Traditional AM/FM radio and online music services like Pandora and Spotify are increasingly battling over the same listening time and space. Radio listening is happening more and more frequently via streaming services so that listening on the laptop, desktop or smartphone is becoming a more contested space. At the same time online music services are increasingly moving into the prime area of radio dominance: the morning and afternoon commute. Smartphones are more frequently connected to car radios, and here.. Lest the revenue implications are not obvious, Pandora is now hiring traditional radio sales executives in major US markets to draw advertising dollars away from radio. The competition is direct and will be fierce, and although the “death of radio” is very unlikely (especially considering its role globally in news and information delivery) there is great incentive for stations need to treat their online stream as seriously as their terrestrial signal. To compete most effectively in the digital space and remain a vital and ubiquitous medium, I believe radio needs to carefully and thoughtfully reconsider its online existence. I would like to suggest several specific ways in which radio can catch up with the data revolution which would allow it to deliver a better product to both its online and terrestrial listeners.
The radio/streaming services debate is sometimes cast as a battle of “humans vs. algorithms.” But this distinction is problematic and not what will ultimately drive listeners to one service or another. Pandora employs hundreds of real people as music analysts to carefully categorize their library. Radio stations rely on scheduling software that uses algorithms to make sure a song isn’t played in the same time slot every day and that different song categories are rotated in a very precise way. Some internet-only radio stations rely on a mix of hosted shows and the shuffle feature available on any digital music player. And some FM radio stations have one employee keeping the lights on and just rebroadcast content beamed in from a conglomerate headquarters. To paraphrase Fred Jacobs, Pandora isn’t radio, but the radio industry may no longer count as “radio” either. Clearly hosts are important to the radio experience as they add context and personal connection, but a host with a weak playlist, interrupted frequently by commercials, is increasingly a low-value proposition.
I see three ways the radio industry might look at this situation. The first is the “race to the bottom” view. If Pandora is hiring sales executives then they plan to run more ads. Pandora will therefore gradually start to sound more and more like commercial radio (minus the hosts), so that the difference between the products is actually lessened, reducing the probability of listeners moving away from radio and toward online services. It seems unlikely that even the paid subscription services like Spotify Premium can remain commercial free. The business model as it currently stands looks challenging, and the temptation to bring in extra revenue will be great. I call this the “ads at the movies principle.” People were furious when ads started creeping onto movie screens, but over time this somehow became less offensive and is now accepted by a majority of people (Owczarski 2013). Whether these predictions are true or false, this is the view that radio need do nothing and everything will turn out ok in the end.
The second way to look at these changes is to believe that radio has an inherent format advantage. The view is that playlists with significant human input still sound better than algorithms and that hosted shows provide a superior listening experience. While the playlist argument might be true today, at some point the algorithms will catch up. Additionally, having a better sounding product does not equal success. Just as the 128 kbps MP3 sounds objectively worse than a CD, MP3s still won the format war because people either didn’t pay close enough attention to be bothered by the difference or were willing to sacrifice audio fidelity for ease of storage and transportation (Guberman 2011). It’s easy to imagine people turning to online streaming services not because the playlist is as good as radio, but because there are fewer commercials, or because skipping songs is more interactive, or because it’s the first thing they see on their phone or their connected car’s dashboard.
The third option that I see is to embrace the challenge to make ever-better playlists by embracing a data revolution in traditional radio. Radio needs to constantly tweak its own music programming formulas based on solid listener data. Doing so will require that music directors learn new skills to research and analyze data. Much like trends in baseball, I would argue that looking at the sales charts that drive Billboard rankings is like looking at batting average. Radio stations need to start looking at OBP and BABIP to compete more effectively with the Pandoras and Spotifys of the world, where there are real people working every day to make music recommendation algorithms more closely match their own intuitions.
How do we find the OBP of radio? Here’s one idea: songs don’t age the same. Given the high play rate of hit songs and the high exposure level to listeners, some songs age more gracefully than others (Morimoto and Timmers, 2012). Most people might be sick and tired of hearing a hit by artist A after one month, but a hit by artist B is still enjoyable 3 months later. Currently most stations put a song in heavy rotation for a fixed period of plays. The song might play twice a day until it reaches 240 plays, and then be moved into a less frequently played category. This is inefficient: very few songs will actually have crested from “most people like it” to “most people are sick of it” after 240 plays. The vast majority of songs will either have been played too much, meaning people are more likely to stop listening to the station when that song comes on, or not enough, meaning continued plays would please listeners.
So how to get this data? Pandora and Spotify have been collecting exactly this kind of information since their inception in the form of thumbs up/down and skips. If radio stations added an upvote/downvote button on their internet players or apps and encouraged people to use it to send their feedback about a song to the station they would have an absolutely fascinating data set. This data should be anonymous to protect listener privacy, but would be no less valuable as such. Let’s take this idea one step further: some stations are too small to act on this hypothetical information. For example you wouldn’t want to make important programming decisions based on 2 upvotes and 5 downvotes. What if radio stations anonymously pooled their up/down information, creating a vast and open online dataset of listener preferences over time? Music directors could research trends on the new Katy Perry song and have a much more accurate indicator of when to play it less compared to Billboard sales or a pre-determined number of plays. To briefly mention technical implementation: if a station is not sending artist/title info in the audio stream they obviously wouldn’t be able to contribute data, but all stations should be doing this unless they are still using CDs or vinyl. Code would need to catch misspelled songs, but artist/title could be checked against a database of all known songs for likely matches. If stations are unable to cooperate in this manner an alternative was suggested by my colleague Andrew Whitacre: stations could offer to pay Spotify or Pandora a small licensing fee to get access to their already existing huge data sets of skips and votes.
Here is another way radio stations could take data seriously and benefit: digital audio platforms excel at knowing exactly how many people listened to a station and exactly how long they listened for. Nielson (formerly Arbitron) charges tens of thousands of dollars to every subscribing radio station to provide estimates on the number of listeners, when they listen, and demographic information about them. In the digital space why would radio pay for something as basic as this? One answer is the “trusted third party.” Stations could easily puff up their numbers to lure prospective advertisers, whereas Nielsen ratings are less open to manipulation. But businesses like Triton Digital and TuneIn are already acting as trusted third parties for listener stats in the digital space. It would be quite interesting to perform a careful statistical analysis for one or more stations comparing online listening numbers to Nielsen ratings to see what the correlation looks like. If free web analytics tell you as much as Nielsen ratings then radio stations should think about leaving the expensive PPM system and trusting their web stats.
One last idea about online players: the player should (anonymously) track the song that is playing when someone closes it. This is the classic “tune out” in radio speak. Such a data set could show exactly what was on air when people stopped listening and could reveal problems with hosts, segments, or even songs that have been in the library for years that are just not well received.
These are just a few ideas about how radio stations might reconsider their online existence. Radio is a rich and historically important medium, but resting on laurels at this juncture risks losing even more revenue and more jobs than necessary to online services. Radio stations can better compete against algorithmic services if they take control of their data, embrace an open data model, and build better features into digital players. Most of all, let’s get everyone who is passionate about radio thinking deeply about how it can not just survive but thrive online.