Jeff Smith, Head of Music, Radio 2 and 6 Music at BBC, speaks about his work at the Sónar+D conference in Barcelona in June. (Photo courtesy of Sónar+D)
By Cherie Hu
On a sunny Thursday morning in Barcelona last month, dozens of professionals gathered under a wooden awning at the Sónar+D conference to hear prophecies from opposite ends of the music curation spectrum: algorithmic playlisting and old-school terrestrial radio. It was a jarring but much-needed juxtaposition, suggesting that curation as we know it is both more fruitful than ever and in danger of dying out.
The voices on the algorithmic side were Spotify’s Matthew Ogle and Ajay Kalia, the streaming service’s Product Owners for Discover Weekly and Taste Profiles, respectively. As many in the music industry are aware, playlisting is Spotify’s dominant currency—its 100+ million users have made over two billion playlists to date, an average of over 20 playlists per user and counting—and Spotify has used this currency not only to draw a detailed map of every user’s taste, but also to deliver high-quality, automated recommendation services.
One of its most popular recommendation features, Discover Weekly, analyzes the relationships between songs that result from playlisting activity in the aggregate (a feat of “collaborative filtering”), and creates a custom playlist for each user that refreshes every Monday morning. Since its launch last year, these automated playlists alone not only have accounted for five billion streams, but also have become an important force in some emerging artists’ streaming stats, with at least 8,000 artists receiving over half of their Spotify streams from Discover Weekly. Indeed, as Ogle and Kalia discussed in-depth, algorithmic playlists help both artists and listeners by connecting the right music with the right ears, all while centering on granular personalization—in Kalia’s words, the operative motto is “it starts with you.”
Does this mean machines are the best musical cartographers? Jeff Smith, Head of Music, Radio 2 and 6 Music at BBC, would suggest otherwise. He took the stage after Ogle and Kalia that morning and, among other topics, described a typical BBC radio playlist meeting to the audience (you may be thinking: people still hold meetings for playlists? Can’t we just let algorithms do the work?) Such meetings take place at a round table of 12 to 15 representatives from the station’s daytime and specialist music programs, and Smith has one deceptively simple request for each of them: champion the music they are playing over the air, without using any data analytics.
“I could sit around looking at plots,” said Smith, pointing to his phone, “or I could listen to what human beings around the table want to fight to get onto that playlist.” He cited “Paranoid Android,” the lead single from Radiohead’s 1997 album OK Computer, as a key example: it was put straight to the radio’s A-List at the time due to sheer love for the song, but would not have been close to making the list with purely data-driven methods...
Read the full story by Cherie Hu over at forbes.com