How much of your cloud spend can you actually allocate cleanly?
Been doing FinOps at a mid-size SaaS shop for ~2 years. Multi-cloud, mostly AWS with a chunk of GCP and a growing K8S footprint.
The thing nobody tells you when you start: tagging is a losing battle.
We tried everything. Tag policies via SCPs. Auto-tagging Lambdas. Slack-bots that pinged engineers when resources spawned without an owner tag. Quarterly tag audits. Even a "tag czar" rotation across the platform team.
After all of that, our peak tag coverage was 73%. The rest was:
- Shared infra (NAT gateways, transit, observability stack) that genuinely doesn't belong to one team
- K8s nodes running 12 services from 4 teams
- Snowflake warehouses are shared across BI and product
- Legacy resources nobody owns, and nobody wants to claim
- Stuff spun up by automation that doesn't propagate parent tags
So roughly 27% of our cloud bill was "unallocated" every month. Which finance hated, and which we spent about 6 engineer-weeks per quarter manually fixing.
What actually changed things wasn't better tagging. It was acknowledging that some costs are inherently shared and should be split by usage signals, not labels. We started allocating:
- K8s costs by namespace, CPU/memory usage, weighted by request hours
- Shared infra by % of total egress per business unit
- Snowflake by query-level credit consumption per role
- DB shared instances by connection-hours per service
Coverage went from 73% tagged to 96% allocated. It took about three weeks once we stopped trying to fix the tags and accepted they'd never be perfect.
So the honest question for this sub:
- What % of your cloud spend can you confidently chargeback today?
- How much of that is from tags vs from usage-based splitting?
- Anyone actually hit 100% via tags alone? I genuinely want to know if it's possible at scale.
Curious where everyone else has landed on the tagging-vs-splitting tradeoff.