A single latency number hides most of the information needed to understand where time is actually being spent.
Split the request
Name resolution, connection establishment, encryption, network transit and application processing all contribute to the final experience.
Treating them as a single measurement makes it difficult to tell whether a change improved the system or merely moved delay from one stage to another.
The useful question is rarely “is latency low?” It is “where did the time go?”
Use a budget
A latency budget gives each stage a reasonable portion of the overall response time. When behaviour changes, the search space becomes much smaller.
This becomes especially valuable when the network path can change without any application configuration changing at all.