A high-throughput insurance comparison platform serving millions of users was running on on-premise infrastructure with in-memory session state coupled directly to conversion rates. Every millisecond of retrieval latency on the primary portal had a measurable impact on quote completions. The architecture needed to move to cloud without taking the platform down or losing performance evidence in the process.
The Architectural Pivot
The migration was sequenced by impact-to-risk ratio, not by technical convenience. Redis distributed caching displaced in-memory session state first: the highest-impact, lowest-disruption change, delivering immediate, measurable performance gains that established the commercial case for the phases that followed. OAuth 2.0 and ADFS migration replaced on-premise Active Directory federation with a cloud-native identity layer. SoC-aligned service decomposition broke the monolith into independently deployable units. Circuit breaker integration patterns were introduced at service boundaries to isolate failure modes and prevent cascade degradation. Zero-downtime deployment pipelines made the transition invisible to users. NewRelic APM provided the distributed operational visibility that turned each phase’s performance claims into auditable evidence for non-technical stakeholders, proving the migration’s value in the language of business outcomes rather than infrastructure diagrams.
The Friction
The friction was not technical. It was the challenge of influencing enterprise architecture from the engineering trenches. Each proposed modernisation pattern (Redis caching, OAuth 2.0 federation, circuit breakers, SoC decomposition) required justification to a stakeholder environment conditioned on legacy infrastructure and risk-averse to architectural change. The path through that environment was execution sequenced for immediate, visible commercial impact. The Redis phase did not just solve a caching problem; it produced a performance result that earned the credibility to propose the OAuth 2.0 migration. Each phase proved the commercial case for the next, building an internal architecture mandate one delivered outcome at a time.
The Rulebook
- Sequence migrations by impact-to-risk ratio, not technical elegance. The first phase must produce a visible commercial result. Without that evidence, the later phases will not be authorised.
- Instrumentation is not operational overhead; it is the evidence layer that converts technical delivery into a business case. Deploy APM before you claim performance gains, not after.
- Influencing architecture from a mid-level engineering role requires the patience to build credibility through execution. The blueprint is a long-term argument; each delivered phase is a proof point in it.
