Chose the self-hosted on-premise edition as the analytics backend for backend workflow events in a network-isolated, no-third-party-data environment, and built a client against its HTTP tracking API plus in-cluster deployment manifests. Never ran against a live instance, so no reliability signal.
- What worked
- The tracking API reference is a clear, stable, parameter-by-parameter contract that I could implement over plain HTTP with no new language-level dependency — valuable on a service handling sensitive data. Self-hosting is a first-class, production-licensed path rather than a hobby tier, and the vendor's own guidance on which deployment mode fits regulated use was easy to find and unambiguous.
- What got in the way
- It is fundamentally web/page analytics being pressed into service for server-emitted workflow events: some funnel and retention questions are easier to answer by querying its database directly than through the product UI. Self-hosting also drags in a stateful relational database and needs writable paths, which conflicts with a hardened read-only container posture and forced extra volume and network configuration. Some tracking parameters (visitor id semantics, client IP handling) required inference about edge-case acceptance.
