Throughout music history, we have witnessed countless recipes and pathways for artists to reach success. While every artist’s formula is unique, hence why they stand out from the rest, there are some technical concepts that are important for artists to understand, including the notable Spotify algorithm. For artists who want to increase their listener and fan base, Spotify is an essential DSP (Digital Streaming Platform) for music releases, and understanding the algorithm can be a saving grace or a roadblock for their discography.
Unlike social networking platforms such as Facebook and Instagram, which have a central algorithm that impacts stories, ads and influences that users receive, Spotify actually does not have a central “Algorithm.” Instead, the service employs a series of machine-learning models that can be thought of as algorithms that help recommend music to users.
This process is referred to as the Spotify Recommendation Engine that is built with three machine-learning methods:
1. Collaborative Filtering
2. Natural Language Processing
3. Audio Data
Collaborative Filtering
The term “Collaborative Filtering” represents that main machine-learning model that drives Spotify’s recommendations, and incorporates a big database, similar to a spreadsheet, that is called a “matrix.” The rows in this matrix identify Spotify users through their User ID, while the columns feature the Song IDs. This totals to over 230 million rows and over 60 million columns!
Each song ID and User ID is given a “vector” that is calculated by Spotify, which shows how the song and the user relate to each other. This process allows Spotify to identify and track the songs, artists and users that are relevant to each other, making recommendations based on those matching preferences.
Natural Language Processing
The term “Natural Language Processing” might not sound like a music term to the average user, but for artists, it’s an important process in understanding the Spotify algorithm. NLP is an artificial intelligence process in which languages are translated for computers to understand the sound and descriptions of an artist and their releases. Spotify runs through thousands of sites and blogs that contain content about the artist and their songs, establishing relevant connections between similar artists and those with similar sounds. Blogs, PR releases, biographies, social media metadata and playlist placement campaigns can all increase an artist’s chances of securing a spot on a highly streamed playlist.
Previously, user-generated playlists were not targets for Natural Language Processing, but they now currently are if Spotify considers your playlist to be strong, meaning they have many listeners and are updated frequently. Considering there are over 2 billion playlists on the platform, there are many playlists that aren’t effective indicators of which artists and songs should match together. As an artist, be sure to target playlists that contain these traits.
An example of this is if a user who has searched and streamed a lot of J. Cole’s music, then they will more than likely get many recommendations for Jay-Z, as Cole was once signed by Jay, has features with him, and has stated he is heavily influenced by his sound and style many times in interviews.
Audio Data
Audio Data, which is the third machine learning model impacting the Spotify algorithm is also known as Signal Processing and Analysis, and is where the music, itself, is dissected. The Audio Data stage deciphers the characteristics of songs, such as the key the song is in, the tempo, the rhythm, the time signature and many other important sound elements.
While Collaborative Filtering and Natural Language Processing models dive into the impact of the music on users and how it relates to the impact and popularity of other releases, the Audio Data model directly plunges into the music and pinpoints resemblances to other artists’ sound. Using this audio data, Spotify then will connect the similarities of one song to the data of another song ID. Interestingly, this process can help Spotify to accurately place a song ID even if it was not referred to on the web or has not yet been streamed.
In order to qualify for audio analysis on the first week of a song’s release, an artist needs to submit the song at least seven days prior.
Algorithmic Playlists
Now that you have an idea of the three main machine-learning models that modify and effect the Spotify Algorithm, let’s take a look at the placements where you can use the algorithm to your advantage as an artist.
If you’ve scrolled through Spotify’s Home, Search or Browse section on their platform, then you’ve seen tons of top playlists that have been created by Spotify’s Editorial Team, such as Rap Caviar and Happy Hits!. While all of these playlists can increase an artist’s listener base, there are only two playlists that can be modified due to an artist’s marketing labor: Release Radar and Discover Weekly, These specialized and custom-tailored playlists are generated by Spotify’s Recommendation Engine and are produced by user’s listener habits according to the algorithm.
Landing a Spot on Release Radar
Release Radar, which is updated every Friday, requires audio data to play the primary role. Majority of the songs on this particular playlist are typically newer releases and do not have enough collected data to be recommended to users. This can feature artists that users follow, artists they listen to or any artists that Spotify thinks that user will like. For releases that don’t have enough listening data, Spotify developers have deemed this issue as the “Cold Start” problem.
Not only can your song be recommended to users based on its audio data, but any monthly listeners you’ve been able to rack up also have a chance of hearing your new song, as data shows that monthly listeners are connected to your Artist ID.
There are two main criteria points that a song must meet in order to be placed on Release Radar:
1. The song must be relatively new – 2-3 months old at the most
2. The song must be eligible to be recommended to a user based on listening data or audio data.
Unfortunately, listeners (who are not following your Spotify profile) are not informed of new releases the same way that followers are, but they are still very viable candidates for receiving your music in their Release Radar tailored playlist. Followers, however, receive a call-to-action email to listen to their new release on Spotify, giving them a more direct, in-your-face reminder to check out an artist’s music.
The best way to potentially being included on the Release Radar playlist is to submit your song for consideration at least 7 days before the song’s release. You can do this by using the “Spotify for Artists” dashboard under the “Music > Upcoming” section. To propel the hype of a release, be sure to publish performance videos on social media so you can retarget users with the Spotify link to the song. It’s also important to understand that Release Radar is cumulative, meaning that users who save your song from Release Radar increase your chances of reaching more users during the next Release Radar drop. According to most music marketers, songs rotating into the Release Radar playlist for a user typically peak in third of fourth week of release and are pushed out of distribution around the 12th week.
Landing a spot on Discover Weekly
The Discover Weekly playlist plays a different role than the Release Radar playlist in how it selects songs, which involves more of a guessing game for each user. Spotify has developed a training system for its algorithm which utilizes user-generated playlist estimates to create its track list. Spotify will trace a user-generated playlist that has one song removed, which pushes the machine-learning models to guess which song is missing from the playlist.
In order to achieve this, the streaming platform uses a playlist comprised of a user’s listening history, which is where the model must “guess” the missing song. Once there is enough accuracy recorded, Spotify then initiates these recommendations to be placed on the Discover Weekly playlist for a user. Essentially, Discover Weekly is a space for Spotify to predict what is missing from the soundtrack of a user’s life.
The two criteria points for recommended song placement are:
1. The user it’s being recommended to has never played the song.
2. The song must be close to the user in the “vector” space, based on existing listening or audio data.
Discover Weekly pulls its placements based on its popularity ranking, its playlist placement, its media/blog coverage, its metadata and its streaming data.
What is the Spotify Popularity Index?
One of the most crucial factors contributing to artist’s placement in the Spotify algorithm is their Popularity Index. The Spotify Popularity Index, which measures music’s influence and impact, distributes a 0-to-100 score that ranks how popular an artist is relative to other artists on the platform. This measurement helps artists to get more placements on editorial playlists that will help increase their reach on the algorithmic playlists. Anytime an artist releases a new song to the streaming platform, it receives its own SPI that impacts an artist’s index. The Popularity Index doesn’t just take the number of streams into account for a song’s ranking, but also the save rate, the playlist placement rate, the skip rate and the share rate.
Remember, the more you increase your numbers on the Popularity Index, the better chance you have for the algorithm to recommend you to new listeners and provide you with algorithmic playlist placements. The first week and any early data collected will help a release’s initial popularity and produce more streams for a longer period of time. If you need to check your Spotify Popularity Index, there are a few resources that can be used to monitor, including Musicstax, Chartmetric, Songstats and for those with more coding knowledge, Spotify for Developers.
Take into account: If your song receives a score of 20+ within three weeks of the release date, this will help the algorithm to work in your brand’s favor, as you are now potentially eligible to be placed on Release Radar. With a score of 30+, which includes 15,000 streams and a 40-50% save rate, you may be able to secure a spot on Discover Weekly. A great technique to increase these odds are through Pre-Save campaigns that give artists permission from responsive fans to save their new track to their Liked Songs collection as soon as it hits the streaming service. By securing an ample amount of saves on a song’s first day can artificially influence the algorithm to put you at a higher ranking in the Popularity Index.
More on the “Cold Start” Problem
Recommendations that come from a new release can be determined off of a slew of components, including the tags of the song, the artist and album information, the lyrics, the audio signal as well as any reviews or interviews on the Internet. When there is no usage data to analyze, such as when a new user signs up for Spotify, and has no previous listening data or history, this begins the “Cold Start” problem for the DSP. For artists, this starts when they create their profile, especially if they have a nearly non-existence digital presence outside of the app. If there is no user data to analyze, the collaborative filtering process is pretty much voided.
Tackling the “Cold Start” issue has been a priority for Spotify developers as they continue to evolve their algorithm. The platform has created an analysis engine that examines songs to identify their tempo, key, pitch, genre and other musical factors and then uses this audio data to compare it to other similar songs on Spotify and connect users to recommended songs.
To avoid: Be sure to initiate some sort of an online presence before a song’s release. This will allow Spotify to use the NLP machine learning model to try to establish and identify your sound, even without any prior releases. Whether you have a PR release, an interview or a biography out on the web, Spotify will be able to use this data to figure out who and where to recommend you to later.
User-Generated Playlists
While algorithmic playlists take a lot of technical understanding and go by numbers and statistics, user-generated playlists involve more of an artist’s interpersonal effectiveness, as well as their judgment. User-generated playlists are typically made by a Spotify user, may be in a specific or no particular order, and may contain songs that wouldn’t be previously matched together through the algorithm.
Many businesses, organizations and influencers curate playlists that will help expand their brand, with the option to make them public or “secret.” In order to be successful with placements on user-generated playlists, artists need to be comfortable with researching and reaching out to third parties. Artists will need to not only understand the playlist and the listener base, but also the taste of the user who created it. As a maker of music, yourself, you will need to reach out to the creator, validate their list, and establish a relationship with them. In many cases, you may need to be ready to pay up, as many popularized playlists give away slots at a cost.
Keep in mind: The downside of user-generated playlists is that some highly streamed collections on Spotify can be popularized due to bot traffic or unethical tactics. These playlists are typically used by non-premium members in countries where advertising CPMs are low, which results in a low payout for the streams stemming from these playlists.
Editorial Playlists
With one scroll of the Home screen on Spotify, users will come across a batch of Spotify playlists that have been created by the company’s Shows & Editorial team. These employees consist of international music experts and genre specialists that curate and manage the platform’s own playlists, which are typically theme/genre-specific, have large followings and are more notorious to the Spotify brand.
Some of the more outstanding editorial playlists include Rap Caviar, Most Necessary, Are & Be, Hot Country and Channel X. There are a large deal playlists that cater to eras in musical history, such as the 80s, the 90s, the Golden Age of Hip Hop, 2000s R&B and many, many more. It’s been noted that Spotify editors tend to test out new music on “smaller” playlists such as Most Necessary, which has 1.8 million followers, before placing them on a large-scale playlist like Rap Caviar, which has over 10.3 million followers.
Having a clear understanding of the Spotify algorithm can make or break your impact on the streaming platform, as well as how far your music is reaching. While the sound of your music and your talent as an artist is always essential to your brand, catering to your current and future fan base with powerful, engaging marketing tactics is the best way to navigate towards success. While your accomplishments don’t have to start and end with Spotify, it can certainly help you maneuver away from only receiving recognition in your local neighborhood to establishing your legacy around the globe.
