Problem
A B2B SaaS business had scaled to twelve product teams sharing a single, ageing deployment pipeline owned by a three-person infrastructure team. That team had become an unavoidable bottleneck: every new service, database, or environment went through them.
The measurable symptoms were stark:
- Mean time from merged commit to production was three days.
- Roughly 40% of infrastructure team time was spent on repetitive request handling.
- No two services were instrumented the same way, so incident triage relied on tribal knowledge.
Solution
Rather than adding capacity to the bottleneck, we removed it — turning the infrastructure team from a service desk into a platform team.
- Golden paths for the three service archetypes that covered 90% of real workloads, scaffolded through Backstage software templates.
- GitOps delivery with Argo CD, so the desired state of every environment lives in Git and drifts are corrected automatically.
- Self-service infrastructure via Crossplane compositions, letting teams request a database or queue through a Kubernetes manifest reviewed by policy rather than by a person.
- Opinionated observability baked into the service template: OpenTelemetry tracing, RED metrics, and a default dashboard and alert set from day one.
We deliberately shipped the platform to two pilot teams first and iterated for six weeks before a wider rollout.
Architecture
The platform is a thin, well-documented layer over standard CNCF components — no bespoke control plane to maintain.
apiVersion: platform.example.com/v1alpha1
kind: Service
metadata:
name: billing-api
spec:
archetype: http-api
runtime: node22
resources:
database: postgres-small
cache: redis-small
slo:
availability: 99.9
latencyP99Ms: 300
That single manifest expands, via Crossplane compositions, into the namespace, workload, database, network policy, dashboards, and alerts — all reconciled by Argo CD.
| Concern | Choice | Why |
|---|---|---|
| Delivery | Argo CD | Declarative, auditable, self-healing |
| Abstraction | Crossplane | Kubernetes-native API for infrastructure |
| Discovery | Backstage | One catalogue for services, docs, and templates |
| Telemetry | OpenTelemetry + Prometheus | Vendor-neutral, consistent across archetypes |
Outcomes
- Commit-to-production reduced from three days to under twenty minutes.
- Infrastructure request tickets fell by 85% within one quarter.
- All twelve teams migrated to the golden paths within five months.
- Every production service now emits consistent traces and metrics, cutting median incident triage time roughly in half.
- The platform team now spends the majority of its time on platform improvements rather than ticket handling.