YouTube, Twitch, & Live-Video Monetisation

This is a snippet of the transcript, sign up to read more.

Maybe we could just jump into those things, however you think is most logical to go through them.

I think Live was very difficult to fit into that scheme because it doesn’t have quite as much of the infra and, therefore, the opportunity, to serve as effective a way as VOD. At the same time, I think the search and discovery engine of YouTube, which is many layers of algorithms, also has a lot of training and, therefore, prediction power for how VODs would perform, but not so much on live video. By the nature of that data imbalance, it just never knew how to properly predict the value of live streaming. The deck was basically stacked against live streaming being a very successful thing. That is the background.

This is a snippet of the transcript, sign up to read more.

I’m loving hearing the journey. It was like a startup within the company and then what happened?

There are a lot of experiments that we run on the search and discovery side of things. A lot of those are because there are gaps in the search and discovery apparatus. What I mean by that is, we assume that the search and discovery engine works well because it can predict and, therefore, optimize for maximum watch time. There are things, like live, where it doesn’t know how to predict those things accurately and so, therefore, it just prefers VOD. In those instances, we can give it a 1,000 times boost, to the ranking of how live content appears. We would apply those experiments to things that we feel have a tremendous user value. Those would be the hypotheses and we would go and experiment for them.

This is a snippet of the transcript, sign up to read more.

Can we zoom out a little bit. This is an example where the data tells you one thing but there is an intuition that there should be an experiment that explores whether the data actually isn’t correct. Where did that come from, in general? Was it you? Was it someone else? Where did that hypothesis come from and the will to do this experiment? More generally, do you feel, with YouTube, there is a healthy balance of being data driven and also using human intuition?

I believe there is actually room to try to spearhead something that is counter-intuitive or that the data doesn’t really cover today. I think there is an allowance for small teams to come up with this unlikely hypotheses and run small experiments to validate it. I think YouTube was very good at this, when I was there, especially on the search and discovery side, where eking out a point whatever percentage gain, is actually massive, at YouTube’s scale. YouTube did take a very engineering type approach to this, which is the whole concept of explore/exploit. If the engine, right now, is working fairly well and is able to exploit the signals that we do have, that’s great but, once in a while, you do need to go and explore. Yes, that could land in some sub-optimal results but those learnings, in and of themselves, are very valuable.

This is a snippet of the transcript, sign up to read more.