This data is used to create a feedback loop for ML models powering FB's ad auction and delivery. Marketing campaigns all have objectives that can be monitored via this data. FB will adapt auction dynamics to maximize the value for all parties, and this data helps FB properly set the auction bids or artificially inflate/deflate the price for individual users based on the predicted likelihood of a campaign objective being met.
It's used for campaign measurement - DoorDash reports an action in their app, and FB says "that person saw or clicked on a FB ad before taking that action. give us credit for it!" Facebook is effectively grading their own homework here and ignores other marketing campaigns that may have contributed to the action, but there is a deeper science around interpreting this measurement (marketing mix modeling) and third-parties who can help validate.
The data can also be used to create Facebook campaign audiences. "Show this promotion to customers who ordered from DoorDash 4 times last month." "Show this promotion to those customers' FB friends who do not have the app installed." "Show this promotion to users who have installed the app, but haven't placed an order yet."
So no money is being exchanged for the data, but both parties are able to maximize their partnership value from it.