You are still right, how is there user supposed to know if one tool is reputable and another isn't. Worse than that, one may be fine today but then it gets acquired and someone starts putting the pieces together and uses your historical data in good_tool to link you in bad_tool.
I don't have a good answer. As the other commenter mentioned, regulation may help.
Beyond that, some kind of standardized policy that can be checked and tested would be nice.
We can't track a user across different domains, or customers.
To be super clear- yes you can. You don't. That's very different. With full JS access on a site you have the ability to collect a lot of information. As another poster mentioned, it only takes 30 bits of entropy to identify all 3 billion internet users.Technically, though we can't, since we'd have to dedicate engineering time to making the changes necessary to do that kind of tracking, and we're not going to :P
However (and without revealing too much about specific IP), there are various mathematical techniques that can be used to discuss the privacy disclosing properties of queries against a set of data and ways in which you can force the data to be both fuzzy and self-destructing.
The reality is that customers (rightly) won't see a difference between analytics companies until one of them actually puts their money where their mouth is on privacy, and removes personal data from being an asset than can be sold or combined later (both by limiting number of queries to prevent reconstruction and with fuzzing applied to prevent over-targeting).
For those looking for some technical details: I work at a marketing company (we're early stage start-up, so I don't want to name names, as we're not really ready for attention), and we're currently in the process of developing a technology which fuses the research in to what questions you can ask of a database (and how many) before the privacy is depleted with non-fully-homomorphic encryption, which destabilizes after a period of time. This approach will let us build containers which homomorphically process data, and decay in a number of questions less than the number which would deplete the privacy of the data in the container. (For various reasons, operating in homomorphic space gives you increasingly fuzzy answers the more questions you ask, and thus you cross a probability bound on your answer being useless near when the privacy is depleted.)
While this isn't perfect, since you still need to trust that we're both a) not storing the base keys and b) did our math/implementation correctly, it's a substantial improvement over the trust asked for by marketing companies now. (Especially since contracts could require us to do those things fairly easily, thus exposing us to breach of contract if we failed to. Once those actions are properly completed, there's no taking them back, so our technology mainly protects the case of "We're the good guys now, but tomorrow is a different day".)
That still doesn't solve all the political dealings that happen but it would be a good strong start to cleaning up a lot.
Follow this up with certification by a main industry body and legal foundation for how companies in the sector should operate with data and we can finally make progress. This is no different than how finance companies or health companies go through audits and such with both process and data handling.
While digital advertising doesn't need to be as onerous and involved as that, the same workflow would help along with clear rules for collecting, analyzing, acting on, purging and securing data as part of consumer wishes and best practices. The next step would be to make sure transacting with a non-certified company carries risks and perhaps fines so that media buyers actually work with good vendors based on merit instead of inside deals.
There should also be technical certifications so that companies can actually prove they know what they're doing and dont just cram megabytes of crap on a page to serve a single ad. It's just too easy to whip up some crappy ad server software these days which ends up bloating sites and ruining trust and user experience.
Unfortunately there are issues with regional laws applying to a global business and the large amount of shady networks headquartered internationally but it's a good first step.
We get lumped in with super questionable networks in tools like Adblock and Ghostery and I have to explain to people when I tell them that I work in analytics that we aren't evil because the default assumption is more and more becoming that analytics and ads are all doing shady things. That sucks. I wish there was a way to combat unsanctioned tracking without diminishing the value that legitimate, and user controlled, tracking and attribution tools provide.
As long as people are willing to do anything to squeeze out more money or beat their competition, questionable techniques will find their way into the market, and that's bad for us all :(