[masswerk]

Maybe, there's a biased view in ad business and some of the perceived benefits and effects are rather tautological? (This is why we have studies.)

You could also conclude from your remarks that there is a pervasive idea around ad teams that former generations (in the times of media analysis) were just delivering complete failures. However, this model had delivered for more than a century. How could this model perform with todays instruments and data?

[renewiltord]

The best thing about startups is that they test questions like that in a way that these studies can't. Because the participants have a very real and strong incentive to succeed they will perform that continuous search and hypothesis adjustment till they hit gold or die. If you truly have a Thiel hypothesis, you're going to get very rich.

In software, we call this "talk is cheap; show me the code", but of course here you don't need to show me the code. It's just that you're letting this golden opportunity go to waste. Up to you, I guess.

[masswerk]

The problem being only that these startups are still acting in the bubble of common beliefs of the field. So these are actually testing the beliefs, not their real-world effectiveness. (Also, at this stage, you have to comply and conform to the delivery networks right from the very beginning with little chance of competing with the big, established ones.)

Edit: Moreover, you had to compete with the paradigm of low effort, high interchangeability and big data. (Meaning: interchangeable code, interchangeable users, interchangeable professionals, interchangeable clients, and lean know-how stack as it's "all in the data". While this adheres to criteria of optimization, it doesn't necessarily mean that it represents an optimum of effectiveness, as well.)