Governance Before Optimization

The instinct with a FinOps engagement is to start with the dashboard. Pull the billing export, build the chart, show finance where the money goes. That instinct is backwards. On a multi-account estate spanning GCP, AWS and Azure, the first project isn't visibility. It's vocabulary.

At a global staffing company I worked with, the billing data was accurate and completely useless. Every team named its resources differently: some by project code, some by environment, some by whichever engineer happened to provision the instance that week. Cost allocation software can ingest any tagging convention. It cannot ingest twelve of them at once and produce a number finance can act on.

So before a line of automation shipped, the work was organizational: a single tagging taxonomy, agreed on by twelve platform teams who each had a reasonable argument for keeping their own. That negotiation took longer than building the Python jobs that eventually read the tags. It also mattered more. Automation applied to inconsistent metadata just produces inconsistent reports faster.

Once the taxonomy held, the rest followed a familiar shape: scheduled jobs against the CloudHealth and native billing APIs, auto-tagging for resources that slipped through provisioning without one, and a chargeback report finance could read without a translator. The waste we found, about 30% of spend, was real, but it was findable only because every dollar now traced back to a team, an environment and a purpose.

The lesson generalizes past FinOps. Any system that reports on infrastructure inherits the infrastructure's naming discipline. Skip the governance step because it's slower and less visible than shipping a dashboard, and you'll ship a dashboard that lies convincingly.