Success storyIllustrative
How a 250-person infrastructure software company brought cloud spend back to plan
Cloud spend ran about 30% over plan for two quarters, and nobody owned the gap. The Finance and Product & Tech agents tied every dollar to a team. The owners did the rest.
3 min read
- Finance
- Product & Tech
- Operations
Illustrative story: a composite scenario built on sample data to show how the agents work. It is not a customer testimonial.
The challenge
Cloud was the company's second-largest cost after payroll, and over two quarters it drifted about 30% above plan. Every monthly review flagged it. Every review ended the same way. Finance saw one invoice, engineering saw a dozen accounts and hundreds of services, and nobody could say which team the extra spend belonged to.
So nobody owned it. Finance asked engineering to look into it, engineering asked for a breakdown, and the next invoice arrived before either happened.
What they set up
- The Product & Tech agent read cloud cost data and the engineering tracker, read-only, and mapped each account, cluster and service to a team and a product area, using the tags and ownership records that already existed.
- Where tags were missing, it listed the untagged spend separately instead of guessing. The platform team assigned owners to all of it within a week.
- The Finance agent compared spend by team with the plan every week and explained the variance in the weekly email to the CFO and CTO.
- The Operations agent tracked each fix as a commitment with an owner and a due date, and reminded the owner in Slack before the date slipped.
What changed
- cloud spend above plan, from the start of the quarter to its end
- 30% → 4%cloud spend above plan, from the start of the quarter to its end
- each with a named owner and a monthly budget for cloud spend
- 9 teamseach with a named owner and a monthly budget for cloud spend
- spend by team, instead of one invoice a month
- Weeklyspend by team, instead of one invoice a month
None of this needed a cost-cutting program. Each fix went through the team that owned the resource. Idle environments were shut down by the people who created them, database changes went through the normal change review, and the retry loop was patched in the next release. The agents stayed read-only and only drafted the requests. Every shutdown and every change was approved by its owner before it happened.
The lesson
The overspend was never a mystery to the engineers. It was an ownership problem. Once every dollar had a team's name next to it, most of the gap closed without anyone having to ask twice.