In theory a "user" is more valuable the more you know about them because the extra info lets you better target them for higher response rates. Better privacy means you know less about the user, meaning you are less likely to bid high for that user's eyeballs, meaning the user is worth less ad spend.
The actual story is a little more nuanced: when you target an ad, you are usually bidding for users in an auction. You're generally willing to pay more for more specific filters because you're more confident of a higher response rate.
In ecosystems with a lot of data on every user, specific filters actually work, so you end up with higher bids on fewer users, which should drive the average spend per user up without necessarily driving total spend up. With less data, you have to use less specific filters, so you end up with lower bids on more users, which drives average spend per user down without necessarily driving total spend down.
There's also a countervailing force, depending on the market. First an example that supports what everyone is talking about:
* User X loves cola
Pepsi and Coke both bid more to serve their cola subscription services or whatever. Ad price goes up. Now add more data:
* User is a Coca Cola fanatic.
Even though this is the same user, the ad price goes down, because Pepsi doesn't want to bid.
Now add another Pepsi fanatic for symmetry, and you can see how "more data about users for better targetting" can in some cases reduce auction prices (while of course still increasing advertiser value.)
But if users were trending towards either side then the number of users in the middle would presumably be diminishing, driving up prices to target them even more.
Your scenario just seems to single out one particular example where more information can reduce spending on one group, while increasing it on another.
I think this is only on fake-scenario. In a real DMP, the user would be "loves cola","loves coca cola","loves fizzy drinks","drinks sugared drinks", and about 30 other related segments, and one of those would definitely sneak by.
Humans create audience segments, and it's in the interest of dataset producers to have each user be a part of the maximum number of members they can be to maximize monetisation.
Eventually there will be machine learning models involved (maybe there are already? I don't know ads.) Those models might move closer to "perfect targeting" where advertisers could compete less over certain users.
And no matter the incentives of the data providers, "increase customer value" is probably always a Nash equilibrium. If better targeting leads to less revenue, eventually you'll probably do it anyway.
The actual story is a little more nuanced: when you target an ad, you are usually bidding for users in an auction. You're generally willing to pay more for more specific filters because you're more confident of a higher response rate.
In ecosystems with a lot of data on every user, specific filters actually work, so you end up with higher bids on fewer users, which should drive the average spend per user up without necessarily driving total spend up. With less data, you have to use less specific filters, so you end up with lower bids on more users, which drives average spend per user down without necessarily driving total spend down.