Lots of things,

Writ large you can look at data across all people and spot trends (upward or downward), if you have a bunch of die hard customers that suddenly change behavior that's a warning sign. or you can see the effect of larger ad campaigns. If Target could figure out you might be pregnant from buying some hand lotion I have no doubt a good analyst with this data set across millions of users could extract a wealth of information. It probably also helps them judge revenue, if they know they can expect 2.16 visits over the next 6 weeks from 10,000 people and about how much they're going to spend you can extrapolate your future earnings with more precision than ever before.

At the micro level it's used to manipulate behavior, knowing your favorite items to give you discounts or to ping you when you might be most vulnerable. Maybe you have data that shows a customer spending more on particular days amid fairly random purchases, you might be able to construe those days are after payday and the customer is more confident purchasing larger ticket items. you might then target coupons for the days they might be getting paid.

Or maybe you see a trend where the person is clearly buying for 2, or 3 people and that changes. or suddenly they're getting happy meals implying they may have a kid now and you can send ads for the latest Disney toy.

Gather enough boring data over long enough and you can learn a great deal about someone.