Scott Pinkmountain Talks Working as a Music Analyst for Pandora
For almost nine years, I’ve been a music analyst for Pandora. I also happen to be a practicing musician. My work has run the gamut from pure improvisation, free jazz and avant-garde large-ensemble music to country, rock, and Afro-psych, and has led me to collaborate with folks who share similarly omnivorous tastes, like Anthony Braxton, Sam Coomes (Quasi), Arrington De Dionyso (Old Time Relijun), and trumpeter Nate Wooley. I was excited about being an analyst at Pandora, because it was the first job I got where I needed to use my musical abilities to do my work. I had no clue what I was getting myself into.
I most likely thought what nearly everyone says about my job — “That’s so cool, you get to listen to music for a living.” Which, technically, I don’t. As a co-worker pointed out to me, we don’t get paid to listen to music, we get paid to analyze it. There are about 25 of us. Our work gets turned into algorithms which are paired with other similar algorithms to create Pandora’s personalized playlists. We take a song like “Billie Jean,” and identify that it’s just below mid-tempo with a strong backbeat, prominent riff (in the synth bass), minor key harmony, mostly i-iv chordal motion, highly expressive and identifiable vocalist singing about love (gone wrong), effected guitars (mostly rhythm, but a touch of riff), synths, back up vocals, hand percussion, recorded with high production values. Stylistically, it’s a combination of pop, r&b and funk with the main compositional elements being groove, that memorable riff, and the lead vocals, and culturally it’s as iconic and mainstream as a song can possibly be. It gets paired with songs that share whatever elements the algorithm makers determines to be of most value to listeners (i.e., what they like about this song that might make also make them like the next one).
The process of listening to music can be many things: emotional, physical, sensual, cerebral, meditative, etc., but analyzing music is solely intellectual. You have to step back and disassemble the musical object: figure out how it works, what are its ingredients, determine if it’s achieving the goals it set out to accomplish. Then move on to the next song. Then the next. Then the next. The number of songs I’ve listened to closely with headphones on while parsing their structure, chord progression, instrumentation, vocal qualities, melodic shape, lyrical content, production values, genre influences, compositional priorities, and cultural placement is somewhere around the number of black flies in Wisconsin.
I worried, is this going to ruin me as a musician? Would it turn me into a giant-foreheaded, unfeeling automaton, generic standardized music-stamping robot of a songwriter? Would it make me unable to enjoy a song as a whole experience anymore, and only hear it as deconstructed constituent parts piled greasy on the workbench?
Initially, that’s exactly what happened, at least with listening to music (as opposed to making it). I vividly remember lying down for a nap sometime in my first few months at the job. As I dozed off, my newly trained mind started analyzing the sounds of nearby construction work, according to standardized parameters. Instrumentation: circular saw, nail gun, hammer. Vocals: male, non-text lead vocalizing, gritty. Rhythm: irregular, no backbeat. Live performance, blend of electric and acoustic sonorities.
I found myself analyzing everything I heard: live shows, Muzak, children’s songs. We analysts talked about this a lot amongst ourselves in the break room, fretting we’d never again be able to get lost in a song, caught up singing along to a boneheaded chorus or wah-wahing along with a guitar solo. But then it just… went away. I must have internalized the process enough that it no longer occupied any frontal lobe RAM and I was able to get on with the business of listening to music while banging head and pumping fist.
But listening to music is different from making it. I’d studied music composition before I landed my Pandora job, both as an undergrad and a graduate student, so analysis was something I was interested in, had trained to do, and believed was of at least some value. But I’ve now ingested galactic amounts of music, and I would do anything to turn off the spigot and un-hear much of what I’ve heard. The fear here, of course, is for my own musical identity — that my creative spark will be extinguished in the flood of other people’s music.
I’d been at Pandora a couple of years when I was gearing up to make my album The Full Sun. I had a batch of songs and knew I wanted to make a “big” record, but I was completely overwhelmed by how to move forward with the material — I didn’t know where to start because my head was crowded with too many options and influences. So I took an analytic approach. I sat down and made a list of records that really mattered to me: Mingus’ Mingus Ah Um, Yo La Tengo’s And Then Nothing Turned Itself Inside Out, Neil Young’s Harvest, Coltrane’s A Love Supreme, the Zombies’ Odessey and Oracle, Elliott Smith’s XO. Then I tried to figure out what they have in common, how they were built, that contributed to making me like them? Obviously the songwriting was spectacular, and that was something I could only strive for, but there were processes I could reproduce: musicians playing live together, really clean, transparent production, a kind of epic, broad-scope intention or ambition, innovative arrangements, keeping the soloing emotional and direct, avoiding technical excess. The Full Sun wound up being pretty much all over the map, stylistically, but it didn’t get criticized for being derivative in part because analysis helped me gain access to, and learn from, the core of the music, as opposed to staying up on the surface level and simply mimicking what the music sounded like.
Analytic or intellectual intentions only enter once I’ve written a song. Music is feel first, logic second; switching that order, trying to create based on ideology or principle, has not yielded awesome music from me. We analysts joke a lot about taking the raw analysis from a highly requested track on Pandora, and reverse-engineering a song from the data. So far, no one has actually done this and I doubt anyone ever will — because it’s almost guaranteed to suck.
Taking in so much music can wipe out any will to make more music. Sometimes, hearing the very best stuff can make me feel defeated, the very worst of it can turn me off from music in general, and the rest — this vast, overwhelming, non-stop deluge of far more music than this planet will ever need — can make me question the value of adding to the din. But those are important experiences to have. For one thing, getting crushed is essential for any artist, to be defeated by someone else’s greatness and still feel intensely compelled to make music. And a realistic understanding of just how much stuff is out there makes me think twice before I hit record, maybe sending me back to cut an unnecessary verse, learn a new chord voicing, rout out a few more clichés, read more non-fiction.
I notice myself feeling less and less self-conscious about what I’m doing as a musician, a songwriter, a composer. Things occur more intuitively both in the writing and the editing. Maybe that’s because I’ve written so many songs. But maybe it’s somehow a reaction against the nature of my job at Pandora. Or maybe it’s because of it. But whatever the reason, when I make music, I no longer consciously think about what the key the music is in, what’s the role of the different sections, how should it be recorded. I think it’s all just… in there now, for better and worse — the overwhelming spectrum of good and bad influences, the brilliant production tricks and bland off-the-shelf presets, the thousands of clichéd rhymes and the one-of-a-kind chord progressions, the genius and the numbskulled, the sacred and profane. My main responsibility, and all I really can do as a songwriter, is to filter it through myself with honesty.