But many of our videos get predictable traffic like this Gates vs. Jobs rap battle at 33 million views (and climbing) or my favorite Mr. T vs. Mr. Rogers at 36 million.
He lost me at "predictable." If an audience's taste is so predictable, go get funding for a multi-million dollar blockbuster. I think all he has shown is that he is lucky: that anyone can bruteforce various kinds of cheap content cheaply, but his bruteforcing just happened to hit the right tone.
You can A/B test your videos, your copy, your thumbnail images.
Nevermind that A/B testing is hilariously inefficient (ANOVA anyone?) and virtually impossible for creative content (what, are you going to permute individual lines in a ninety page script?). Hollywood has statisticians. Focus groups and audience testing has been going on for decades. Web distribution expands your population size, but it also greatly increases the variables to control. If it were as easy as he says, then of course there would be truly data-driven content by now.
And video views now produced and managed by these 5 companies totals more than 4 BILLION per month. With a B.
But who's paying for that content?
Hollywood looks at YouTube and says, "Ok, what does content funded by Google's auction adwords ads look like? Oh, it looks like those link farm websites, which is precisely exactly content paid entirely by auctioned ads."
Don't we want something cooler than literally spam?
One could point to Hulu and say, perfect platform. But there is no original Hulu content. It is still largely funded by cable ads. And Hulu's profitability depends entirely on how an owner like Disney values the free broadcast licenses Disney gives Hulu. If Hulu had to pay for its shows, it would be deep in the hole.
But I digress. Data-driven content works for Wikipedia one-line banner ads. It isn't going to work for video longer than 30 seconds for a very, very long time.
1. Delivering content in movie or TV episode sized chunks is an artefact of inefficient distribution mechanisms. Smaller units of content are easier to test, easier to produce, and less risk if they tank.
2. Any script contains numerous decisions. You can test the main ones. It's viable to do this if you aren't making big bets on big chunks of contents.
Content dictates length, not the other way around...