The possible silver lining is that companies will quickly realize the data is garbage (which is causing then losses) and stop using it.

Data that appears "good enough" under the purely economic lens can leave an awful lot of people unfairly out of service.

In a low margin business a single bad customer can easily cost more than what is earned from ten good customers. In that situation, an oracle that rejects eight good customers per rejected bad customer would already be good enough.

Depends on how quick is "quickly".

If anything, they seem to be doubling down on this. From my discussions on the subject with people "in marketing", they are really convinced that not only is the data "accurate enough", but that collecting and exploiting it is actually for the client's benefit.

I wonder how one would go about quantifying the usefulness of all this tracking, because I doubt it's cheap. The best I could get was "we're able to see changes in sales which correlate to marketing campaigns". I can totally buy that and it seems somewhat more advanced of an answer than the more common one of "If didn't work they wouldn't be doing it".

What I wonder though is how they would go about quantifying how much better a tracking-based campaign worked than a "traditional" one would have.

What if most of the marketing is useless in the first place? And most of the data are useless too.

How would they know?

That's a pretty strong assumption. Data that is garbage isn't necessarily going to cause losses. In fact, making the market more opaque (less efficient) will add to profit margins. If all companies are using the same data/algorithm, then even incorrect decisions form a focal point. If all the companies think someone is a bad risk, then that person will end up paying more regardless. Even if one company defects and accurately prices their risk, that company will just lower their rate a little and that customer still won't be getting the rate they would have if full market competition were taking place.

Does "risk avoidance" mean they have to act on the data?

Yeah that's probably the bigger problem!