This is a gross mischaracterization of what the linked paper says.

"It's extremely difficult to measure the impact" (the claim that the cited paper puts forward) has quite a different meaning from "no measurable impact." The paper is entirely about the difficulty of measuring the impact, and studiously avoids, as far as I can tell, any inference about what the impact actually was in these experiments. For example, Table I gives the mean of the control group sales and a standard deviation, but no mean of the treatment group sales, which you would need to do a statistical test of whether there was an impact; similarly, Table II reports standard deviations of the sales effect and ROI, but not means; Table III presents power calculations based on hypothetical ROIs and the real measured standard deviations, but gives no clue to the real ROIs were. Nowhere does the paper support your claim that there is no measurable impact.

In addition, the situation is very different for small companies that most people have never heard of. The article is citing studies done with major corporations with millions of customers and that are already well established in the collective consciousness. Measuring the effect of an advertising campaign taking you from 3.23 million customers to 3.231 million customers is indeed very difficult, particularly when you might fluctuate by tens of thousands of customers on a weekly basis. Measuring the effect of an AdWords campaign taking you from 200 customers to 250 customers is much easier.