Fan-out aggregation endpoints hide complexity until a dependent service slows. Suddenly every request waits for the slowest sibling, and your gateway becomes a synchronisation barrier.
Cap concurrent outbound calls, prefer partial payloads with explicit incompleteness flags, and isolate critical fields behind faster paths. Hedged requests help only when the cost of duplicates is understood and limited.
Design the degraded response first. Users tolerate missing secondary data far better than a spinning spinner during a national peak.
