (RL) allows for greater diversity of consumption, resulting in higher user satisfaction metrics.
“[Reinforcement learning] is this fascinating technology that creates a simulator and allows you to play a certain type of game with rewards that train the agent [for effectively recommending new content to a user]. And that is the latest frontier of stuff we’ve published in our research and that we’re experimenting with in production,” Sultan said.
The company has invested heavily in its research and development this year.
“We invest very much in content understanding and user understanding,” Sultan said. “We are investing in reinforcement learning.”
Per its most recent Form 6-K, in the six-month period that ended June 30, 2022, Spotify spent 30 percent more on funding R&D activities compared to the same time frame in 2021. The company noted that it has recently allocated an additional $100 million to expanding its personnel in R&D. Spotify spends 12 percent of total revenue on R&D, according to the filing.
The company paid nearly $90 million in July to acquire Sonantic, a startup that produces life-like artificial voices that can read text in an array of different emotions.
Understanding The End User
Sultan said Spotify has a wide range of teams involved in personalization.
“We go from research all the way to production. So that means that we have engineering teams, design teams, product teams, data scientists, user research scientists, all working together to go from research to product,” he said.
Those teams work together to understand end users’ desires through a mix of qualitative and quantitative research.
A data science team and user research team sit under one umbrella, and form Spotify’s Product Insights team.
“We have a very strong data science team that is able to look at our data across the board, and extract amazing insights in a very rigorous way. So, we know what is happening, but we also know if it’s causal, or if it’s seasonal, or if it’s a fluke, or if it’s correlated with something but it wasn’t really the cause. So we can dive really deep into the data, which helps with one type of evaluation,” he said.
A group of 30 user researchers globally helps to analyze qualitative data to draw insights about preferences and trends.
Advice for Companies Trying to Create Better Personalization Experiences
While some industries have fewer opportunities for personalization than others, Sultan said many companies can still find ways to personalize experiences for their markets.
You can go very far with some simple, rule-based personalization
He said being data driven and focused is one key to creating a strategy for personalization.
“You need to understand what is happening with your users and with your product, and have it really well instrumented, so that you can query it and understand really what’s going on,” he said.
He also recommended starting simple and getting more sophisticated in stages.
“You can go very far with some simple, rule-based personalization, if your product is not millions of choices, and millions of users that need to be matched,” he said. “I think it’s really important to build it up in stages.”
Making a Bet
Sultan said Spotify is a widely loved product, which means a number of people have ideas and opinions about where it should go next.
“That’s the beauty of working on a product that you and your friends and family love, is that they will all have opinions. That is also true inside the company. A lot of people will have ideas. And in some cases, they will be amazing,” he said.
I think the challenge in general with innovation…is that there are a lot of innovative things that are great, that should be launched in the world, that you cannot arrive at through a purely rigorous, linear, mathematical way. You need to suspend disbelief; you need to make a bet.
He said a key issue for innovators is knowing when to be data driven about what they should build next — and when to take a leap based on intuition.
“I think the challenge in general with innovation — which is my favorite challenge, actually — is that there are a lot of innovative things that are great, that should be launched in the world, that you cannot arrive at through a purely rigorous, linear, mathematical way. You need to suspend disbelief; you need to make a bet,” Sultan said. “There is some part of what you need to do that will be data driven… and some part that will be a bet, that we don’t know if it [will be] successful.”
Spotify has strong mechanisms in place for gathering insights about users and running A/B tests. But Sultan said the harder part of balancing data and risk is making bets that could be wrong — or could lead to an important new strategic direction for the company.
“In order to be good at making bets, we spend a lot of time as a team and as a company, debating strategy, building trust in each other’s judgments, knowledge of the systems, creating psychological safety, so people are willing to make a bet and be wrong,” Sultan said.