Spotify Teardown: Inside the Black
Box of Streaming Music !
Hello Friends,
The multi-authored monograph Spotify Teardown is a book very much of this moment. As well as considering the streaming platform’s “front end” through consulting news coverage, company blogs, financial results, interface analysis and more, it also uses numerous experimental methodologies in an attempt to get at Spotify’s mysterious “back end”, and its particular algorithm-enabled wrangling of the digital music commodity.
Internet researchers stand blinking in the cold light of a “post-API” age. The open access to data which platforms permitted in order to benefit from third-party additions and augmentations has been jettisoned for a new business imperative of holistic control, and thus a primary avenue by which scholars sought to access and interrogate the world of big data has been closed off.
“Where Is Spotify?”,
Referring both to the company’s geographical placing (and how it has leveraged its Swedish-ness for preferential political treatment), and also its place within and across the industries of music, IT, finance, and advertising. We join founders Daniel Ek and Martin Lorentzon in 2006 as bored ad-tech millionaires with vague plans for developing a content delivery system; a key argument is that, from the off, their strategy has been “opportunistic” rather than “innovative”. Spotify’s initial appeal depended upon an ambivalent relationship to illegal peer-to-peer sites like The Pirate Bay, “sometimes presenting itself as the continuation of the ongoing illicit disruption, while at other times insisting on a binary opposition between illegality and legality”. The proposition that the platform began as a “de facto pirate service” has provoked the ire of Spotify’s legal team, but seems accurate. Organising the firm’s history by rounds of investment clarifies how the big record labels were able to determine Spotify’s direction – e.g. pushing towards paid subscriptions rather than ad-supported free content. It’s an exceptionally well-researched chapter, which exhumes Spotify’s buried narratives in order to thoroughly counter the platform's revisionism.
concerned with how Spotify “turn[s] files into music”
A premise that lacks a coherent theorization of what music is – and concludes that it “involves an exceedingly interrelated data stack and a complex streaming infrastructure of software services, metadata, and user-generated data”. Chapter 3 excellently employs interface analysis to consider how Spotify “packages” music. Default playlists are instilled with a “self-help ethos” that denotes a functionalist use of music to manage emotions and, crucially, this is linked to Spotify’s desire to provide advertisers with insights into the affective states of listeners.
Playlists also make “chrono-normative”.
assumptions about users’ work and leisure practices. Chapter 4 employs economic sociology to hypothesize that if Spotify is “embedded in finance”, this ought to have observable consequences for their actions across other markets. But the search for such evidence is marred by a cart-before-horse approach wherein “brokerage” and “arbitrage” are immediately adopted as “sociological types” ; as such, the chapter becomes less of an empirical hunt and more of an interpretative game. Each chapter is followed with details of a related “intervention” that the team carried out, mostly using bots, packet sniffers and data scrapers to attempt a “teardown” in the technical sense of identifying and separating out the technical components at work within and around Spotify.
Some interventions are less techy, such as setting up a “fake” record label and submitting “fake” music to Spotify via third-party aggregators. A lot of the attempts to get at the “black box” are marred technical issues of various kinds. More disheartening though is that the more successful experiments rarely generate anything particularly surprising about how algorithms work – it’s all pretty much as one might suspect looking from the outside. It’s also hard to know what’s at stake, and quite where it connects to the book’s “critical” perspective. One experiment reveals that user–platform data traffic contains “hundreds of minor malfunctions”, but this is then acknowledged to both be “highly common” and also “unnoticeable” to the user, somewhat undercutting the anthropomorphic dramatization of software “break[ing] down and misbehav[ing]”. The authors are generous in their discussion of their failures and their successes, and in that sense it is a book about methods as much as it is about Spotify.
But it is actually light on epistemology, and heavy on step-bystep elucidation of research design. I found the tone of the book at times unnecessarily, distractingly provocative. Spotify does not “invite”, but rather “summons its users”; Daniel Ek is described as “pale and sweaty” twice on one page. This animosity is perhaps understandable, given that Spotify sent a cease-and-desist letter to the researchers, and then appealed – unsuccessfully – to the Swedish Research Council for the project’s funding to be withdrawn. The introduction and conclusion are structured around this horrible abuse of power, and the unenviable and stressful situation the research team found themselves in. But as the authors lean in to this skirmish as a source of critical credibility, they adopt a role of activist prankster-provocateurs which threatens to overplay their hand.
In the “Songblocker” intervention,
concerning their app which inverts an ad-blocker (i.e. silencing songs and playing adverts), seven pages are spent outlining the myriad ramifications of what seems to me a rather straightforward, if well-executed, gag. Spotify Teardown offers two visions of post-API internet research. In one, the “black box” continues to be approached as the primary source of answers to the question “what are platforms doing?”. Based on the experiments offered here, this may be an arduous struggle – lots of effort expended for minimal insight, and plenty of rumination on failed experiments. In the other vision, original and important answers to the question “what are platforms doing?” are found in examining the rich data that they leave strewn “outside”: their reshaping of the cultural industries, the impact of financialization on personal communication, and so on. These approaches overlap, of course. But we should be careful that, because one object is out of our view, we do not assume it to be an oracle and, consequently, follow platforms down a cul-de-sac in which we play by their techno-centric rules