In 2012 I created a killer prototype that demonstrated that you could accurately reconstruct most people's flight history at scale from social media and/or ad data. Probably the first of its kind. This has been possible for a long time.
A quick sketch of how it worked:
We filtered out all spatiotemporal edges in the entity graph with an implied speed of <300 kilometers per hour or <200 kilometers distance, IIRC. This was the proxy for "was on a plane". It also implicitly provided the origin and destination.
These edges can be correlated with both public flight data and maintenance IoT data from jet engines to put entities on a specific flight. People overlook the extent to which innocuous industrial IoT data can be used as a proxy for relationships in unrelated domains.
In rare cases, there was more than one plausible commercial flight. Because we had their flight history, we assumed in these cases that it was the primary airline they had used in the past, either generally or for that specific origin and destination. This almost always resolved perfectly.
This was impressively effective and it didn't require first-party data from airlines or particularly sophisticated analytics. Space and time are the primary keys of reality.