Glenn McDonald, as the cover says, was formerly Spotify’s Data Alchemist. He joined when his previous company, The Echo Nest, was purchased by Spotify in 2014.
In many respects, this book scratches a lot of itches for me, getting into some of the details about how Spotify works, and in particular some of the algorithmic details.
Except, I’d have liked more than we get here. For reasons no doubt related to NDAs, and company secrecy, we don’t really get into a lot of that, but there’s still a lot to understand.
One of the most illuminating sections is where McDonald defends the pro rata model that Spotify and most others use to payout artists and rights-holders versus the user-centric model.
Back in 2015 I wrote a long piece about these two models – I referred to them as the big pool and subscriber share models. But they’re the same thing.
In the pro rata/big pool model, all the revenues from all a music streaming service’s subscription fees (and advertising revenues from free tiers) go into a large pot, and is shared in accordance with how all users listen to that music. As McDonald puts it, even if I’m not listening to Ed Sheeran, some proportion of my monthly fees are going into his pocket, because he’s very popular.
Many have argued that this might seem unfair. If I’m instead spending all month listening to a single niche artist, why don’t all my revenues go to that artist?
But McDonald makes that point that more active listeners to a service like Spotify actually mean that it’s better for those niche artists to use the pro rata/big pool model. In essence, these are the keener music lovers who listen to wider repertoires and for longer. And he has had access to the data to prove it. Because in truth, if I’m a lover of more niche artists, I’m probably listening to more and a wider variety. On a subscriber-share model, I’m delivering less value than if we mix me in with everyone else.
The book also tackles subjects as diverse as the impact of fan armies, the various kinds of frauds that have been attempted and much more.
I think if there was one area that isn’t really broached though, it’s the change in how music is now made in a world where you’re only paid if a listener hears 30 seconds of a song. Songs are now shorter as a consequence (fewer songs in a set-time period means more payouts), and the structure of songs has changed so that the chorus or hook of a song comes right at the start. It’s kind of like how YouTube has changed movie trailers to such an extent that there’s a pre-trailer trailer of the trailer you’re about to watch to grab you in the first 3 seconds of a three minute video!
This all feels a bit like those old 70s and 80s US action TV series that opened with a “Coming up in this episode…” montage showing clips of all the cool car chases and stunts you’d see if you hang around. (The remake of Battlestar Galactica knowingly did this too, which was a call-back to the original 80s series, while Mission Impossible movies do this to an extent in their opening title sequences, again calling back to the original 60s/70s TV series).
The book is very chatty in style, and there are plenty of diversions into the way the music industry works. McDonald is also honest about times when functions and algorithms didn’t work and how he fixed them. There are also diversions into why Christmas songs get a big jolt on 1 September, or into forms of music that many of us don’t know – Brazilian Country music or Australian Hip Hop. A service like Spotify has to classify everything! McDonald has a website that lets you explore more than 6,000 genres.
Another area that gets prodded a little is how the model probably doesn’t work very well for classical music where movements are not 3 minutes in duration. And of course the metadata can be all over the place with a repertoire where the conductor or the orchestra is more important than the “artist” or composer and the title of the piece.
And it’s not really within the remit of this book, but I’m still interested in the financial model that underpins subscription music services. In 2024, Spotify charges £11.99/$11.99 a month for a normal subscription, but that’s slowly morphing into a bundle which means that some of that money is going to audiobook authors and publishers or, in the UK at least, video learning courses. Yet, as McDonald notes, the annualised cost of Spotify is much more than most people used to spend on music. I still think back to the “fifty quid bloke” coined by David Hepworth in 2003 at a BPI meeting. These were those people (and I used to be one of them) who traipsed into a large HMV or Tower Records, and came out with perhaps a couple of CDs, a DVD and maybe a book. They’d spent £50. And they’d do it every month. I swear that Tower Records in Piccadilly Circus was only open until midnight to trap those like me who’d spent an evening in the pub and suddenly thought a quick shopping trip on the way home on the Piccadilly Line was a good idea…
But the music industry made a lot of money from these people. And now… well they spend £11.99 a month like everyone else. Not £50 a month. That’s a big loss.
At the other end, we’re trying to get someone who bought the equivalent of two albums a year, to now spend £144 a year on music. That’s a big ask.
At the end of this book, McDonald makes the case that algorithms need to be more open about how they’re achieved. There is certainly a belief amongst some that other imperatives are coming into play when we get the music we’re presented with – because when you have most recorded music in history available, you definitely need a lot of help to navigate it.
But as becomes enormously clear in some of his anecdotes, even presenting an algorithm in a non-computer code manner is not simple (e.g. screening out Christmas songs when it isn’t Christmas, even if you love that artist).
McDonald has a love of music, and this book shows that. I might not share his love of metal, nor his hatred of jazz, but this is well worth a read if you want to get into the weeds of streaming music today.