Spotify is the world’s most popular audio streaming service with 433 million users (including 188 million paying subscribers) across 183 different markets—and artificial intelligence is absolutely essential to the company’s success.
“Machine learning is at the heart of everything we do at Spotify,” said Tony Jebara, the company’s VP of engineering and head of machine learning, at TensorFlow World 2019.
Jebara should know. Before working at Spotify, he used to fill a similar role developing AI-powered recommendations at Netflix.
And, like Netflix, Spotify depends on several forms of artificial intelligence to recommend content that satisfies users and creates value for paying subscribers.
Understanding Spotify’s use of artificial intelligence is valuable for two reasons:
First, understanding how Spotify uses AI helps us understand how to better use the platform to market our podcast, The Marketing AI Show, which is distributed through Spotify and other platforms that use AI in similar ways.
Second, even if you don’t have a podcast, Spotify is a sterling example of how important it is to build AI Native and AI Emergent companies—no matter your business.
In the near term, businesses that don’t embrace AI will become obsolete. Spotify offers a helpful case study to understand what’s possible when building AI-first products and companies.
We’re uniquely positioned to break down how Spotify uses artificial intelligence.
In our daily work at Marketing AI Institute, we track thousands of AI vendors and evaluate and pilot hundreds of solutions across different areas of marketing and our business operations. We also help businesses in many different industries understand how AI technology works and how they can use it to grow.
Spotify is just one example of the power of AI in action, so let’s take a closer look.
How Does Spotify Use Artificial Intelligence?
Spotify uses AI in the form of machine learning and deep learning in a few core ways.
First, Spotify’s AI models recommend audio content to users. Spotify’s AI powers music and podcast recommendations designed to create long-term user satisfaction.
To do that, Spotify leverages user data, from playlist creation to listening history to how you interact with the platform, to predict what you might want to listen to next.
These recommendations are served up through Spotify’s Home screen, which is divided into rows of cards containing both recently listened to audio content and new, recommended content based on your behavior.
Specifically, Spotify leans heavily on reinforcement learning, a type of machine learning model that uses environmental signals to optimize toward a long-term reward. The long-term reward Spotify dangles in front of its machine learning model is long-term user satisfaction.
So, at every turn, Spotify’s recommendations are designed to increase your overall happiness with the product, which keeps you coming back to listen more and more often.
Spotify’s AI-powered recommendations are the company’s competitive advantage.
Spotify doesn’t have a monopoly on music streaming; you can easily listen to your favorite songs in many places. What it does have is a superior way to surface the right audio at the right time for you.
This ability to hyper-personalize at scale is impossible without AI—meaning, in a very real sense, Spotify as a business would not be possible if AI didn’t exist. Spotify is the very definition of an “AI Native” business—a business that unlocks unprecedented value in a market by reinventing a product or service from the ground up using AI.
In this way, every single user sees a hyper-personalized, highly customized version of the product. According to the company, half a trillion events are processed daily to inform machine learning models. And the more data these models gather, the better they are at making higher-quality recommendations.
Because this is all done at the individual user level, the company says that Spotify isn’t actually a single product; it’s 433 million different products—one for each and every user.
Playlists Created and Curated by AI
Second, Spotify takes recommendations further by using AI to create entire playlists. These AI-powered playlists take a handful of different forms.
The Discover Weekly playlist is an automatically curated playlist of new and existing music you’re likely to love based on your listening habits.
There’s also the Release Radar, which automatically creates a playlist each week of the latest music from artists you follow and personalized recommendations based on those follows.
The Home screen also contains a variety of recommended lists of podcast episodes you might like and mixtapes based on moods, activities, and environmental factors like the day of the week or time of day.
This is another area where the company’s reinforcement learning comes in:
Through playlists and recommended content lists, the model attempts to nudge you towards audio options that will make you more satisfied.
After all, the more satisfied you are, the more you’ll listen.
Natural Language Search
Third, Spotify uses artificial intelligence to power natural language search.
Previously, Spotify used exact words typed into a search bar to match content to queries. This approach was better than nothing, but it was limited: it didn’t always provide high-quality results because it could only match terms very close to those used in a song, album, or podcast title.
Natural language search is different. Using AI technologies like natural language processing (NLP) and deep learning, natural language search understands the semantic correlation between words, so it doesn’t need to exactly match your search with the words in a title to find what you’re looking for. That’s because natural language search can understand synonyms for different words, paraphrasing, and any content that means the same thing as what you searched.
Today, Spotify has rolled this feature out for podcast search, making it dramatically easier to find podcasts on relevant topics—even if your search doesn’t exactly match what’s described in the podcast metadata.
Spotify’s Artificial Intelligence Investments and Acquisitions
Spotify has acquired AI companies, or companies related to AI, at a steady clip for almost a decade.
- In 2013, Spotify acquired Tunigo to power better music recommendation algorithms.
- In 2014, the company acquired Echo Nest, a music intelligence company that Spotify used to improve recommendations.
- In 2015, Spotify acquired data science company Seed Scientific.
- In 2017, Sonalytic was acquired by Spotify. Sonalytic uses machine learning to detect audio and recommend music.
- Also, in 2017, Spotify acquired Niland, an AI startup, to optimize music searches and recommendations.
- In its most recent AI-related acquisition, Spotify acquired Sonantic, an AI-powered text-to-speech generator.
In 2018, Spotify also started a regular event called Machine Learning Day, which brings together company researchers to discuss core topics in AI.
How Spotify Could Be Using Artificial Intelligence
There’s no question AI is critical to Spotify’s business. So, where is the company going from here?
At Marketing AI Institute, we help companies of all types build smarter businesses with AI. When we do so, we evaluate a company’s entire business and determine ways they can use AI to accelerate revenue and reduce costs.
We don’t currently work with Spotify, but we do have some possible AI use cases that the company may want to consider exploring as part of its product roadmap.
- Create AI-assisted or completely AI-generated songs and albums.
- Identify the most engaged fans to target with highly personalized marketing offers related to albums, events, and merchandise.
- Identify top marketing channels and recommend the next actions with the highest chance of growing a fanbase and/or generating revenue.
- Inject hyper-contextual ads into streaming sessions and playlists based on user behavior and interests.
- Launch hyper-personalized virtual concerts with realistic AI-powered artist avatars that perform a fan’s favorite songs just for them.
- Predict what type of music to create for fans next.
- Automatically summarize podcasts and create text-based descriptions of episodes using natural language processing and text summarization.
- Automatically rate podcast speakers on speaking time, narration quality, and other voice factors to improve podcast quality using sentiment analysis and voice recognition.
- Create AI-powered profiles of podcast guests to accelerate podcast host research.
- Identify the most popular or engaging podcast topics, episode formats, lengths, etc., so creators can optimize the content and format of their podcasts for maximum consumption.
- Predict which guests a podcast host should have on next, based on popularity, relevance, and influence.
With Proprietary Data
- Use AI on music data to predict which musical acts stand the highest chance of success, then sell this information to agents, labels, and brands.
- Use AI on podcast data to predict which podcast topics and formats stand the highest chance of success, then sell this information to podcast product companies and influencers.
- Use AI with user behavioral data to advise brands on the best, most effective ads in a given niche or market.
And that’s just the tip of the iceberg. I’m sure there are dozens, if not hundreds, more ways that Spotify could be using AI in the near future to transform business as usual in the music and podcast industries.
So, now that you know how Spotify uses artificial intelligence to thrive as an AI Native business, isn’t it time your company benefited from AI as well?
We’ve created a resource that helps you do just that…
Piloting AI for Marketers teaches you step-by-step how to use AI to make your business more productive and successful. Piloting AI features 17 courses and 6 hours of content for one special pre-sale price. Click below to learn more.
As Chief Content Officer, Mike Kaput uses content marketing, marketing strategy, and marketing technology to grow and scale traffic, leads, and revenue for Marketing AI Institute. Mike is the co-author of Marketing Artificial Intelligence: AI, Marketing and the Future of Business (Matt Holt Books, 2022). See Mike's full bio.