The Internet’s vast size means users can afford to be fickle.
With an estimated 4.66 billion web pages available to browse, users have a virtually unlimited menu to choose from. This means even the smallest obstacles – a clunky interface, or a detour to download a required plug-in – can send users running away from a site. Amazon discovered that just an additional 100 milliseconds of waiting led to a 1% decrease in sales from their users.
And when it comes to watching video online – an increasingly central part of what people do on the Internet – nothing deters users like buffering breaks1.
Just how much of a delay will a user put up with before moving on? We wanted to understand exactly how buffering impacts video watching behavior, so we put our data to work.
At Mux, we collect data on video performance and viewer experience – everything from video load times to playback errors to buffering interruptions. We have data on hundreds of millions of video watching sessions.2
Our data shows that just one buffering event decreases the amount of video watched by 39%. The more time a video spends buffering, the less video people watch, and even a small amount of buffering is a killer.
We began our analysis of buffering by looking at just how often buffering actually occurs. Do most videos load perfectly, or are buffering interruptions the norm? The chart below shows the percentage of videos punctuated by buffering events.
While just over half of the videos in our dataset were not interrupted, 49% paused for buffering at least once. Of videos that incurred buffering, about half were interrupted only once (24% of the total sample). Double-digit disruptions were relatively rare (4% of sample), and triple-digit disruptions even more so (1%).
When buffering does occur, how long does it last?
For each video in our dataset that was interrupted by at least some buffering, we found the total amount of time spent waiting for the video to start again. The following chart shows the percentage of these videos that fell into different categories of total time spent buffering.
According to our data, most videos buffer relatively quickly. Thirty-eight percent of the videos in our sample buffered for one second or less. Longer buffering durations were rare, with only 13% of videos buffering 15 seconds or more. Of course, some of these buffering interruptions would last longer if the user allowed it. Most video watchers won’t wait patiently as a video buffers for a long period of time.
So how much buffering will the average user put up with?
To answer this question, we pared our sample down to videos that lasted between one and five minutes, and grouped views into viewing sessions. When it comes to video-watching behavior, most publishers should focus on sessions – bouts of binge watching videos – and not just individual video views. YouTube doesn’t want you to watch just one video to the end; YouTube wants you to spend your whole evening on YouTube, and it's OK if you jump from one video to another.
The following chart shows the relationship between the number buffering events in a session and the median time spent watching videos.
The typical viewer who does not experience buffering has a total video watching session length of 214 seconds - just over 3.5 minutes. But just one buffering event triggers a huge drop in viewership. The first buffering event reduces median session length to 137 seconds, and median session length drops all the way to 111 seconds with four interruptions, a decline of 48%.
After four interruptions, sessions rise in duration with more buffering breaks. This makes sense; viewers who persevere through the first interruption or two are invested in finishing their video. There is also a relationship between session length and buffering: the more someone watches, the more opportunity there is for buffering to occur.
What happens when we simplify the data to compare session length for viewers who experienced any buffering to no buffering?
Again, just one buffering event triggers a huge drop in viewership. People who experienced any buffering had an average session length of 130 seconds – a 39% reduction.
Our analysis confirms that people really, really hate waiting for videos to buffer. According to our data, just one short buffering interruption leads to 39% less time spent watching video on a site, as viewers search for more fulfilling Internet diversions.
The good news is that video publishers can eliminate many cases of buffering. Publishers can adjust video bitrates, move to adaptive streaming formats (like HLS or DASH), change CDNs, use multiple CDNs, change video players, improve adaptive algorithms, fix bugs that lead to buffering, and more.
The data shows that buffering is something to take seriously, and any publisher who wants to succeed needs to care about performance. After all, the promise of the Internet is instant gratification, and the benchmark is broadcast TV. Why would people wait for a video to buffer?
This is the first of many explorations of video performance data. We at Mux will continue to study the effects of video performance on user engagement, including other performance metrics (video quality, startup time), new questions (how different is user behavior across different types of video?), and new approaches (active experiments rather than just analyzing past data).
We also offer custom data analyses. If you are a video publisher and would like to see data like this for your own video, get in touch.
Buffering (or "rebuffering") causes video to stall because a video player can’t download data fast enough to keep up with the rate of playback. When this happens, playback stops until enough video is downloaded to play again without stalling. Buffering can have a number of causes, like a viewer's connection cutting out, bad decisions in a video player, or CDN problems. ↩
This particular study measured the performance of tens of millions of streams across a number of video publishers. All of the streams were browser-based (desktop and mobile web), most were short-form content. ↩