The latency budget is a product decision
Reliability targets become useful when they are connected to what a customer is actually trying to get done.
A latency target is easy to write down and surprisingly hard to make useful. The difference is whether the number is attached to a customer moment or left floating in a dashboard.
Start with the action that needs to feel dependable. A checkout confirmation, a deploy preview, and a search result can all have different tolerance for delay. The budget should describe that experience before it describes the implementation.
From there, split the budget across the critical path. This gives every team a constraint they can reason about, and turns slowdowns into a shared product conversation instead of a late-stage escalation.