Building an audience in the Periscope era
When I want to find a movie to watch, I usually search Netflix or iTunes or sometimes I might thumb through the pages on Amazon Prime.
But whatever I do, there will likely be a recommendation engine running in the background racking up clock-cycles as it sorts through a content catalogue trying to work out what choices to offer me.
The sequence of events is simple: I express a requirement (to watch a movie); the system attempts to fulfil requirement (selects a list of movies I might like); I choose one to watch (or not)
I like to summarise this approach as “User goes in search of something to watch”
It works great when the range of available content is fairly static. New movies get added occasionally and others drop off through expiring rights windows — but otherwise nothing much changes.
The system is perfectly happy just sitting there waiting for users to send it requests to find something good to watch.
However, the situation is somewhat different (and more problematic) with live user generated content on platforms such as Periscope and Meerkat.
Here, the fleeting nature of live content means that the audience for a clip consists of those users who happened to be looking for something like it the moment the stream started up.
Relying on the coincidence of interested users finding interesting content to build an audience for your live content is as good (or bad) as leaving it to chance and hope.
While live UGC is often casual and off the cuff, this spontaneity can make it feel truly genuine, highly personal and potentially valuable. But clearly, getting this content in front of the people who will appreciate it most is not something a traditional recommendation system or blind luck can really accomplish.
A different approach is required.
The “pull” method described earlier worked well in those circumstances because the range of content available was only lightly dependent on the time at which the request for recommendations was made.
With live content, the opposite is true: what’s available to watch is highly dependent on time. This suggests a “push” method might be a more valuable solution.
Consequently, I like to summarise this approach as “Content goes in search of an audience to watch it”
Whenever a hopeful user sparks up a new stream and begins to broadcast it to the world, it is important to start accumulating intelligence about it. The more detailed the better.
What’s being filmed?
Who’s in the frame?
What music’s in the background?
Where’s the action taking place?
What are people saying?
…you get the idea
This data can be used to match potential viewers with the subject matter based on viewing histories and social graphs. High quality matches should then trigger the despatch of notifications and the arrival of engaged viewers.
Facebook use a similar approach — if someone you follow starts streaming through their platform you receive an alert to go and watch.
However, this limits discoverability to those sources of content you already know you like. Once you’re there and watching, the experience is flawless but the value of any recommendation system that operates purely within this domain of known preferences is consequently constrained.
Better is the careful compilation of real-time stream metadata that will allow truly insightful and unexpected connections to be made between the familiar and the unknown.
In this way each and every new creator will be able to draw the ideal audience for each fresh piece of content they produce.