Commitment purchases based on real usage

Reserved instances and savings plans get bought on an estimate and then paid for regardless

The problem

Committing to a year or three of usage is where the largest cloud discounts are, and it is also where the largest waste is. The commitment is a bet on future usage, and it is usually placed with a poor view of current usage. Somebody looks at last quarter's bill, adds a margin for growth, and buys.

The usual answer

Then the bet goes wrong in one of two directions. Over-commit and the organisation pays for capacity it does not use, with no way out until the term ends. Under-commit and the excess runs on demand at full price, which is the outcome the commitment was meant to avoid. Either way it is discovered late, because measuring commitment utilisation across three providers means understanding three different discount models and their overlaps. The common response is to buy conservatively, which leaves the discount on the table, or to stop reviewing commitments altogether and renew what was bought last time.

How we approach it

Nimbus Stream's advisory pulls current data from the provider portals and recommends purchase commitments and resource changes against what the workloads are actually doing. On-demand, reserved instance and savings plan usage sit in the same reporting, so it is clear what is covered, what is running uncovered at full price, and what has been bought and is going unused.

What changes

Commitments get sized to real usage instead of an estimate with padding on top. Renewal becomes a decision informed by the last term's utilisation, and the resource recommendations catch the workloads that should be resized before they are committed to for three years.