What do music streaming apps need to have?

21 October 2014

The music streaming space is both an old one and a messy one. Products have come and gone, and many have shown users a new way of interacting with and listening to the music they like.

The ups and downs of different products suggest that the space is still actively evolving. However, I believe there are a few core user needs that any music streaming product will need to address in order to be successful, and no one product currently has a home run in all of these needs.

Have what I want: Content

This is the most basic need that must be fulfilled: the music service must have songs that users want. It must offer the most current pop songs as soon as they come out, and it must have a wide library going back in time and spanning all genres. If the product is intended for an international audience (and, really, even if it's not), it also has to provide a thorough library of international content.

Getting access to all of these songs is understandably a huge barrier to entry, and from anecdotal evidence, the legal issues surrounding all of this are formidable. However, based on the example of services like Pandora or 8tracks, it seems like some of these legal issues can be skirted (or lessened) if the service doesn't give users full choice over playing the songs from the library.

In these situations, the user doesn't know the full breadth of the service's library, since the service is, in some sense, controlling what she can listen to. Not having everything a user might want is thus not a dealbreaker, but in these situations the service has to offer some other strong benefit, like strong recommendations (discussed below).

As another side note, there is also a growing content base outside of music publishers, where users create their own tracks (e.g. covers, remixes, etc.). Getting good coverage of this kind of content can be tricky, because there are very likely network effects that encourage users to all contribute their creations to one particular service (YouTube? Soundcloud?). However, Spotify, which has great coverage over published, licensed songs, is addressing this additional kind of content with instructions for unsigned artists to get their music onto Spotify. This isn't quite as easy as uploading a YouTube video, but opens the door to a lot more content.

Show me what I want: Curation

Finding individual tracks and putting them together into a custom playlist is now too time-consuming for many users. Today, music streaming services need to offer some kind of curation, where users can choose a genre, base track, or "theme" and receive a playlist of similar songs in return.

The appeal of this kind of feature can be seen in radio channels. Different channels tend to focus on specific genres (Top 40s, Country, Classical), and users can tune in to the channels that appeal to them. One benefit of listening to a radio channel is that you'll hear new songs from that genre, and somebody else has taken the effort to find those new songs or artists for you.

In the music streaming world, radio channels can either be replicated (Spotify has a "radio" feature in its main navigation), or can be replaced with playlists. In fact, Spotify now has dozens of "Browse" playlists, under broad genre, mood (e.g. "Melancholic"), or situational (e.g. "Focus") groupings.

This idea, of course, is not new. Songza focuses its whole product around the idea of contextual playlists, where the suggested playlists change depending on time of day. Songza can compete with Spotify in terms of number of playlist options, but there is unfortunately no control at the individual song level, unlike with Spotify.

This kind of curation does not have to be done by the music provider. 8tracks is a great example of when users can generate curated content, and on top of that, the best playlists created by users can get voted up via "likes". Given user creativity, 8tracks has thousands of playlists, and they are loosely organized by a tagging system. One powerful thing that 8tracks has going in its favor is that it allows users to add user-generated tracks to the playlists, thus expanding the overall library significantly.

Tell me what I want: Recommendation

A step beyond curated playlists would be a strong recommendation system. Pandora is an early giant in this space, and they base a lot of their value on correctly predicting what kinds of songs users will like. An inherent difficulty of creating a strong recommendation engine is that a music stremaing service needs a large amount of data showing revealed user preference in order to beef up its recommendations. This is probably why Pandora started off by manually cataloguing all its songs with the "Music Genome Project".

One inherent tension that a recommendation system has to face is how to incorporate an "exploratory" element into its output. For example, the recommendation algorithm might suggest that there's some similarity between country music fans and jazz fans. How should it incorporate jazz music into a country fan's recommendations, if at all?

The flip side of this tension is that you have to decide what to do when the recommendation system "runs out" of strong recommendations because the user has expressed too narrow a range of liked songs. If you continually listen to a single Pandora station and tell it whether you liked or disliked each song, then the algorithm seems to adjust by eventually playing you a much more limited set of songs - only those you liked.

Recommendation is a tricky thing to get right, but much like the radio, it helps reduce the burden of usage for the user. Once users can trust that the music provider "knows" their taste, they don't need to spend time digging through albums and tracks to create their own playlists. Relying more heavily on recommendation and curation also means that the music provider can get away with slightly less breadth in its music collection.

Give me what I want when I want it: Convenience

The final major pillar that a music streaming service has to provide is convenience. Users want to listen to "their" music in many different settings and through different kinds of hardware.

At the minimum, this means that today, music streaming services should run well on both web and mobile. It also means that there are certain audio file quality thresholds the files have to meet, although for most users this will probably be a relatively low bar.

This principle also means that users want to be able to listen to music when they're offline. iTunes solved this problem by letting users buy ownership of the songs they wanted. But now, Spotify has worked out a clever "offline mode" feature, which lets you easily mark which songs and playlists you want to access when offline.

Currently, Pandora appears to offer some integrations that other music services do not - some cars and smart TVs, for example, come with a built-in Pandora app. However, Spotify is quickly winning the convenience war outside of these rather rare situations, and I predict it's only a matter of time before Spotify replaces Pandora in these built-in integrations.

What I don't really need

Given the principles above, I also tried to think of some features or principles that are actually not important for users, at least via observation of my own usage and that of my friends.

The first one I came up with is a social sharing element. I am not convinced that listening to music is always a social activity. News about a new artist or album can spread quickly, and there's certainly virality in what becomes a hit. However, I don't think that letting a user's friends know what he's listening to solves an inherent need for anybody. In the most likely scenario, I would bet that most users are indifferent to what exactly their friends are listening to. In the worst scenario, publicly showing what a user is listening to can even be embarrassing.

Another feature that is no longer important for users is ownership over the music. Spotify has done a lot to revolutionize this point - as a Spotify user, I have chosen to give up ownership of songs in exchange for obtaining access to any song I want, when I want it. There is some evidence that digital music sales are on the decline, and given the plethora of streaming options, I am not surprised.

A third element which I think is not impactful with music is a commenting or thought-sharing feature. SoundCloud is notable for letting users comment on specific times in tracks, but I think the additional benefit of this user-generated content is low. Music can be a fairly personal preference, the experience for each song is generally fairly short, and there is not really a demand for discussion for most songs. What is potentially more helpful is the knowledge of what's popular or what's been "liked" often, which is a feature that other services, like Spotify and 8tracks, have.


Overall, music streaming has evolved shockingly quickly over the last couple of decades. The above was an attempt to think about what exactly drives success for a music streaming product, and what does not. However, I am optimistic about even more advancements and evolutions ahead for this industry, and I can't wait to see how the above principles change (or don't change!) over time.


comments powered by Disqus