Fourteen weeks. Workload by workload.
The work decomposed the single commit into three workload commits, each with its own consumption shape and commercial lever.
1. Compute and GKE commit
We modelled Compute consumption against seasonality, mapped GKE node groups to flexible CUDs, and built a commit shape that captured the floor without paying for peaks.
2. BigQuery slot reservation
We separated baseline analytics from data science exploration. Baseline went onto reserved slots at the discount the volume justified. Exploration stayed on-demand with a documented monthly cap.
3. Cloud Storage growth band
The media store growth was the easiest to forecast. We negotiated a growth band rather than a fixed commit, with re-rating at each anniversary against documented storage volumes.
4. Migration credits, defensibly sized
We reduced the migration credit pool to the size of the actual roadmap and added a contractual right to recover unused credits at term-end against future commits.
Lesson
A single committed-spend number is a negotiation convenience for the vendor. Splitting the commit by workload turns three different leverage profiles into three different conversations.
