Published on January 31st, 2020 | by Clayton Durant0
Paul Maloney, AccuRadio’s Head of Music Programming On Building Playlists In The Age of AI
Machine-learning algorithms made possible by a combination of deep learning and artificial intelligence have dominated 2019. Moreover, the dramatic rises in consumer expectation levels are forcing businesses to simplify and personalize everything in a bid to remain relevant to their tech-savvy customers. This reality is no different in music. Every major platform, from Spotify to Apple Music have to create unique experiences to over 200m plus consumers. In an age where AI is mature enough to simplify this customization process, there are other companies like AccuRadio who are taking a more “human-curated” approach to music programming.
I got a chance to sit with Paul Maloney, AccuRadio’s Head of Music Programming in which we discussed why human playlisting is more important than ever in the age of mature AI and cognitive technologies.
It was really interesting to see that Spotify has a skip rate of 40-50% while your platform only has a skip rate of 4%. Can you walk our readers through the process of how you are curating playlists?
We have a team of veteran radio programmers that are each responsible for a group of channels per one or more genres of music. We mostly come with the experience of broadcast music radio… the shorthand version of which might be “less is more” and “always be playing a better song.” Also known as “play the hits”. Listeners stick around for what they recognize and are comfortable with and come to us for audio “comfort” — to relax, destress and focus — needs best served with favorite, familiar music.
In terms of data, can you describe how data plays a role in the playlist curation decision making and packaging process?
Well, certainly national chart performance, broadcast radio monitors, and a memory of what tested well when our programmers worked in broadcast radio all counts as “data.” We have song ratings from millions of songs each week, but most of our staff don’t really rely on them in creating channels (with the exception of the Listeners’ Top 100 channels, which are, by definition, based on user ratings). Our programmers select music for their channels based on what they believe “fits” rather than how a song may have performed on other channels.
Do you think human programming is scalable? I mean, Spotify has to serve alone over 70M paid users. Their algorithmic playlists are used to create custom experiences for each user. How can human programming scale when each user has different tastes?
We “offload” some of the work of customizing the experience to the listener — through their star ratings, song and artist bans, channel blends, and Five-Star Radio picks (a custom collection of each listener’s top-rated songs). Algorithms win on scale, human input wins on quality. While our current programmers are pretty maxed-out, we are working towards employing full-time “brand managers” for every genre to handle music programming for a dozen channels as well as features, marketing, social media, production, etc.
When it comes to artists looking to get into your playlists, how can talent, their managers, and labels connect with your programming directors?
What things do your human programmers look for when they are trying to find new records to put into playlists?
Since we focus on going deep on favorites and familiarity we look for performance and familiarity on other platforms: FM radio, on-demand services, TV… whether currently or in the past. The “fit and feel” of a song in the context of other things on a given channel is another factor. Our programmers work to understand audiences for various channels and genres and try to deliver what makes sense to their ears. A skilled and passionate programmer will include a little “spice” in any channel — that is, a song or artist that “works,” but perhaps in a non-obvious way — but the bulk of what’s on a channel are those songs that fit the idea behind the specific channel.Tweet