Skip to content

Streaming's Endgame (Part Two)

In part two of Streaming's Endgame, I offer a historical perspective on the evolution of music's digital era, from the iTunes Music Store to Spotify's global domination.

In part one of this three part series, I examined the increasingly divergent fortunes of artists and corporations in the modern recorded music business. As 2021 comes to a close, the industry's corporate titans enjoy renewed optimism, impressive revenue growth, and record-setting public market valuations. Conversely, individual musicians – those actually making the product being sold – find it increasingly difficult to create or sustain a living in the streaming economy.  

While a multitude of factors are driving this bifurcation in outcomes from the streaming era, several play an outsize role. Supply has exploded, while demand has remained relatively fixed; catalog continues to grow into an ever-larger share of listening time; and pro-rata payouts, power-law distributions, and a fundamentally flawed equity structure between creators and platforms have left an elite few looking down on an ever-larger pool of artists fighting over scraps.

The past 20 years, of course, haven't only changed music's supply side. The industry's demand side – the experience of finding, consuming, and listening to music – has been shaken at its very core just the same. Since the late 1990s, piracy, the iTunes Music Store, and on-demand streaming have progressively altered the experience and economics of consumption with increasing consequence for those who view music as a profession.

In the final two pieces in this series, we'll shift our focus from the broad technological trends which have driven music's transformation to an in-depth look at the history and core attributes of streaming platforms. In the process, we'll tease out the foundational principles and forces underlying our increasingly digital world: aggregators and the power they exert, the rise of algorithmic curation, winner-take-all dynamics, and the commoditization of content with increasing scale. Lastly, we'll shift from analysis to prediction and prescription, considering the next generation of disruptive trends on the horizon and how they could – and should – change our industry for the better, while being cognizant of the challenges ahead.

I believe we are nearing the end of the streaming era's first chapter, which can broadly be defined as the period from Spotify's U.S. launch in 2011 to current day. Where we go from here – the Streaming 2.0 era – will be defined by the choices made by platform creators, industry leaders, investors, and music fans themselves. I will outline two possible visions for this future and what it might look like; by the end of this series, I hope to give you the tools to consider how we can collectively build a better industry for fans, artists, and rights-holders alike.

Before we can conceptualize where we're heading, however, we must first understand where we've been. There's no better place to start than the infinite loop of innovation in Cupertino, California – the home of Apple Computer at the turn of the 21st century.

One Infinite Loop

After the initial shockwaves of piracy rattled music's corporate giants in the early 2000s, the man who would eventually be hailed by many as America's greatest innovator, Steve Jobs, believed he had a solution.

In November 2001, Apple launched the original iPod music player – a now iconic piece of hardware, and the first among a series of products that would turn a nearly bankrupt brand into the world's most valuable company over the next two decades.  Jobs doubled down less than a year later, releasing the iPod second generation with 4x the storage capacity of its predecessor: a whopping 20 gigabytes, enough to store 4,000 five megabyte MP3 files.

While 4,000 songs may sound like a pittance in 2021, it was a staggering amount of portable storage in 2002. Few were more shocked than the major record labels. From their perspective, the iPod was undoubtedly a remarkable consumer product – but where, exactly, were listeners getting all that music from? With the lackluster adoption of label-sanctioned platforms such as PressPlay and MusicNet, the industry lacked digital distribution channels at scale. Most consumers were either curiously inclined to spend money on storage they didn't need, or were amassing a large amount of music from illicit channels. Excluding a small cohort of fervent CD collectors who enjoyed ripping their collections to MP3, of course, the answer was the latter.

By 2003, questions surrounding the iPod's enablement of piracy became hard to ignore. Eager to make amends with the major record labels, Jobs had a second invention up his sleeve: a digital music store that consumers would actually enjoy using. That April, the iTunes Music Store launched with a fully-licensed catalog of over 200,000 major label songs available for purchase for just 99c each – a truly massive shift in how the industry's copyright owners envisioned their business.

The iTunes Store wasn't merely a new, technologically advanced sales channel for delivering goods to consumers. For the major labels, it represented a humbling inversion of the core power dynamic of their operations – albeit one they had no choice but to endorse. Just three years in, this century was already shaping up to be drastically different than the last for the industry's leading corporations.

The 20th century music business was almost comically absurd in its anti-consumer practices. Price fixing, racketeering, payola, and various other activities befitting companies with near zero fear of competition or disruption were widespread. Following UMG's merger with Polygram in the late 90s, the cost of producing a single compact disc (including all packaging and materials) dropped to less than $1 – which led the conglomerate to pass savings of exactly zero percent on to their loyal customers.

To many observers, the music business of the mid-1990s seemed unstoppable. The major label hegemony remained firmly in control of highly profitable intellectual property, while wielding complete dominance of its exclusive distribution channel. In reality, however, it's always most still right before the storm. In the span of a few short years, the narrative flipped from record profits to staggering losses, and from $16.99 CDs to 99 cent downloads. The majors hadn't merely lost their grip on a retail channel: they'd lost an understanding of their place in a rapidly digitizing world.

In part one, we noted that the RIAA's panel of engineers made a grave error in judgement when debating the consumer appeal of digital music formats – but why did the majors, with all of their resources and knowledge, make the same mistake? While hubris, complacency, and the innovator's dilemma all played a role, I believe the fundamental error was an application of 20th century principles to 21st century realities. Since the advent of the modern music business, labels had been in complete control, determining who put music out, and where, when, and how they did so. In their eyes, obviating the MP3 as a threat was as simple as not backing its developers in a business meeting; what could possibly happen without their blessing?

Historically, labels controlled format development and distribution down to the minute details. The internet, however, had other plans: for the first time in history, users were in charge of content distribution and consumption, not boardrooms. This profound shift, nearly identical to the one seen in a range of industries later upended by technology, would drive the rapid adoption of radically new ways of paying for, and listening to, music.

Infinite Shelf Space

While Spotify receives much of the credit – and blame – for the seismic changes the digital era has brought to bear on music, the iTunes Music Store was the first to introduce several key dynamics that would reshape the consumption and monetization of music at a core level. Nearly 20 years after its launch, these mechanisms continue to have an indelible impact on the streaming economy and artists around the world.

Chief among these is the notion of 'infinite shelf space', which former Wired editor Chris Anderson referenced in his massively influential theory of The Long Tail published in 2004. Anderson argued that the era's nascent digital platforms would transform the careers of creators the world over: retail's transition from physical stores (with fixed shelf space) to digital storefronts (with infinite shelf space) meant that any creator of any thing would soon be only one click away from their target consumer.

The iTunes Music Store was, at its launch, the most realized example of Anderson's infinite vision. Gone were the limited shelves of Tower Records and HMV; the endless expanse of digital storage meant the service could have every record, by every creator, ever created, so long as it could be digitized one way or another. While Anderson's observation was technically accurate – many artists were now just one quick search away – the past two decades have turned out quite differently than he anticipated. Far from enabling one-click access to the long tail, music platforms have instead driven the masses towards an increasingly powerful hit and scale-driven market owned by a tiny group of elite creators. This transformation, crucially, began on iTunes.

How exactly did Anderson's vision of a creative utopia go awry? As it turns out, the infinite scale of digital platforms does indeed create a seemingly endless pool of instantly accessible content – but when a store holds every piece of media in the world, how on earth do you actually find any to consume?

"All curation grows until it requires search, and all search grows until it requires curation." - Ben Evans

The history of the internet is replete with startups who endeavor to catalog some rapidly growing source of content for easier navigability. Invariably, however, they fall into a circular logic trap that goes something like the following:

  1. 'Let's take all this content and put it in a structured list, so it's easier for people to find!'
  2. The list quickly grows from a manageable number to a deluge.
  3. 'Uh oh, this list is too long for a human to navigate. Wait – we can let people search it!'
  4. Search works fine, assuming your algorithm is half decent, and if a user knows what they want. But how do you solve the cold-start problem – a user who needs some guidance?
  5. 'I've got a solution – we'll curate a list of what we think is best!'

This final step – which nearly every large platform in history has conquered or been crushed by – is where the long tail theory and infinite shelf space begin to go off the rails. While the wide-eyed optimists of the early Web 2.0 period envisioned an always open, instantly navigable, creator driven storefront that would transform the arts, the iTunes Store turned out to look an awful lot like the Tower Records it was bound to replace. There's a perfectly good reason HMV used to put the week's hottest new releases front-and-center, in their most valuable square footage: people need help finding content to consume!

Beginning with the iTunes Store, this foundational challenge – and the problems our industry has devised to solve it – would catalyze a chain of events that would, in time, shift the onus of discovery from the user to the platform, unbundle and rebundle the album, and fundamentally commoditize the song as a format, turning the artists responsible for those songs into interchangeable commodities themselves.

Unbundling and Rebundling the Album

As early as 2005, scholars began to notice a curious phenomenon on digital music platforms: they seemed to be obliterating the primary means of musical consumption for decades, the album. This process, which began in earnest with the advent of Napster and peer-to-peer file sharing, represented a profound shift in how music was packaged. At their core, albums offered consumers a binary choice in music consumption: never mind if you only like one or two songs, you can either buy an entire compact disc for $16.99, or don't listen at all. Napster, of course, was only too happy to remove this long-ingrained restriction, allowing users to download full albums, singles, and anything in between. And so began the unbundling of the album into their constituent parts: songs.

"There are only two ways to make money in business: One is to bundle; the other is unbundle." - Jim Barksdale

As Jeremy Morris notes in his superb 2010 synthesis on the commoditization of music via digital platforms, the iTunes Music Store took this principle one step further, rebundling individual songs from many different albums into new, neat little playlists with attractive cover art. A consumer in 1995 went to Tower Records to buy a full album from their favorite artist; a decade later, the same listener was going to iTunes for platform-curated playlists amalgamated from scores of different musicians.

As with many transformational changes, this unbundling and rebundling of the album into playlists didn't seem terribly important at the time; it just seemed like a quick way to find a solid mix of New Years or Halloween-themed tracks. The core principle, however – the notion that songs could be untethered from their creator's intended context and reassembled at will – would in time create a deep shift in the locus of curation from the artist to the distribution channel. Crucially, it would do so on a scale exponentially greater than that available to those hawking homemade cassette or CD mixtapes to their friends in the 1980s and 90s.

While the iTunes Store initially vowed to be music's piracy panacea, the era of windfall profits Jobs promised Doug Morris and other industry leaders failed to materialize quickly enough. One year after its launch, in 2004, Apple highlighted the 100M song purchases the store had driven. While this sounds like an impressive topline number, at only 99c per song, it represented just $100M in gross sales – one percent of Universal Music Group's total annual revenue.

This lackluster revenue was compounded by the fact that iTunes then made up 70%+ of the legal digital music market: it wasn't a small retailer in a booming sector, but rather the dominant player in a highly nascent space. As a result, the industry continued its descent towards the nadir of revenue it finally reached in 2014. It would take a subsequent technological innovation – on demand streaming – to save industry giants from eventual oblivion. Many artists, however, would not be as fortunate.

There's Something in the Water

A common refrain in pop music production – a field I've spent the past 15 years trying to master – is that there's something special about the Swedes. The country, whose relatively small population (10M) belies a truly massive impact on the Billboard charts, has produced a slew of writers and producers with otherworldly talent, none more-so than Max Martin, the most successful songwriter of all time. From Britney Spears' ...Baby One More Time in 1998 to The Weeknd's I Can't Feel My Face and Blinding Lights, odds are the catchy melody that became stuck in your head sometime in the past 25 years was Martin's handiwork.

In the past 15 years, however, the Swedes haven't just made their impact on the charts: Spotify, one of the country's largest consumer-facing corporate exports, has shaken the very core of the music business itself as the unquestioned leader in on-demand streaming. Understanding Spotify's rise – and current impact on artists – first requires a brief examination of the unique period in which it began.

As Maria Eriksson notes in Spotify Teardown: Inside the Black Box of Streaming Music, the company was founded during a time of particularly rapid technological progress. Web 2.0 – the internet we know today, with more interactivity, rich media, and user generated content than Web 1.0 – was rapidly eating the online world, driven by increased broadband adoption and sophisticated technology for platforms to build upon.

In a piece of oft-overlooked history – one critical to better understanding Spotify's future ambitions – the platform's 2006 origin story did not necessarily revolve around music. Rather, as Eriksson and her co-authors note, Spotify merely began as a media distribution platform, without any particular kind of media in mind:

"The kind of data to be distributed was, for Spotify, a secondary consideration at this point. “The media may here represent any kind of digital content, such as music, video, digital films or images,” explained the US patent application that Spotify filed in 2007...Speaking with the business press, the founders presented Spotify as a general “media distribution platform,” indicating that the ultimate aim was to use it for video distribution."

There's just one problem with video, however: files are huge, requiring immense amounts of bandwidth and storage compared to audio. While broadband was indeed growing in the world's leading economies, including the United States, our digital infrastructure simply wasn't capable of supporting the file sizes necessary for a truly instant 'everything store' for video just yet.

Source: Pew Research Center

We can understand Spotify's initial focus on music, then, as more of an intuitive read of what the consumer market could bear, rather than an ideological or passionate act of platform determinism. Understanding Spotify's early days as a format-agnostic startup will prove key to making sense of its eventual pivot beyond music, which we'll discuss later, as well as its apparent callous indifference to the plight of recording artists in the next decade. To the company, audio wasn't necessarily the media format most deserving of legal distribution at scale. Rather, the files were small, the internet was still developing, and music just happened to be there for the taking amidst the fallout from widespread piracy.

Once Spotify settled on a format to disrupt, the company went about building what would become the first truly compelling user experience in music's digital era. Two features were critical in this pursuit: sub-300 millisecond latency for playback, and the ability to search for and play any song under the sun. Spotify's aggressive push to minimize user latency – the time between pressing the play button and actually hearing music – was and still is a groundbreaking technological achievement, and one essential for user adoption. As CEO Daniel Ek has often noted in interviews since, the founding team felt it was essential  for the service to give the user what they wanted, and to do so immediately.

The company's second insight – that any song ever recorded should be at your fingertips – presented a marked contrast to the monetization model of the iTunes Music Store. Although the iTunes Store similarly sought to be a catalog of the world's music, the two services had fundamentally different visions about how to give users access to this library. While iTunes espoused the pay-per-download model, in which a user must purchase each song or album they want to listen to, Spotify instead opted for an unlimited, 'all you can eat' streaming model, supported initially by advertising and later by monthly subscriptions. Following Facebook's "move fast and break things" mantra, the company initially chose to fill their library of on-demand songs not by licensing them from rights-holders, but rather by mostly using pirated song files from its employees computers.

The transition from a-la-carte purchases to limitless streaming would prove to have enormous effects on the economics of the music business within a decade. While iTunes' usage-based framework creates incremental revenue for rights-holders as a user consumes a broader swath of content, Spotify's unlimited model offers a fixed payout regardless of consumption patterns for paid subscribers. Although ad-supported users do in theory offer incremental usage-based revenue as consumption increases – more time in app equals more opportunities to show ads – such listeners typically contribute far less to a platform's bottom line due to the economics of digital advertising. Assuming a $15 CPM (cost per thousand impressions), an ad-supported user would need to see 666 video ads per month – 22 per day – to contribute as much revenue as a $10 per month subscriber.

There's just one problem with arguing against Spotify's all you can eat model: it happens to be a phenomenal user experience. It's no coincidence that the streaming behemoths of our current age – Spotify and Netflix – both ask for your credit card precisely once before unlocking unlimited access to a sprawling catalog of music and video content, respectively. In the digital era, friction is the enemy of user experience, and Spotify's frictionless consumption model would, in time, make singular song purchases a relic of the past for all but the most determined consumers of music.

Notably, Spotify's approach to unlimited consumption was and still is an outlier in the streaming economy. Unlike rivals in verticals such as video and audiobooks, the platform's flat monthly fee provides access to virtually every single piece of content ever produced in its chosen media format. Netflix doesn't offer every video ever made; Audible Plus doesn't offer every audiobook; nor does Xbox Game Pass offer every video game. Music is singular in its truly 'all you can eat' nature, which undoubtedly contributes to the financial struggles of its content creators.

Over the next decade, Spotify wouldn't just break the industry's long-running reliance on incremental, consumption-based revenue models. It would also fundamentally alter the means by which music is discovered and consumed by listeners, through the growth of algorithmic curation and the ascendance of the playlist, on its way to becoming the über aggregator for music.

User vs Platform-Centric Design

As Spotify set out to save the music business, the company initially took an approach that would have made Chris Anderson proud: it sought to be a searchable index of every song ever made, with the user firmly in control of the listening experience.

"[In 2008-2012], user interaction was organized around tracks, search options, and community-activating features, such as self-made playlists...During its beta period, Spotify consolidated a kind of on-demand doctrine as a service centered on the search box, giving access to “whatever you want.” The user was effectively conceived of as a sovereign individual, who already knew exactly what he or she wanted to listen to and did not need help with music recommendations." - Spotify Teardown

We can understand Spotify's initial design, then, as the most realized vision of Anderson's infinite shelf space to date as the decade drew to a close. In this paradigm, the platform exists as a transparent distribution layer between the sovereign user (listener) and the music they seek to find. The user is in full control; the streaming app is merely there to facilitate the fastest and smoothest playback experience possible.

This approach quickly led to impressive growth for the company, with rapid international expansion and praise from music and tech press alike. While competitors focused on the size of their catalog and curatorial services that would help listeners manage the vast expanse of music now available, Spotify prioritized the user experience – with a particular focus on speed – and assumed that the listener already knew what they wanted to hear. Nowhere was this contrast more acute than with its most formidable rival, Pandora. In contrast to Spotify's search-based approach, Pandora enjoyed rapid growth by offering users a 'lean-back' streaming experience, with a steady stream of radio-like playback based off of an initial artist selection by the listener.

In 2012, however, this battle of dueling product visions was challenged by a new entrant that would, in time, help to dramatically change the face of music streaming: Songza. The upstart service – which would never prove to have an enduring business model of its own – sat somewhere between the polar extremes of Spotify and Pandora, asking the user to begin a listening session by selecting a mood or activity in contrast to a specific artist or song. By mid-2012, Songza was the top free iPad app, while ranking #2 overall on Apple's iPhone free app charts, signaling significant traction for the company's unique 'vibe' based approach to curation. Notably, the options presented to the user for selection were heavily influenced by data, including a user's listening history, time of day, and day of week – representing one of music's first data-driven music discovery platforms at scale.

Despite Songza's limited long-term prospects as a standalone service – it was acquired by Google for an estimated $15M in 2014 – the platform would prove to have a seismic long-term impact on our modern music landscape. Seemingly overnight, the same press outlets that had praised Spotify's sovereign-user design were criticizing the product for its lack of listener guidance and merely being an enormous repository of digital recordings. Over the coming years, the pressures exerted on Spotify by Songza and similar challengers – and the platform's response – would catalyze a profound shift in the economics and consumption of recorded music (emphasis mine):

"[Spotify] gradually reoriented itself toward the aim of providing not only access to music but also recommendations for music that users would not have requested themselves. This meant that Spotify began to transform itself from being a simple distributor of music to the producer of a unique service." - Spotify Teardown

As Eriksson and her co-authors note, this shift from user-centric to platform-centric design represented much more than an evolution of the product: it was a fundamental reimagining of the role a platform should play in the consumption of music content. User-centric design places the locus of control on the 'knowledgeable' listener, prompting the user for some degree of conscious consideration of what they want to listen to. By contrast, platform-centric design shifts this responsibility to the platform itself, imagining the listener "as being in urgent need of musical advice and guidance from experts."

This paradigm shift also turns out to have another key benefit for the platform: user lock-in. Although major labels initially experimented with exclusive releases in the early days of the U.S. streaming boom, this proved to be short-lived. After a decade of attempting to persuade fans that music does indeed have some intrinsic monetary value, telling a paying subscriber that they can't listen to a major new release turns out to be a very bad sales pitch.

Today, every major streaming platform has a virtually identical catalog of songs: the only differentiating factor is the platform itself, which is comprised of its design, user experience, and algorithmic or editorial curation. As we'll see shortly, it is no secret that Spotify's algorithm gets better at recommending content the more you use the service. In practice, this creates a powerful feedback loop for user retention, as changing platforms incurs a heavy switching cost on the user. As much as one may dislike particular aspects of how the company deals with artists, it's awfully hard to wean yourself off of that phenomenal Discover Weekly playlist that lands in your inbox every Monday.

This shift wouldn't merely upend much of the modern music business: over the coming years, this same concept would come to undergird many of the world's most widely-used social media platforms, including YouTube, Tik Tok, Instagram, and Twitter. The early Web 2.0 period was a time of the sovereign individual; the latter would prove to be a time of the sovereign algorithm.

Collaboration for All

Spotify, for its part, would prove to be uniquely adept at both catalyzing and navigating this transformation at scale. Although the means to this end took several years to iron out – the service initially pivoted to 'trusted source' curation from experts, trendsetters, and artists – by 2015, the company had found its footing in the land of algorithms with the launch of Discover Weekly, a hyper-personalized weekly playlist delivered to users each Monday.

As Spotify Teardown notes, Discover Weekly was a critical turning point in freeing users from being 'trapped behind a search box' – the playlist would gain such rapid momentum (5B+ streams in its first year alone) that the company rolled out two additional algorithmic playlists, Release Radar and Fresh Finds, shortly thereafter.

Shifting the locus of discovery from the user to the platform would prove to be a direct attack on Anderson's vision of infinite shelf space for the long tail of creators, and it would also become the primary catalyst of the increasingly extreme power-law distributions I outlined in part one of this series. While in theory the artist you want to listen to is still just one click away, what about all this shiny new music you haven't heard yet? Don't you want to hear just a bit of that first?

Understanding the progression of this change, however, only tells us one part of the story. How exactly does a platform like Spotify actually create the recommendations it gives to listeners? And just as importantly, what do the nature of recommendation algorithms mean for the average musician?

Spotify's recommendation algorithms – like those of many leading streaming platforms, regardless of media format – are largely powered by a technique known as Collaborative Filtering, which attempts to create an abstraction of user preferences through the use of massive datasets. Just as you might have gone to a friend with particularly good taste for a song recommendation in the pre-streaming era, collaborative filtering algorithms attempt to surface content to a listener based off profiles of users who appear to be similar.

In such a recommendation system, each user builds up a 'taste profile' over time as content is consumed. From your very first play on Spotify (or Netflix, another market-leader who uses the technique for recommendation algorithms), most actions you take – such as a song play, skip, share, artist follow, or playlist addition – begin to feed into your personal taste profile, which also gives particular weight to the genres you listen to.

Once a baseline of data is attained on a user, the recommender system can then query its database of other users who look similar in taste; the final step occurs when the algorithm determines what songs or artists these other users – your 'nearest neighbors' – have engaged with, but which you have not. And just like that, a highly curated and personalized playlist appears in your app every week, like magic.

How accurate are these systems? That depends largely on both the size of the dataset a platform has to work with and the skill of its engineers – and Spotify, of course, excels in both categories. Netflix has previously disclosed that collaborative filtering techniques result in better recommendations than explicit user feedback about likes and dislikes: as is often the case in our modern digital world, data can tell a very different story than we tell to strangers.

Source: How Spotify Recommends Your New Favorite Artist by Clark Boyd

At first glance, you may well be wondering how collaborative filtering and algorithmic recommendation engines could possibly lead to the immense engagement gap between popular and long-tail artists we see in streaming today: shouldn't such systems be leading us into a golden era of artist discovery and empowerment? In theory, they should. In practice, however, I believe the opposite to be the case. Recommender systems, when combined with the exploding supply of artists needing exposure, instead commoditize the creator through the extraction of objective data points from the art form they analyze. A deeper look at Spotify's algorithm is instructive in understanding why.

Crucially, algorithmic recommendation systems do not, at their most core level, recommend an artist as the unique creator they are. Algorithms do not (yet) understand a musician's life story, personality, or other such subjective and personal qualities which fans often care deeply about. Instead, they abstract a musician's songs into a batch of attributes, qualities, and objective data points, all of which can be ranked numerically from zero to one, such as their 'danceability', 'speechiness', and 'energy'. None of this is a secret: Spotify generously offers a robust Application Programming Interface (API) for developers, which allows one to query their massive database for songs, artists, and algorithmically generated recommendations based off these very qualities.

As an example, let's take a look at what Spotify's API says about my artist profile on the platform, with a specific focus on artists it thinks you're likely to enjoy if you like my music. Thanks to the good folks at Glitch.me, you can do this without any programming experience:

If you've heard my music before, something should jump out at you very quickly when scanning this list: most of these artists are exceptionally odd recommendations given my history as a house music producer. Of the initial list of ten artists closest to me, only two are in similar genres (Kap Slap and L'Etranger), and several are from a particular style of Hip-Hop that is highly dissimilar to my music. No human who enjoys my work would recommend most of these artists if asked to give a list of comparable musicians: while perhaps an outlier, such data shows some of the potential weaknesses of algorithmic recommendations.  

Things get more interesting, however, if we run a similar query on the top artist recommended to my listeners, mike., who happens to make some really good (and very popular) melodic Hip-Hop. Mike., most importantly, has 4.5 million monthly listeners – which puts him, according to Spotify Loud & Clear, in an elite group of the platform's top 2,000 creators out of over eight million. Mike.'s recommended artists happen to be a lot better than mine, with a solid list of very strongly related acts including G-Eazy and Bryce Vine – two creators who you are highly likely to enjoy if you enjoy mike.'s music.

Why would Spotify be so much better at recommending music for mike.'s fans than for my own? Well, we need only take a glance at the platform's API documentation for its 'Get Recommendations' function to get a hint:

In other words: data is king. For artists and songs with lower volumes of data for Spotify's algorithm to analyze, recommendations may be less useful; for artists with very low stream counts, the algorithm may cease to function completely. But this particular point is worth exploring further; while I am certainly no global superstar, especially as I release less music these days, I'm also not what you'd necessarily think of as a 'ultra long tail' creator. Several of my songs have enough streams to rank in the top 10% of Spotify's catalog; I've released music on every major label; and I've even made Spotify New Music Friday in the U.S. and Viral 50 in some foreign countries, which a small subset of the eight million artists on the platform can claim.

I've been in tech long enough to know that it's all too easy to build a weak argument on what is nothing more than an outlier in a system (an 'edge case'), so perhaps there isn't much worth reading into here. But Spotify's own documentation – which is transparent about the need for large quantities of data for accurate recommendations – is revealing when trying to understand how the rise of algorithms may potentially disadvantage the long tail of creators.

In a recurring theme in this series, it is worth noting that the practical outcome of Spotify's shift toward algorithmic curation likely has vastly different impacts on artists versus listeners. As an artist, I am increasingly concerned about the central role algorithms now play in curation on the industry-leading streaming app; as a listener and fan of good product design, I find hardly anything to complain about. Spotify's product, as I noted in part one of this series, is not to be faulted for its listener experience, which is brilliant. In many cases, however – particularly the abstraction of living and breathing artists to lines in a database where their songs are classified by 'speechiness' – the same features that make the end user experience so phenomenal also create the most troubling trends for creators.

Spotify's recommendation algorithm is more often than not superb from the listener perspective. I am consistently impressed with the quality of recommendations in my personalized playlists, a fact likely driven by my long-time use of the service as a fan: if data is king, the company certainly knows my taste by now.

In most cases, however, the songs I discover via Spotify's recommendations are quite popular; they're released by artists firmly in the upper echelon of creators we explored in part one of this series, and nearly all are from artists I've at least heard of before. This bias towards popularity also isn't a critique of Spotify alone: Netflix is a great streaming experience because it typically shows me well-established and widely viewed shows I'm likely to enjoy, and Amazon has an uncanny ability to recommend really good products for my particular search within the first few results (the company also uses collaborative filtering).

It is far from my place to tell anyone that they shouldn't enjoy the end-user benefits from algorithmic curation. My only request is that you start to ponder the ways in which these increasingly ubiquitous curatorial tools likely bias your consumption towards the elite few in place of the long tail.

Keep It Short, Stupid

Having written music in multiple genres ranging from house to film soundtracks, I've always found that pop music – a deceptively simple format – is by far the hardest to write in. While no genre of music offers an easy means to writing a good song, nearly all styles aside from pop offer the songwriter or producer far more leeway in the number of elements introduced to the listener.

Whereas house producers (myself included) are often able to rely on overwhelming the listener with a 'wall of sound' approach, which can utilize many sounds in a particularly dense section of a song, pop enforces strict constraints on the songwriter. If you listen closely to nearly all commercially successful pop records – both in and before the streaming era – you'll notice there are rarely more than two (or a maximum of three) sonic elements hitting your ears at any one time. With the ever-critical lead vocal occupying one of these, writers are often challenged to keep the listener engaged with just one or two instrumental parts: a kick drum and a guitar or bassline are most frequently employed.

This restriction has, over time, birthed a frequent refrain among pop songwriters: Keep It Simple, Stupid (KISS), which is often used interchangeably with 'don't bore us, get to the chorus'. Both phrases, however cliché, summarize the point succinctly: a successful pop record is fundamentally about stripping away excess and letting the vocal shine.

When considering the impact and eventual legacy of the rise of algorithmic curation, few trends are more important than the rapid decrease in song length and shift in song structures, both of which are a direct product of streaming platforms, particularly Spotify. This change is especially critical to note, as it is far from subjective or theoretical: the streaming era, and the concurrent rise of algorithms, is directly altering the sonics of music itself.

How drastic is the change? From 2014-2019, the average length of a song in the Billboard Hot 100 decreased by 20 seconds, or nearly ten percent. Numerous hit songs, especially in Hip-Hop, the genre now dominating pop culture, are clocking in at 2 minutes and 30 seconds or less; in 2021, it's not uncommon to find wildly popular songs with a total runtime under two minutes. At the time I'm writing this article, six of the top ten songs on the Hot 100 clock in at under three minutes, down from an average of 3 minutes and 50 seconds in the pre-2014 era.

In a related shift, the structure of songs are also evolving due to the pressures of our streaming age. With attention spans shorter than ever, long intros are out –  and immediate choruses, where a song's most pivotal melody is teased just as you press play, are becoming far more common.

Why is streaming driving these changes? The answer lies in our earlier examination of Spotify's algorithms, which as I noted, are built from data comprised of every action you've taken on the platform since your very first listening session. As it turns out, one of these data points carries particular weight: a user skipping a song before it completes. In our streaming world, few actions are as devastating to the artist as hitting the skip button. In the eyes of the algorithm, such an action is a clear demonstration of a user rejecting a particular song. The weight of the skip button is not lost on professional songwriters – I can tell you firsthand that it directly impacts the process of writing a modern record, putting producers and artists in a constant arms race to see who can hack Spotify's algorithm most effectively.

The drive to please the algorithm – which is further heightened by the increasing role TikTok now plays in music discovery – is largely responsible for so much of today's music sounding eerily similar. To be sure, music, and pop music in particular, has always been a game of copycats. Riding trends until the wheels fall off is a tried and true approach to maximizing the odds of chart success dating back to well before the streaming era. Historically, however, such adaptations were typically confined to single genres: rock music tended to copy other rock music, and pop writers tended to copy whatever was working at the moment, whether that was Timbaland's frenetic syncopated beats in the late 1990s or Max Martin's power-pop ballads of the early 2010s.

The gravitational pull of streaming, however, has obliterated these genre-based restrictions. Musicians in a particular genre may still be copying the instrumentation or sonic signature of their contemporaries: what is different now, however, is that nearly all artists are seeking to outdo each other in a race to please our universal algorithmic overlords. Whether you are a creator or fan of music, this trend is worth your attention.

The Aggregation Age

In this piece, we've examined the evolution of digital music consumption over the past twenty years, with a particular focus on the enormous changes brought to bear by the iTunes Music Store and Spotify. These changes, which include the notion of infinite shelf space, the unbundling and rebundling of the album, and the rise of platform-centric algorithmic design will likely impact music for years to come.

Looking at the potential pitfalls of the algorithmic age, however, only reveals a portion of music's evolution. While algorithms may well be biased towards the fat head instead of the long tail, it will likewise be essential to examine the impact Spotify has on the lives of artists through its role as the über aggregator of music. The ways in which dominant platforms can exert profound pressure on their suppliers – willingly or unwillingly – is worth deeper examination as we begin to ponder what the next decade of music will look like.

In the final part of this series (out Monday 12/20), I'll take an in-depth look at the central role of aggregators in our digital economy, consider likely trends on the horizon, and close with thoughts and suggestions on how we can build a better future for creators in the Streaming 2.0 era.


As always, you can find my list of worthwhile reads and other fun things to check out on my Reading List.

Next

Above The API - March 11 2022

On Meta Disruption