CategoryWriting

The double edged sword of personalization, as told through Spotify Discover Weekly

It’s one in the morning. The party has moved from the dining room to my living room. If I had to point to why I gave up on semi-professional cooking to keep it a pure passion, it would be dinner parties like this. Fifteen or so people, probably ten different accents mixing in with that reverberant, waterfly-landing-on-water rippling beat of “Zealots” by The Fugees. Berlin is magic. I remind myself to notice the full texture of this moment in the hopes that it increases the chances that some shred of this night sticks in my memory forever. 

My “Kitchen Vibes” playlist on Spotify is at this point somewhere close to a prized possession. When I was cooking pop-ups in San Francisco, my co-creator Peter and I put the list together through trial and error over countless hours of events. It bumps and glides, it flows from top to bottom, it mixes well-known with pleasant surprise. It starts smooth and groovy with Wilson Pickett, Gramatik, and others and by the end it’s meant to be the soundtrack to the “I probably had one too many glasses of wine and this event is fucking incredible and please God let me tip these chefs”… think “Int’l Playeres Anthem” and “Shaolin Monk Motherfunk.” Shuffle in the privacy of your bedroom, but if you want to host an event please start from the top.

The reason I tell you this is because I’ve heard “Kitchen Vibes” hundreds of times, and I happen to know that when the Fugees drop the beat it means I need to look around and see if the party is still on the up and up, because my playlist is nearing the end. And on that night in Berlin, the party is most certainly still on the up and up. One day I’ll have a ready-made sequel playlist, but for now it’s time to hit the search. 

My Discover Weekly is the only thing in my list of playlists that is truly alive, changing on its own every week. Perhaps that’s why I feel it’s staring back at me at this moment as I look through my Spotify library. To be honest I can’t even remember what’s in there right now, but I do know that I’ve been playing it on repeat. I can remember at least three isolated instances of dancing by myself this week – it flows and there are definitely jams. I put it on.

I don’t know whether it’s mostly better machine learning models, period, or also the progress of more data on me entering Spotify, but my Discover Weekly playlists are absolutely incredible these days. When I was young, every once in a while I’d get a new album that just hit right, and for about a week I’d play it on loop. These days I have that experience probably one out of every three weeks. Increasingly, it feels like an album that flows rather than just a collection of songs, which makes sense because the point is to play it like one.

Anyone who loves to host events or parties knows what it’s like to watch the energy build or dissipate out of a room, and also how easy it is for small things to change the direction of that flow. Adjusting the lights, changing the music, or forcing a soft reboot (for example by moving people to another room) quickly turn the tide.

And on that night in Berlin, as the third song plays from my Discover Weekly, I’m watching the energy fade as if something impermeable was suddenly removed from under the cracks in the floor of my old apartment. 

In the moment, I have no time to think. I’ve done this enough – I just react. Without looking panicked, (nothing kills a party faster than the realization that everyone is wondering whether the party is dying) I casually excuse myself from my conversation and slide back over to my phone. I pull out one of my life-rafts, Rumors by Fleetwood Mac. I put it on and everyone climbs in; we’re floating again. Eventually the party really turns and other people take over the music to play songs that people can dance to in my living room. An incredible evening, with just one brief energy crisis.

Days later, I’m looking back at that moment and I realize that my Discover Weekly would not be a well received playlist if it was released publicly. I’m not saying my music taste is anything more than fairly basic, but that playlist is so personalized to me and to a moment in time that mass appeal has been lost in favor of personal relevance. 

There’s a concept in science that most scientists’ unique research interest can be described by placing them at the center of some number of overlapping topics. For example “art therapy * PTSD * earthquakes” (I actually met this person recently). But since that center of the venn diagram is a lonely place, they tend to collaborate and attend conferences and such in parts of the diagram where you’ll find more people.

I love my Discover Weekly, and I’m listening to far better and more interesting music than I would without it. My library would be what people used to call “cool” in that it’s unique, diverse, opinionated, not just a reflection of the Billboard 100. And each week my Discover Weekly is the center of my ever more complex venn diagram – the exact reflection of my taste.

But as with science, the bullseye of my interests is a lonely place. When I used to find albums or artists I loved, I could easily find other people that loved them too. My taste was more basic, more influenced by whatever CD was placed front and center on the display table or popular amongst my friends. These days I hardly know who or what I’m listening to – I only know I like it, a lot. 

For me, music makes the complexity of the discussion around recommendation algorithms clearer and easier to grasp. There may be fake news but there’s no fake music, but the exactness of my Discover Weekly gives me the feeling that so far, personalization is turning the dial in a somewhat zero-sum game of relevance vs connectability. Spotify is learning how to deliver amazing music directly to the center of my unique intersection of interests, but I’m sacrificing the connective power of the outer layers in the process. I can no longer see over the top of my musical ditch to connect with other people. But please don’t force me entirely back to the surface – I love the sound down here.

Music helps frame the challenge ahead and the need for balance. Can Spotify figure out how to continue advancing the algorithms and deepening people’s personal taste while also maintaining the social element of music discovery which relies on some inefficiency in personalization? The gradual abandonment of the social features in Spotify would suggest it will be an uphill battle. Can personalization and connectability be more than zero-sum? As an optimist in the power of great products and a lover of Spotify, I think so.

Today, I listened to my Spotify Weekly for an hour or so while I was writing. It’s a good one. But then later I also started building the sequel playlist to Kitchen Vibes, and I invited a few friends to build it with me. That feels like a good balance for now.

Subscribe to get the monthly review

From Einstein to Bezos: What Science Can Teach Us About Creating and Disrupting Growth Models

This post originally appeared on Reforge, and can be read in its entirety there.

One of the things that has struck me in my transition from neuroscience labs to tech is the deep connections in tech to the way the world of science operates (including heavy adoption of scientific terms and techniques), but a general lack of depth in the understanding of what science can really teach us about how to do these things well. Even at ResearchGate, the largest platform at the intersection of tech and science, the awareness of the analogy of our work to science rarely extends past experimentation techniques. I believe this misses a great deal of insight; there is a lot more that we in tech can learn from how science has worked for hundreds of years to build knowledge and drive human progress. 

Read more at Reforge…

How to Run Your Growth Team Like a Scientist

This post originally appeared on Reforge, and can be read in its entirety there.

Recently, a colleague was telling me about a “successful” experiment they were excited about. They hypothesized that the best time to encourage users to upgrade from a free trial was while they were using a certain common feature.

To test this they put a big button on that page and saw a measurable lift in conversion compared to users who didn’t see the button. Based on this result, they were attempting to drive more free users to that feature.

I asked if they had experimented with putting the same button on other commonly used features, and they hadn’t. They hadn’t controlled for the risk that the lift was simply a result of putting a prominent CTA in front of more users.

Read more at Reforge…

The Growth Experiment Management System that Tripled Our Testing Velocity

This post originally appeared on Reforge, and can be read in its entirety there.

All the fastest growing companies move at lightspeed.

Uber, Airbnb, Amazon, and Facebook are some of the fastest growing companies in history, in part because they built high velocity testing into their cultures from day one. Lindsay Pettingill, a data science manager on Airbnb’s growth team, reports that her team has increased their experiment cadence from 100 to 700 experiments per week over the past two years.

Read more at Reforge…