What’s On

YouTube’s Cristos Goodrow on why views are a ‘bad’ metric

Highlights
  • 80% of views come from outside of US
  • Watchtime is up 60% y-o-y
  • Number of people watching per day is up 40% y-o-y since March 2014
  • On mobile, the average viewing session is up more than 50% y-o-y
  • Users who start at the YouTube homepage is up three times y-o-y

How often does it happen that you want to watch just that one video on YouTube, but the ones in the related videos tab are just too enticing to pull away? Well, this is what YouTube won an Emmy for in 2013 – for having a recommendation system that makes people want to spend more time on it. In a one-on-one with Digital Market Asia, Cristos Goodrow, Engineering Director at YouTube, explains the origins of the recommendation system and how YouTube plans to be the hub for content discovery.

About 80 per cent of views on YouTube come from outside of US. Asians are watching more hours of online video now than ever before. The number of hours Singaporeans spend on YouTube has doubled (110 per cent) in a year.

Surfacing the right content for the users is a key task for YouTube as a content discovery platform. When Mr Goodrow joined YouTube in 2011, the company had launched a machine learning system to improve the recommendations system. “The objective of this system was to maximise the number of video views. So it took more views as a part of deciding recommendations,” he explained.

Why focussing on views is wrong
However, there was a catch in the system which focussed on views, that the team realised after the launch. “When we launched this in 2011, it was great as the views went up a lot and we were very happy about that as views were the metric that we had focussed on then. But as we started to look at the sessions, we noticed that some sessions would have more views, but actually, seemed to be the worst experiences for the viewers,” Mr Goodrow explains.

Many people would remember the days of related video suggestions on YouTube which looked very enticing, but were not the videos they expected them to be – as the headline and the image would be very misleading. Giving an example, Mr Goodrow explains, “We were looking at a session in which a user would have searched for a boxing match. At the top of the search results, they found what looked like a video of the boxing match. The thumbnail was very compelling – but when you click it what you saw was not the actual video of the fight, but someone talking about the fight. The user would watch the video for a minute or 30 seconds, and realise that this video would not have any footage of the actual fight.”

These videos found their way to the top of the related videos list because of the system relying on the number of views as the metric. “The user clicked on four videos before he finally got to the actual video about the fight,” he adds.

This got the team re-evaluating the recommendation system as they realised the system actually caused this as it put more such obstacles between the viewers and what they were looking for – because each one of them got counted as a view and thus led to maximising the views.

“We were so focussed on the views. This system was like catching a rocket booster for your product and whatever direction you point it in, it was going to go much faster in that direction. So if you have not pointed it in the right direction, you are going to be way off very quickly and that is where we were headed,” Mr Goodrow opines.

If not views, then what?
While rethinking the metric for the recommendation system, the team landed on the notion of ‘time spent’ by users on watching a video. “People’s time is precious to them and if they are spending more of their time on YouTube, it is an indication that YouTube is more important to them and is providing more value to them,” Mr Goodrow asserts.

In March 2012, YouTube took the gamble of moving its metric from views to time spent – which Mr Goodrow remembers was a scary moment. “The whole company had applied its focus on this metric earlier and how we got them to go up. Now we were going to purposely drop them by a very significant amount in one day because we now wanted to focus on something else,” he adds. The views fell by almost 20 per cent in a day (as the graph below shows).

youtube-views-1024x546But ultimately the gamble paid off as YouTube witnessed the second largest increase in watchtime (the first one was when the company changed the Search system to Google). Most of the videos that had misleading thumbnails and titles also started to disappear.

On competition & rise in viewership
While YouTube has changed its focus on views as a metric, there has been a lot of talk in the past one year about Facebook’s video platform increasingly catching up. Interestingly though, Facebook’s statistics are focussed on the number of views it has – which stand at four billion daily video views.

YouTube, however, has not released the number of daily views since 2012, but its official figures do highlight a growth in watch time.

“The competition is surely growing, but remarkably this growing competition in the video space has led to a growth for YouTube. It has been a great thing for us – but we can’t relax, we must work as hard as we can,” said Mr Goodrow.

The number of hours people spend watching videos on YouTube is up 60 per cent year-over-year, the fastest growth for the company in two years. The number of people watching YouTube per day is up 40 per cent year-over-year since March 2014. The number of users coming to YouTube who start at the YouTube homepage is up more than three times year-over-year.

Once users are on YouTube, they are spending more time per session watching videos. On mobile, the average viewing session is now more than 40 minutes, up more than 50 per cent year-over-year. The number of hours people spent watching videos on mobile is up 100 per cent year-over-year.

“In the past year, the overall viewership on YouTube has not only increased, but it has accelerated. We are in a very competitive market and the competition has increased and so has the growth of our viewership,” he added.

Mr Goodrow attributes the growth to the recommendation system and also the revamping of the YouTube homepage to make it more compelling. “In the past year, the number of people who came to YouTube, had just opened it up or turned it on and without having to search clicked on a video on the homepage itself – that has increased almost three times. So I think that also plays a role in increasing the amount of viewership as we know that users who come to YouTube directly tend to watch a lot more,” he said.

YouTube of the Future
The video platform works on the premise that the platform has a 100 hours of content for every person, and Mr Goodrow feels it’s the job of his team to make sure that people find it. “Our aspiration is to be the Holy Grail of recommendation systems,” he says.

Mr Goodrow’s team is now looking to have YouTube suggest videos, that users have not shown an interest in before, but would like to watch. “The thing that keeps people from watching more YouTube is that they can’t imagine all the great videos that are there on YouTube – it does not occur to them that YouTube would also have videos on other things they are interested in,” he says.

He concludes, “Somehow we have to figure out a way to help viewers find videos that they didn’t even know would be on YouTube. We don’t know how to do that very well yet – but that is where our aspiration is – how can we make the platform so that the user does not have to put in any energy and yet somehow we show them some suggestions that they actually love.”

Shubhi Tandon

Shubhi Tandon is the Assistant Editor at Digital Market Asia. Fascinated by the evolving digital media industry, she has focussed on tracking developments in the Asia Pacific market since 2014.
Search