Start free trial

By Yair Knijn · August 5, 2025

The engineer who understood the bill just left. Now nobody can.

A year ago your engineering director signed off on GCP committed-use discounts because one staff engineer walked them through the math at a whiteboard, and the math held. The mistake nobody caught was assuming that math lived somewhere other than that engineer's head. It didn't. It lived in a handful of saved BigQuery queries no teammate could read, and in a private map of which project belonged to which product. Two weeks ago the engineer gave notice.

So the director is looking at a renewal date and a commitment they can't defend, because the one person who could reconstruct the reasoning is counting down a final sprint. A single point of financial failure, built quietly by never forcing the question into the open.

The cost knowledge that never made it into a doc

Carried cost knowledge works right up to the day its carrier leaves. The commitment renews on autopilot, the dashboards stay green, and the load-bearing human stays invisible until they walk out. The FinOps Foundation's practitioner surveys keep landing on the same shape: a wide spread of capabilities, flat team size, and one person owning rate optimization, allocation, anomaly triage, and forecasting at once. When that person is also the only one who reads the bill, headcount risk and knowledge risk are one risk.

Commitment rationale you can no longer reconstruct

A committed-use discount is a bet on a forecast. Whoever placed it knew the baseline they assumed, the growth they priced in, the workloads they excluded for a coming re-platform, and the coverage ratio they wanted. The GCP console preserves none of that. It shows a one-year commitment at some hourly spend and stays silent on why that figure and not one twenty percent lower.

That leaves two bad doors at renewal. Renew blind and risk locking in capacity you are weeks from deleting. Let it lapse and pay full On-Demand while re-deriving the analysis. Either door bills you for the missing rationale, and a decision you can't reconstruct is one you re-learn at retail.

Account-to-product mapping as tribal knowledge

The map leaves too: which project, subscription, or linked account feeds which product line and P&L. In plenty of orgs that mapping exists nowhere but one person's recall, propped up by account names like proj-prod-2 that meant something in 2022 and nothing now. Take that person out and a quarter of the bill goes unattributable. "Unallocated spend" stops being a chargeback annoyance and becomes a number you can't explain to finance.

FOCUS, the FinOps Open Cost and Usage Specification, was built to close this gap, with a vendor-neutral billing schema so the mapping lives in standardized columns instead of one skull. A standard only helps once the allocation rules are written against it, though. The schema has no idea proj-prod-2 is the billing service until a human encodes that fact.

Encoding allocation logic as policy, not memory

Treat allocation as code, not as a person. Every rule the departing engineer applied by hand should exist as something a successor can read, run, and argue with:

The test is blunt. If your most knowledgeable cost person vanished tonight, could a colleague reconstruct the next renewal decision from artifacts alone? If not, you don't have FinOps. You have one employee and a lot of hope.

A FinOps runbook that survives a resignation

Cloud Horizons keeps the cost context in the workspace instead of in a person. Allocation rules, the account-to-product map, and the reasoning behind every commitment get encoded where the whole team can read them, so an offboarding becomes a handover, not a loss. If your renewal math leaves the building at five o'clock, close that gap first. See how the allocation model holds up in our approach to FinOps.