How a Canadian public sector institution cut Kubernetes spend by 45%

A Canadian public sector institution cut its Kubernetes spend by 45% with Randoli. The reduction didn’t come from one change. It came from months of evidence-gathering: cost reports broken down by workload and team, rightsizing recommendations, and historical trends showing how consumption actually moved over time.

What made it possible was having cost and APM in the same place. The platform team could weigh spend and application performance together, instead of optimizing for one at the expense of the other.

The problem wasn’t finding savings

The platform team drove the initiative, but it couldn’t rightsize anything on its own. Every change needed the engineers who own the workloads to agree to it. Developers are generally reluctant to accept changes to their workloads, and with good reason. If a rightsizing change saves money but hurts performance, they fear being held responsible for it. A recommendation that rests only on cost numbers is easy to say no to.

Cost and APM, side by side

Randoli put each proposed change next to the production telemetry for the workload it touched. For every candidate, the platform team and the engineers could look at:

• what the workload costs today, by workload and team
• how its consumption has moved over time
• how it performs under real load, through APM

That let them answer “will this save money without hurting reliability” with data instead of a guess, and gave the engineers evidence that the change was safe.

A data-driven path to yes

The conversation changed. Instead of a platform team asking developers to take a risk on faith, the engineers were involved in the process from the start, working from evidence that balanced cost against performance. They accepted the proposed changes because the data showed performance would hold, and the risk of being blamed for a bad change went away.

Why the savings held

Every change was justified on both spend and reliability. That turned the work from simple cost cutting into engineering-led optimization, and it is why the savings held instead of quietly reversing the next time a team hit a performance problem.