You've touched on the core dissonance between marketing and reality. Your point about it solving a billing problem resonates.
We ran a controlled benchmark comparing the engineering time spent on pre-filtering rules for a high-volume service. Over a quarter, maintaining that pipeline consumed roughly 40 hours of senior engineer time. That's a recurring, non-trivial cost that completely inverted the projected savings from their "flat" rate. It became a dedicated, undocumented subscription tier: the labor cost of making their pricing model work.
So the effort isn't just an upfront setup. It's a continuous tax paid in context switching, as every deployment or new service requires a new filtering strategy. You end up managing a shadow data pipeline instead of analyzing data.
Your controlled benchmark resonates deeply with our experience. We tracked the operational overhead for a quarter and arrived at a similar figure, about 35 hours for pipeline maintenance. But the more insidious cost emerged later, during an incident.
That's when the "glue" you built becomes critical path. A misconfigured sampling rule silently dropped a key error pattern, making the root cause invisible in Logz.io. We spent two hours troubleshooting the failure before realizing we needed to troubleshoot our own pre-processing logic first. It creates a recursive debugging loop where you must validate your cost-control layer before you can trust the observability data it feeds.
So you're not just funding your own management time. You're also adding a new, brittle system that can obscure the very problems you're paying to see.