A BaFin-regulated German robo-advisory / wealth-management platform
−40%
P95 API latency
50k+
AUM users impacted
−60%
Redundant network payloads
0
Compliance regressions
Overview
The iOS app for a regulated German wealth platform was under-performing on API response times, directly affecting user confidence in a context where speed is a proxy for trust. I profiled, root-caused, and restructured the iOS networking layer — reducing P95 latency by 40% without changing a single backend endpoint.
The challenge
The platform served 50k+ users managing real money. Slow portfolio load times were generating negative App Store reviews and user support tickets. The backend team had already optimised their API responses — the bottleneck was entirely on the iOS networking layer: redundant serial requests, no request coalescing, and blocking main-thread JSON parsing.
The approach
Profiled the full request lifecycle using Instruments Network profiler and Charles Proxy to establish a baseline and identify the highest-impact serial request chains.
Refactored portfolio and transaction data fetching from sequential URLSession calls to parallel async/await request groups using Swift Concurrency's async let and TaskGroup.
Moved JSON decoding off the main thread with a dedicated decoding actor, eliminating a class of UI frame drops during data refresh.
Implemented response caching with ETags and conditional GET requests — reducing redundant full-payload responses by 60% on repeat visits.
Added MetricKit instrumentation to track P50, P90, and P95 API latency in production, enabling regression detection in future releases.
The outcome
P95 API latency dropped 40% within the first release cycle. Portfolio load times went from ~3.2s to ~1.9s at P95 on a standard 4G connection. Support tickets related to slow loading dropped in the following quarter. Zero compliance regressions — all changes were reviewed against the platform's BaFin data-handling requirements.
−40%
P95 API latency
50k+
AUM users impacted
−60%
Redundant network payloads
0
Compliance regressions
Tech stack