There are ways to estimate advertising ROI, but so much of what ads are attempting to do is inherently unmeasurable without a tightly controlled study, particularly given the highly chaotic nature of the economy and people's behavior.
Fundamentally, there are two "sides" of advertising ROI you can try to measure: the individual, and the aggregate. They each have serious challenges.
At the individual level: sure, some people see an ad, say "hey, I want that!" and go buy it. But that's definitely nowhere close to the majority of purchases—even online—and it's very unlikely to be a very high percentage of ad-influenced purchases. Much more often, the ad gets the product into your brain, and then later, when you're shopping, you see it and say, "hm, maybe I should get it," never consciously realizing that the ad caused or contributed to that. Thus, any kind of consumer survey will be unable to capture more than a tiny fraction of ad-influenced purchasing.
At the aggregate level: what you're trying to measure here is simply your sales numbers over time. And except for the very largest of companies, sales numbers will be pretty noisy. That means it's very hard to tell for sure whether this ad or that one worked, or if any of them had any direct effect on sales at all. (That's particularly true for the subtly-influenced purchases I described above; those might happen months after an ad campaign starts.)
Is it totally impossible to measure ROI? No, it's not. But it takes a lot of work, and has a fairly high degree of inherent uncertainty. The old saw "half of all advertising dollars are wasted—we just don't know which half" is very relevant here.