How TF does Spotify work?

0:00 Why Spotify is a marketing tool, not an income stream

How the f**k does Spotify work? Spotify claims over 30% of the music streaming market share. That's a big deal, and what matters isn't just how much of the total market it takes, it's also where a lot of music discovery happens. It's got the best tools for finding new listeners.

Forget about pay per stream for now and focus on reach. I really don't want you thinking too much about how much you're going to make off of promoting your music with conversion campaigns. The answer is probably you're going to lose money, unless you've got a really sound strategy to monetize elsewhere, because streaming just isn't that good of an income stream, especially if you're having to put money into promoting it. So think about this more as a business: you're going to make your money in other places, and Spotify is a marketing tool. In that way, you can reach a ton of people with your music.

1:07 The three parts of Spotify's recommendation system

This is how Spotify's recommendation system works. It has three components: collaborative filtering, content-based filtering, and natural language processing. Collaborative filtering works like this: you've got John, who listens to artists A, B, and C, and Mary, who listens to A, B, and D. They both listen to A and B, so Spotify's like, maybe we should try showing D to John and C to Mary, because they've got similar music tastes but haven't heard this other artist. It's kind of like when you're on Amazon and it says people who purchased X also purchased Y. The idea is that it's looking for people with similar musical taste as you, then it's looking for the pieces that are different and recommending those to you.

So if you think about this strategically, it's about how you can become part of these common collections of artists, like scenes or niches. Niche is a great way to think about it: you want to place yourself into a niche so you'll be recommended to other people who are into that niche but haven't heard you yet. In the end, both listeners end up with A, B, C, and D in their listenership.

2:25 Content-based filtering and natural language processing

Content-based filtering works by Spotify running the music through something called a spectrogram, which collects data about the actual sound of the music. Often this data is wrong, but that's what it is. It collects things like tempo, key, loudness, time signature, BPM, energy, danceability, all of these different elements, and it makes recommendations based on that. If you're listening to music that fits a certain energetic and vibe level, it wants to weight things that fit in that same space. So it calculates its suggestions based on a balance between collaborative filtering, content-based filtering, and natural language processing.

A lot of people don't know this: natural language processing means Spotify scrapes the web, meaning it actually reads articles, blog posts, any information it can find about you and your music, and correlates that to your vibe to create recommendations. This happens on an artist level and a song level. PR isn't as potent as it used to be, but it's still really important. In fact, I see this the most, and this is a theory, I don't have proof around this, but I see this the most in the 'Fans Also Like' section, where oftentimes it's not just about listenership, it's about artists you've toured with. Even if you're no longer playing with them, that's just out in the world, and it's really hard to overcome that, even if you're pushing listenership elsewhere. So natural language processing is a really important component, and Spotify weights the three of these to make recommendations. We don't know the exact weights, that's obviously proprietary and we don't have access to it, but these are the three main pieces to their recommendation system.

4:28 What you can actually influence: collaborative filtering

The part of the system we can impact directly with conversion campaigns is collaborative filtering. You don't really want to try to manipulate content-based filtering by strategically adjusting the speechiness of your song, just forget it, focus on collaborative filtering. You can do natural language processing by getting more blog posts, et cetera, but it's a costly and time-consuming thing to do, and honestly not the best use of your time and energy in my opinion. You're much better off focusing on conversion campaigns to influence collaborative filtering. A great way to do this is actually creating and promoting playlists with other artists you want to associate with. So how do we do that? We feed the robots.

5:15 Feeding the robots with conversion campaigns

Conversion campaigns filter for active listeners who are taking intentional actions to stream your music after already being introduced to your vibe and your sound. Remember this: they've already seen your ad, they've already seen a video, they've heard the music, they like it, they click on it. It takes them to a landing page with a tracking pixel, and they have to make an effort, so now they're intentional, super intentional. They're not bots, and they're clicking through to listen to your music, so their likelihood of really being into it is very high. This filters out the people who aren't super intentional during the optimization process of a conversion campaign. It maximizes repeat streams, maximizes saves, maximizes playlist adds, maximizes shares, and it minimizes skips, which are actually a negative coefficient in your score internally to Spotify. It's going to deprioritize music that people skip. Makes sense.

6:22 A real campaign's save numbers

Here's what this looks like when you're running a conversion campaign. This is a specific example from when I was spending $30 a day marketing my own music and pushing an album. People were saving the whole album, and since there are 10 songs on the album, every time somebody saved the album that's 10 saves. So 600 a day means 60 people are saving the album, and I'm getting 673 saves a day, which is very good. It means people are going to go back and listen to that music. It's going to boost your streams, boost your listeners, boost playlist adds, boost followers. It's really good s**t. Conversion campaigns are super effective.

7:09 The popularity score explained

Why is this so good at triggering the algorithm, which is a little different than the recommendation system? It's because Spotify uses something called the popularity index, also called the popularity score, internally, which is what they use to determine what to push out into their different algorithmic playlists, which are now out-competing editorial playlists. The popularity index is a measure of a song's relative performance over the last 30 days, mostly influenced by streams and then listeners, and it influences algorithmic placements.

7:43 The threshold ladder: radio, Release Radar, Discover Weekly

Around 10% plus means you're going to start getting pushed out into radio. All of this is approximate and always changing, but in general that's where you'll start seeing a little bit of radio action; the higher it gets, the better you'll do on radio. Around 20% in the first month means you'll get a boost to your Release Radar, which pushes your music to non-followers. If it's less than 20%, Release Radar just shows your music to your followers, and barely even does that. Release Radar sucks unless you're promoting well, so you're not going to get the most out of it because it's only during the first month. There's a four-week period where you can only have one song at a time on Release Radar, and that's why it's not generally advised to release more than one song every month, because only one of them can be on Release Radar, so you're missing out on that algorithmic boost opportunity. You might as well save this in order to get the most out of it for every song. Around 30% is where you're going to hit Discover Weekly, and this is where you're really in the algorithm, and it's going to determine if you continue to get pushed out or not. As those scores go higher, it increases your visibility for editorials, so on and so forth.