Collecting and cleaning data at sizes most teams never touch, then keeping it correct on a schedule nobody has to babysit.
Data problems stop being interesting right around the point they start being expensive. A pipeline that works at ten thousand records and falls over at ten million is not a smaller version of one that scales. The difference shows up as a query that used to be instant.
The work is usually three things: make the query fast at the size you actually have, make the daily job survive its own overlaps and failures, remove the human who currently notices when either goes wrong.
None of it requires exotic technology. It requires someone who has run it in production and knows which of the obvious approaches will not hold.
Datasets that have outgrown what a team can hold by hand. The pipelines move millions of records a day without anyone watching them. They report their own failures.