If a data aggregator can create a timeline of an individuals life, watching personality traits, social graphs, income, travel, routine, biometrics and health, stress and recreation, political affiliation, brand and taste preferences, savings, debt, credit, and social media influence traces, local, regional, and national cultural influences, and so on... that email archive is gold.
You can then create predictive models that let you target products, politics, music, media, and so on. It's not about spying on individuals, it's about manipulating populations. It's about rent extraction and wealth consolidation using tools of influence that negate consent. It augments abuses by law enforcement, corrupting the principles by which democratic governments are supposed to operate by hiding tyranny behind EULAs and TOS and private sector proxies.
Imagine a gpt-3 type model, except that instead of predicting text, it's designed to predict behaviors and psychological effects. That gives you a tool that's got a Darren Brown level of manipulation potential that you can scale. It's never going to be 100% accurate at the individual level, but you can target huge collections of individuals to modulate their lives through advertising and media sequencing.
We’ll this is not what the OP is proposing. Data removal after 3 months or a year seems too fast. I game on steam once every two years - do I have to buy all my games each time?
You can fake relevance if you want to sell the company without actually lying. Coincidentally there's a certain class of company that is in a permanent state of being sold and whose communication is under particular scrutiny wrt truthfulness. Seen from any other angle I fully agree, random user data value tends to be greatly overestimated.
