In many senses, the question seems too broad to me and it's going to be hard to give you an accurate answer to the question What each of these backends really does?. However, we can read the diagram and guess.
A key concept here is separation of concerns. A secondary one could be distributed computing. For brevity, let's focus on the first1.
What you see is a complex backend comprised by several processes each of which is specialized on a specific concern (job, task, whatever).
Think in any business which nature involves treating the data through several stages each of which happens at different times, with different requirements, with different needs of performance, reliability, consistency, etc.
Instead of a monolithic "backend" doing all the above mentioned, the whole business is dissected so that each part is performed in different stages by dedicated, different and interconnected processes.
There could be many motivations to go with this architecture, from economical to technical. The one that caused our diagram we can not say, but we can guess (again).
Looking at the diagram the presence of a queue might suggest a need for high performance, the ETL suggest complexity on the data intake and the workflow suggest business rules to be performed in a very specific way, time and order. The API suggests that there could be clients monitoring and managing the data. The Confidential Data Backend seems a front-controller to allow clients to send new data. Altogether make me thing in a system where there could be potentially many clients sending data and monitoring its state concurrently.
1: Looking at the diagram we cannot say whether the system is deployed as distributed or not. But the separation of concerns is quite evident.