"You can tell the quality of a video by how long viewers watch it."
I see. This insight is neither new nor surprising, and anyone who hires highly qualified consultants to serve up such insights, well, yes. Attentive readers already suspect that there is surely something more to come now, and indeed, here it comes: "If I know how long a video has been watched, say 2 min. 30 sec. do I know how good it was?" Exactly not.
The different length of videos makes all measured values useless.
The problem is the different length of the videos. If the video in our example itself is only 2:30 min. long, then it was watched completely (we call that "complete"), it doesn't get any better than that, perfect, full score, the video was used optimally.
However, if the video itself is only two minutes long, then a measured viewing time of 02:30 min. is a data error, which can have different reasons. In any case, however, the value is useless because it makes no statement about the quality of the video.
Let's assume a third case, it is a movie-like video with a length of 45 minutes. If this video is watched for only 02:30 min. on average, the same measurement value means a real quality problem.
Percentage viewing time is also misleading.
Yes, that's clear, we say then and put the measured viewing time in relation to the running length of the video. If the video is viewed at 100% of its length (completes), the optimum is reached. Our 45-minute video, on the other hand, achieves a viewing time of about 5%.
So is percentage viewing time an appropriate KPI? To answer this question, I need to do two things:
- Review and argue professionally: Is it fair to compare the 5% for a 45 min. show with the 100% for a two-and-a-half min. show?
No. - Conduct a thought experiment, our "nerd test", which is often more important and meaningful for KPI practice in social systems (companies, agencies...) than the professional test.
Nerds always want to be the best. So what, the thought experiment asks, do I need to do to be the best at this KPI? In this case, it's about becoming the best video publisher to score 100% as often as possible. What I have to do as an overachiever to achieve this is quite simple: I publish only the shortest possible videos, preferably something like 2 seconds. That way I'm guaranteed to always score 100% with my videos. This will be a super report, I go to the boss / customer, he praises me, I've left all colleagues behind, career and home are secured - if everyone agrees that I only publish 2 seconds. Are they? Probably not.
So what to do to use data to objectively evaluate video quality?
If we were Mr. Zuckerberg, we would respond as we always do: "Our AI does that for us" - in effect, doing nothing.
But we are a companion. That's why it's not an imaginary AI that evaluates the quality of video content, but a very specific STIX. STIX is our stickiness index for videos. We developed STIX for a large media company based on the analysis of usage data from around 18,000 videos on all web and social platforms that can play out video content (not videoads).
The viewing time of a video can be predicted with high probability based on its length
The curves show the results of a regression analysis that examined the correlations between viewing time and running time. They show a perfect basis for a forecasting tool. The comparison of push media (Facebook, Twitter) with pull media (media library) is highly interesting. There are worlds between the willingness to use social and web, both in terms of the willingness to watch a video at all and in terms of the willingness to watch videos for longer.
The Stickiness Index STIX clearly shows whether a video was better or worse than an average video.
Bigger 1 = better. Smaller 1 = worse than an average video. This is how simple the STIX works and yet it is very reliable. The forecast quality / reliability of the index is 90% .
STIX can benchmark all videos and will be integrated into our content marketing dashboard.
