How are back-to-back systems designed?

Question

Software design

There are two DB (the conditional server and the client). Each database has its own user interface for working with data. The data is generated counter (both on the server and on the client). Between them there is a transport responsible for the exchange and control of data integrity. The structures of the databases themselves differ significantly, the exchange takes place with a strictly limited set of data.
During the implementation of the test case, I encountered the fact that any change (for example, on the client) requires changes both on the server and in the transport. A lot of time is wasted unproductively.
How to rationally design development stages in order to understand from which end to start?
At the moment I see this:
1. On the server side, I implement the data storage structure that is generated on the server (it appears relatively static, changes are unlikely).
2. Based on item 1, I implement the client, completely (database, interface).
3. I implement on the server what should come from the client.
4. I implement transport for the final data set.