Will Apple’s Tastemaker Test Win The Streaming Music Challenge?
Apple made big news last week by hiring one of the music’s best tastemakers, Zane Lowe, the preeminent DJ on BBC Radio 1 who has introduced the world to artists like Arctic Monkeys, Gnarls Barkley, Adele and Sam Smith.
With Zane’s hiring and the reported tapping of other music journalism talents, Apple is betting big on the ‘human curation’ chestnut that Jimmy Iovine used to sell the service to music fans, and more importantly, to Apple last spring.
Curation is believed to be a solution for streaming music’s problem of what to play next. All-you-can-eat music services like Spotify and Beats provide access to tens of millions of songs, but listeners consistently run into the issue of figuring out what they want to hear next. So by creating recommendations, radio stations and playlists that the music fan might like, curation helps alleviate the problem.
Except it isn’t that easy.
Why? First, is there’s a lot of music. Millions and millions of songs are available on these services and figuring out everything about the music is rather difficult. And then there’s the user expectation. A broadcast radio tastemaker like Zane is pretty adept at talking to a lot of people at once, but streaming the music customer expects—if not demands—a unique music experience based on their taste and listening habits.
The Beats Formula
Curation solutions have come in two flavors. Companies either use automated technology solutions, like Pandora’s ‘music genome’ and the Echo Nest’s taste profile. Or you hire a staff of music experts to pick music.
Beats’ Co-CEO Jimmy Iovine and Chief Creative Officer Trent Reznor rightfully pointed out that most services have the soul of a hard drive and that music fans craved more in a music experience.
Beats preferred playlists selected by humans, experts on music who understood what the listener needed music for, like cooking dinner, exercising or studying. The startup went on a spending spree, hiring a team of music programmers to build playlists and pick the perfect song. While others, like Rhapsody and Emusic, had staffs of curation experts long before Beats, Jimmy was the first to make human curation the main selling point.
When it launched, Beats had subscribers select their favorite style of music. Afterwards, the service would feature playlists built by their staff of music experts who hailed from the radio industry and music blogs. Beats playlists were indeed compelling but the depth of the lists appeared to be light and the curation stale. After all, how many times can you listen to the same 15 tracks on the Indie Breakup or 2006 Hip Hop Gems playlist? Fact is hand curation requires a lot of hands to consistently churn out new lists, something the service didn’t quite get right.
Beats management objected to algorithms that automatically choose the next song based on a set of rules. “The promise of algorithms that we’ve all bought into over the past few years, that you enter a band and you are going to hear a ton of music that’s all based on that seed,” Trent Reznor told USA Today last year. “I think we’ve all realized the reality of that is that it’s a shallow puddle, it immediately kind of sounds good and then you realize the limitations and you start to hear the machine in there.”
“(With an algorithm) you are using math to solve an emotional problem,” is the way Jimmy Iovine put it. He is partially correct. When the catalog is tens of millions of songs and you have millions of customers, picking what song comes next can only be tackled by math.
It’s impossible for a service to function without any algorithms. There’s just too much data and you need to rely on something with automated rules to do some of the heavy lifting. Even Beats, despite its marketing message of ‘the music service with music experts’ had several different algorithms that were used in the service or under development.
So marketing pitch or not, everyone (in one way or another) must use math to solve these problems. The success or failure of algorithms and curation depends on how companies employ the products and who’s in charge.
It’s far from me to tell Apple what to do, but hey, that’s never stopped me from dispensing advice of questionable value. Here are my guiding principles for building curation and algorithms in streaming services.
- The Right Tool for the Right Job
As much as I have a problem with Pandora and their marketing of the ‘music genome,’ the company sure went about solving the right problem with their algorithm. Simply put, Pandora is designed to serve up around 40 solid minutes of songs for the person who likes to listen to music. It doesn’t do more than that and that’s a good thing.
Technology products get unwieldy because they are designed like a Swiss Army Knife. My general rule is that technology solutions need to be designed to nail one solid use case at a time. Expansion beyond that gets to be tricky.
A good example: I recently spoke to David Porter, CEO of 8tracks, a radio service that features playlists curated primarily by the service’s pro DJ community. David mentioned that 8tracks had recently hired a data scientist to match his listeners to playlists that they might enjoy.
An algorithm must be very good to nail this use case, but it doesn’t rise to the level of a playlisting algorithm, where a user will think you don’t know music nor them if a Coldplay song ends up in a Jose Gonzalez playlist.
Defining what your algorithms are meant to do and sticking closely to those use cases is vital for success.
- Man Guides The Machine
An algorithm must be built as a tool for curators and not simply a technology product. Therefore it must be tunable and adaptable. There is no such thing as ‘code lock’ on an algorithm.
In my experience, this is not the way many algorithms have been built. Machine learning–the ability for algorithms to improve based on usage–is a big topic right now for many technology companies, but I have yet to see one example of a music algorithm that gets smarter with time. Ensuring curators have input and a modicum of control of algorithms is extremely important.
- Playing Your Position
What makes managing a music algorithm so absurdly challenging is that no single person is qualified to manage it. You must posses a full understanding of music composition as well as its place in culture. You should have the knowledge of how a data scientist goes about their work. And you have to have a keen observation about how consumers behave in the system.
Without any leg of this stool, the product will end up hamstrung. It cannot be managed by one human, unless you have a consumer driven, musicologist, data scientist on staff (not bloody likely), therefore it requires a team of experts to tackle the problem.
Each will bring an expertise and needs to trust other members of the team. Success should be judged on results and data; not taste or perfect code.
- Match Curation to the Taste of Your Listeners
This one is easy to say and hard to pull off. Curation should closely mimic the usage in your system. While a marketing approach will influence who your listeners are, good old data and analytics should be fastidiously monitored and results fully understood by the team.
A curatorial staff must adapt their approach to what the listener is doing, and what brings more value to their experience. And above all, it’s about your listeners’ tastes. Not your own.
Tim Quirk, my former boss at Rhapsody and formerly Google’s global content programming head, authored the objective approach to editorial that we practiced heartily at the service. He recently posted a series of tweets that questioned the practice of tastemakers being the lead programmers at services and believes that curators should function more like ‘park rangers than gatekeepers.’ “Yay curation. But boo anyone who thinks he or she knows better than you what you should listen to,” Tim summed up.
- There Is No Finish Line
The algorithm will constantly need to adapt to the music, the customer usage and the technology. Likewise music trends change over time. After all, few could have predicted the amazing rise (and the fall) of EDM? As long as you have music, you must have a team who lives and dies to have the perfect music catalog, the algorithm and the curation to fully create a great music experience.
The first generation of streaming services focused closely on catalog and access. We’re nearing the end of this era, as pretty much everyone has the same catalog and the apps are very similar. The next phase will focus on the music experience of the services. Curation, whether lovingly hand-crafted by humans, or processing massive amounts of data crunched down by an algorithm, will be the battlefield all the services will vie on over the next couple years.
We can already see this battle taking form as ‘the humans’ vs. ‘the geeks.’ That’s a mistake. A company needs to seamlessly blend these talents together to build curation that listeners will enjoy and create true value.