The Music Business Buddy
A podcast that aims to educate and inspire music creators in their quest to achieving their goals by gaining a greater understanding of the business of music. A new episode is released each Wednesday and aims to offer clarity and insight into a range of subjects across the music industry, all through the lens of a music creator for the benefit of other music creators. The series includes soundbites and interviews with guests from all over the world together with commentary and clarity on a range of topics. The podcast is hosted by award winning music industry professional Jonny Amos.
Jonny Amos is a music producer with credits on a range of major and independent labels, a songwriter with chart success in Europe and Asia, a senior lecturer in both music creation and music business at BIMM University UK, director of The SongLab Ltd and the author of The Music Business for Music Creators.
www.jonnyamos.com
The Music Business Buddy
Episode 41: Inside The Spotify Algorithm
The secretive Spotify algorithm stands as arguably the most advanced digital recommendation system ever created, yet few truly understand how it works or how to leverage it effectively. As a music producer, lecturer, and industry consultant, I've dedicated countless hours to researching and testing how this mysterious system operates behind the scenes.
What I've discovered is that Spotify's algorithm functions through six interconnected elements. Natural Language Processing scans the internet for discussions about your music, collaborative filtering connects listeners with similar tastes, contextual information considers when and where music is being played, audio features analysis examines the technical qualities of your recordings, machine learning models predict listener preferences, and user feedback refines recommendations through skips, saves, and playlist adds.
For music creators, understanding these elements provides crucial insight into how your music travels through Spotify's ecosystem. Rather than trying to "hack" a system too sophisticated to manipulate, success comes from appreciating how your music is categorized, discovered, and shared. The metadata you submit matters—if you misidentify your genre, Spotify's own analysis might flag your submission as risky, potentially slowing your discovery rate.
Beyond the technical aspects, I explore the three types of playlists driving music discovery on Spotify. While many artists obsess over landing on editorial playlists like RapCaviar or Today's Top Hits, algorithmic playlists often provide a slower but more reliable path to genuine audience growth. Even more surprisingly, placements on influential user-curated playlists can sometimes generate more streams than editorial features. By understanding these dynamics and considering your own consumption habits as a listener, you'll gain valuable perspective on how to position your music for maximum visibility and engagement.
Have questions about navigating Spotify's algorithm for your music? I'd love to hear from you—reach out anytime to continue the conversation!
Websites
www.jonnyamos.com
https://themusicbusinessbuddy.buzzsprout.com
Instagram
https://www.instagram.com/themusicbusinessbuddypodcast/
https://www.instagram.com/jonny_amos/
Email
jonnyamos@me.com
The Music Business Buddy. The Music Business Buddy. Hello everybody and a very, very warm welcome to you. You're listening to the Music Business Buddy with me, johnny Amos, podcasting out of Birmingham in England. I'm the author of the book the Music Business for Music Creators. I'm a music creator with a variety of credits myself. I'm a consultant, an artist manager and a senior lecturer in both music business and music creation. Wherever you are and whatever you do, consider yourself welcome to this podcast and to a part of this community. I'm here to try and educate and inspire music creators from all over the world in their quest to achieving their goals by gaining a greater understanding of the business of music.
Speaker 1:Okay, so in today's episode, we are talking about the Spotify algorithm, the great mystery behind it, what it actually looks like, what we need to bear in mind as music creators and how we actually feed it. Okay, let's get going. Okay, so understanding how Spotify's algorithm is built is somewhat comparable to understanding the recipe for McDonald's Big Mac sauce. Ie, it's a secret, right? It's undisclosed propriety information that nobody has the legal right to publish or to talk about, right? There's probably major executives at Spotify that can't fully understand or explain it. I might be wrong about that, but I bet I'm not. However, through studies, through research and through testing, we can come pretty close to understanding how the algorithm is pieced together by exploring some of its key elements. So it's not possible for music creators or anybody else for that matter to hack or to cheat their way to growth. The algorithm is way too sophisticated for that. In my view, it is the most advanced digital recommendation system ever known to man. There are ways, however, that music creators can feed that algorithm, and it starts with understanding you know it's probable fundamental operations. So I see six parts to the algorithm and let's get into it. Okay, element number one natural language processing. I've spoken about this before on the podcast. It's a big part of many different algorithms and Spotify most certainly use it.
Speaker 1:Nlp, if anybody has never heard of it before natural language processing. It enables technology to understand text and speech. Enables technology to understand text and speech. Spotify uses the findings to inform its understanding of a song and an artist and their place in the market. So NLP will scrape the internet and search for information that they can find on a song and an artist, and this could include such things as what people are saying in forums, what music reviewers are saying in publications, what bloggers are writing about the song, and more. Spotify seeks to understand how the song is being heard and understood outside of its platform. So NLP also aids Spotify in analysing the song's lyrics, which can help to inform them as to kind of key themes and moods when they look at kind of paradigmatic analysis of themes Something which is very useful, by the way, in the context of creating recommendations. You know NLP, so it's used for song categorisation and also metadata tracking. So to summarise that it's words, people's words, words in the song, words about the song, all sorts of anything to do where it can grab text from the internet, from web harvesting, to inform them of what is going on with that song outside of Spotify. Remember, spotify want songs to be big on their platform, but they don't want songs to be big on their platform. But they don't want them to be huge on their platform and completely unheard anywhere else. It makes them higher risk. So one of the ways that they negate that risk is by using NLP.
Speaker 1:Okay, element two collaborative filtering. So this is where the platform analyzes the listening history of a user and searches for other users with similar tastes and then groups, them. Consumers don't know that that's happening, but that is happening right. The way I think about collaborative filtering is is to have, like imagine, sort of 20 trees inside of a park. You know they're all. They all kind of stood close to each other. They may not communicate, but their roots are touching right. That think about about it. It's a weird analogy but if you're still with me on that one, it really means that there are combined interests that all of us have that we don't necessarily communicate verbally. So there's certain tendencies that we have when we listen to music. Maybe our guilty pleasures are listening to this band at a certain time through a certain way. All of those things are collaborative filtering and they are drawing the tastes between people. So once those patterns have been identified amongst listeners, comparisons are then drawn between them and preferences are based on like songs and artists and playlists, moods, and once Spotify has a greater understanding of a listener's profile, that listener is introduced to music that they have not yet heard on the platform but that has already been enjoyed by listeners with similar tastes. I hope that makes sense. Remember that. You know, new music is absolutely crucial to the entirety of how the Spotify algorithm functions. But new music doesn't necessarily have to be things that are released last week or this week or next week. It's new to that listener. So it could be that you've got songs that are sat there, kind of semi sort of dormant, haven't been, you know, streamed regularly for three years. All of a sudden they get a huge boost and it's because of something else that you've done that's then triggered that aspect of collaborative filtering.
Speaker 1:Element three contextual information. Okay, so Spotify tries to kind of build in context to the suggestions it serves to listeners to curate experiences that cater to different lifestyles. So, for example, the time of the day, the week, maybe even the time of the year certainly with seasonal stuff can play a part in what it is offering to a listener. So a listener's geographical location also influences this, to offer a bit more of a sort of advanced level of suitability for the right song at the right moment for the right person.
Speaker 1:Element four audio features. So we've looked at the text, we've looked at some of the information around that. Now we're looking at the actual audio. So the recording of a song, so each song's recording is analysed to extract its key features. Some people know this as raw file analysis, which is something that Spotify, and many other algorithms for that matter, do in order to understand the characteristics of something. So, while the mood of a song can, you know, perhaps be open to sort of human interpretation, spotify can scan a waveform of a recording and assess factors such as energy, danceability, rhythm, speechiness, loudness, things like that, things that you can't really kind of see with the naked eye, if you will. So this process helps Spotify to categorise songs into primary and secondary genres, from a technical perspective rather than a human perspective. So this process provides a kind of deeper level of accuracy when it comes to recommendations and can also assist with playlist curation.
Speaker 1:Now let's just think about metadata for a minute, because it might be that you're thinking right now well, hmm, I send that information, that metadata on primary genre and secondary genre, to your distributor. Okay, now I'm going to tell you something right now that I don't know if it's true, but I do believe it is. If you get your metadata wrong, ie if you were to say that something is classic rock and it is actually deemed to be indie disco under raw file analysis, I think it makes the song a bigger risk, because it means that we don't know what we're selling and it also then means that the categorisation on it becomes slower. So, audio features element four is a really, really crucial part, because it will really help Spotify to understand exactly what your recording is going to do for a listener. It's very important that we try and build that understanding ourselves and I know I've spoken about this on the podcast before, but it's a very important point to make Try and know what you're selling, know what mood it curates, know what feelings are associated with it, what the lyrics are doing, what the music is doing and, most importantly, understanding the primary genre category and the secondary genres that support it. It might be that your distributor has categories that you can select, but those selections are not actually applicable to a particular streaming platform. For example, we know that Amazon, apple and Spotify all categorise things differently, so they will actually analyse the raw audio features of a song differently. So all we can do when we send music through a distributor, through a label, is to try and inform them as best we can on its categorisation. It's not the end of the world if those things are inaccurate, but it certainly slows down a song's discovery rate because of how this element works.
Speaker 1:Element five machine learning models. Ok, so machine learning, or ML as some people call it, is at the heart of what Spotify is and what it's all about. So they use a host of different sort of ML techniques to analyse various factors based on user data. This is possibly the most crucial aspect, because it is what makes Spotify so popular, as the platform has been designed to offer this kind of hyper-personalised recommendation to users. So it is assumed that Spotify uses a very highly sophisticated form of reinforcement learning to analyse user behaviour whilst they're actually listening. So this in turn generates predictions on what listeners might like in the future. If the algorithm gets a suggestion wrong, then the user will inform Spotify, unknowingly of course, of this crucial piece of data, which in turn, will help to refine their future recommendations. So the profile of the listener builds over time. It's all crowdsourced data, so the more someone listens, the more they tell Spotify about what they're doing, and the recommendations become increasingly attuned to their taste. So Spotify uses ML to assess a listener's preferences across a variety of factors, such as genre, sub-genre, mood, tempo, instrumentation and a whole other heap of factors that we'll probably never truly understand.
Speaker 1:Ok, finally, element six is user feedback. I haven't mentioned this yet. Right, it's crucial, isn't it? So user feedback is a very, very important part of what Spotify do, especially as it aims whether and how quickly a song is actually sort of skipped or saved or liked etc. Saved to playlists and you know even sort of whole album likes and things like that. So social media shares also kind of create further interactive data, and this all feeds into ML and informs future recommendations. So there's a link point there between sort of element five and element six. So ML and user feedback, okay, so there's a little summary Now.
Speaker 1:What I've tried to do there is to take out all of the kind of science, data, heavy words or hopefully, anyway. There are a lot more technical terms and I'll be honest with you guys, I am not a computer scientist, right. I am a music producer. That's why I do this. So I find it easier to be able to kind of take away some of the jargon and actually understand some of the functions and operations behind something, because it helps us to understand if we can understand what they're doing. It helps us to understand what we are doing, what they're doing. It helps us to understand what we are doing, right. So that's why I've kind of tried to simplify that as much as I possibly can.
Speaker 1:But also, what we've just done there is we've looked at Spotify from what I what I call a back end user, right, so somebody that is an artist or someone that is feeding music to Spotify right, so that's back end use, right, so the kind of access to Spotify for artists, etc. Now, front end use would be a consumer, right. So it's very important to think about how we actually use Spotify and other streaming platforms as a user ourselves. So it's worth considering. You know, just stop and think for a minute and just think about how you consume music, right, and what effect that could be having on your generated recommendations. So, to understand how growth is kind of gained through platforms like Spotify, it's important to understand, you know, how they work and that's what we've been trying to look at there.
Speaker 1:But a good place to start with all of this is to add context to it, is to just start by thinking about how you consume music, how you use those streaming platforms, not as a creator, but as a consumer, as a front end user. So if you, for example, share your streaming account with a family member or a friend, the recommendations are going to be generated, are going to be, you know, quite significantly distorted and a bit confusing, right? So I mean, I'm kind of stating the obvious there, but it's worth bearing in mind, of course. So different users using the same account will compromise the whole experience if their tastes vary. By the way, if you're sharing your account with someone that shares your tastes, it could probably work even better than it already does for a solo user, but anyway.
Speaker 1:So it's also worth considering how you actually find new music and how music is being introduced to you. It will inform you as to how to feed it as a creator. You are certainly feeding it as a consumer, so it's good to think about it. Will inform you as to how to feed it as a creator. You are certainly feeding it as a consumer, so it's good to think about it the other way around. So, as a music creator, it's important, of course, to listen to music, right? Being a fan of music is almost certainly what inspired you to make music in the first place.
Speaker 1:But perhaps, you know, maybe you're using algorithmic playlists to listen to new music, you know, like discover weekly, for example, release radar, or perhaps you are consuming kind of home screen mood based playlists that are edited by the editorial teams and curated by the editorial teams at spotify, I should say, um, if so, you are definitely helping spotify to source on you, which in turn, of course, as we now know, feeds into an enhanced experience for you and others with those similar tastes from collaborative filtering. So perhaps instead, you're not consuming music like that. Let's imagine that you are saving songs to your own playlists. Well, if you're doing that, the outcome is equally as desirable for Spotify, because you are helping. You are helping them to understand you and your habits on an even greater scale, especially when, even though we're all unique, we have patterns right. So when those patterns are then compared to others, it tells them even more about what might be suitable for you. So, um, perhaps also, by the way, those playlists that you are sort of self-curating can be enjoyed right by others with similar tastes, in which case perhaps you know there is an influential tastemaker inside of you who is ready to curate for others. Maybe you even want to make it collaborative and you can collaborate on playlists with other people. You know, this can be a very, very useful marketing tool, by the way, for artists and labels, because it allows you know you to build playlists that have followers which are then readily available to you when you have new music to release. There's a lot of people doing that now. So you know, just pause for a minute, right, and just think about your own behavior in the context of streaming platforms as a consumer. It can be really really useful, because if you link that to how you then use it on the back end as a creator or as a provider of new music, draw the link between the two, it will really help you.
Speaker 1:Okay, and the final part of today's podcast is to look at the three different types of playlists and just understand them a little bit better, and we're going to stay on the theme of Spotify for this, because I think they do it really really well. So there are three types of playlists, right Editorial playlists, user curated playlists and algorithmic playlists. Let's have a look at each of them of them. Okay, so editorial playlists are created and managed by Spotify and are predominantly based on moods, trends, genres, cultural moments, activities, regions, countries, eras and a whole lot of other things. They are often seen as the holy grail of playlist success by artists and labels, and it's easy to understand why, because some not all, but some have a tremendous influence and can play a vital role in boosting a song's awareness. So, just for anybody that doesn't know, since 2018, anyone with a verified Spotify profile so that's with the blue tick verification, which doesn't cost anything can pitch their soon to be released song to the editorial team at Spotify, provided, by the way, that this happens at least seven days prior in advance of seven days in advance of a release date. Right? So this can be done through the editorial pitch tool in the Spotify for artists app or, you know, on the desktop format of that app.
Speaker 1:So some genre specific playlists can have a pretty undeniable impact on a song's visibility. So, for instance, something like Rap Caviar has I don't even know how many million followers over 15, while Rock Classics has in the region of around 12 million, right? So the more universally appealing Today's Top Hits, by contrast, has closer to something like 35 to 40 million followers. So other sort of editorial playlists focus on things like geographical location and its associated music style. So examples include predominantly sort of era based Made in in manchester, which has around 300 000 followers, and I love nyc, I love new york city, which has closer to kind of 120 000 followers, something like that, although this focuses on songs that are themed on this city rather than music from the city or inspired by the city. So, um, spotify has numerous editorial teams based in various global locations who are widely acknowledged as being more approachable than those of some of the other streaming platforms.
Speaker 1:Let's say, editorial playlists can give valuable data boosts to artists, righting benefits beyond just you know added exposure and the kudos gained from that placement. So inclusion on a large playlist not only guarantees a song a higher stream count, but also boosts the artist's profile by adding a range of new collected data, which in turn feeds further visibility through algorithmic growth. This is especially true, by the way, if the save rate reaches a particular threshold. So while Spotify knows that many of the streams on an editorial playlist may well be deemed as sort of passive listens, it still deepens the crowdsourced data approach to a song suitability for additional audiences. So don't be disheartened If you get on a massive editorial playlist and you go, wow, look at all these streams, look at all these saves, and then, boom, it's gone. Don't worry, that's what happens. That's how it works. It's okay because it has also fed an awful lot of data about your audience, or about your potential audience in regards to those that streamed it on those playlists and how they interacted with it. So that's all well and good, you know, if you've got editorial support or if you're, you know, signed to a major label, that helps that happen. But what about if you're not right? So there are other types of playlists. There are two categories, right.
Speaker 1:The next one to think about are algorithmic playlists. Now, these are absolutely crucial to the Spotify engine. They are unique to every listener and introduce new music to listeners every single day in some form or another. So Discover Weekly is served up to listeners every Monday and comes with a sort of you know 25 song sort of digital mixtape, if you will, and comes with a sort of you know 25 song sort of digital mixtape, if you will. So these are songs which Spotify knows that their listeners have not heard before. Well, maybe they have, but they haven't listened to them on Spotify before.
Speaker 1:Let's say, release Radar is refreshed every Friday and predominantly features artists that listeners follow, along with some additional options which the algorithm deems to be of interest to the listener. If that annoys the listener, then that means that it doesn't know the listener well enough. So if you get something like that and you go I don't like that, I'm going to skip it. Then that's useful to Spotify as well. Right, they know that they don't always get everything right. It's not about getting it right or wrong on paper. It's about getting it right or wrong for you and customise it to you right, wrong on paper. It's about getting it right or wrong for you and customize it to you right.
Speaker 1:So other forms of algorithmic playlists are compilations based on what listeners have already enjoyed. So we know things like on repeat and songs of the year, summer rewind, time capsule. These things exist, right, and that's why. So it's worth noting that even those types of playlists, of course, introduce new songs to the listener. For example, summer Rewind might curate an experience that allows the listener to reminisce about a specific season, but songs that the listener did not hear during that season will still be placed in to that unique playlist due to the power of collaborative filtering that we looked at earlier. So in many cases, the new songs blend so seamlessly with the rest of the playlist that you know we might not even necessarily notice that new music is being piped into our lives Now.
Speaker 1:The other thing is, of course, the platform recognises that listeners have eclectic tastes and tries to cater to this through its daily mixes. So these are usually between I don't know four, nine different daily mixes that are available to each unique listener and they're numbered right. So Daily Mix 1, daily Mix 2, and so on and so forth, with a kind of a summary of the artists featured on the playlist underneath. So if you have quite a lot of those, that means that you listen to a lot of different music. So you know, they're not named by genre, but the style of music is clearly indicated by the acts that are in that description, right? So you know, I guess it blends, um, and also, of course, fresh music. So those are all very, very good gateways for new music to break through to new listeners.
Speaker 1:And it's very, very good, of course, to be looking at Spotify for Artists and looking at the source of your streams and then seeing if some of them come through daily mixes. If that's the case, that's why and that's how it's working. So, um, now you know, while many music creators and and labels etc, you know, kind of regard editorial playlists as the sort of the holy grail of spotify playlists, it is perhaps actually the algorithmic playlists that prove a slower but more accurate and effective route to organic growth. So, you know, a particular threshold of streams may need to be reached before a song can secure a spot in an algorithmic playlist. Rumours suggest that around 20,000 streams are needed for Discover Weekly, but saying that further research also reveals that songs with way, way, way fewer streams have also found their way onto that discovery tool. So you know, the yardsticks are not just about the metrics of streams. There are, of course, all the other facets that we looked at earlier in the algorithm that contribute towards all that.
Speaker 1:And finally, there are user curated playlists. Now, user created playlists can be created and managed by anyone right, and therefore can vary significantly in terms of influence. They include everything from you know, privately curated fan playlists through to you know artist playlists, right up to the playlists of influential third-party curators with a large audience. So those are the kind of curators that you would be able to pitch to in places like Groover and SubmitHub, for example. Also, many independent record companies have their own playlists, which would fall under that category, and even major labels, by the way, curate their own playlists, and they look like they're kind of you know, not in any way associated with a major label, but they are um. So you know also, there are now, um, you know, independent companies that manage their own playlist range and there are other playlists that started out maybe as like some kind of fan page but have grown into something more influential, um, you know, kind of like how youtube works, right? You, when you start off, even though their algorithm is kind of built more for content creation than for music, you know, youtube is still the giant. So, undeniably, spotify have looked at YouTube and gone. What have they done? Right? And that's one of the elements. Now, a placement on an influential tastemakers playlist can sometimes generate more streams and traction than a placement on an influential tastemaker's playlist can sometimes generate more streams and traction than a placement on an editorial playlist, which is kind of hard to imagine, but it's true. I can assure you. I've looked at the data, I've seen it, it's a thing. So don't underestimate the power of user curated playlists, because actually it's one of the key things that underpins grassroots growth and gets it climbing up the algorithmic ranks into editorial playlists.
Speaker 1:Ah and relax. Okay, we're done. So I hope you found that useful, everybody. I do try my absolute utmost to try and simplify things as much as possible. I think that's a good thing to do in music and in life in general, really, I suppose because it's good to not get too caught up in. I don't know how this works, therefore I can't do it, so hopefully an episode like this today enables you to be able to go. You know, I kind of get that now. Ok, right, I can bear that in mind. Oh, boom, yeah, it's worked. Brilliant, ok, oh, wow. So actually you listening to this episode today has been a crucial investment in your time, and I wish with all my heart good luck to you in everything you're doing today and moving forward. Ok, that's enough from me today. I wish you a great day, feel free to reach out anytime you like, and may the force be with you.
