Four people spent ninety minutes arguing about a cloud bill in a meeting I sat through last year. Spreadsheets. A downward arrow on slide six. A target number picked before anyone opened a console.
Not one person in the room had deployed a service in five years.
The engineer who knew about three idle replicas running in a region we served nobody from was two floors down, closing a ticket. Nobody asked him. Nobody asks him now.
Cost control turned into a spectator sport
Flexera's 2025 State of the Cloud Report found 84% of respondents name managing cloud spend as their top challenge. Budgets overrun by 17%. Spend is expected to grow 28% in the coming year. 87% now treat cost efficiency as their main measure of cloud success.
The response? 59% expanded a FinOps team.
Look at the shape of the fix. A bill nobody understands, so we hire a group to understand it... and we seat them one more step away from the code generating it.
I like FinOps teams. They build the visibility nobody had before. They win the arguments with finance. What they do not have is the memory of why a service writes 40GB of debug logs every day. Only the person who wrote the log line holds it.
The big rocks are gone
The State of FinOps 2026 survey covers 1,192 practitioners representing more than $83 billion in annual cloud spend. One of them puts the current problem plainly:
"We have hit the 'big rocks' of waste and now face a high volume of smaller opportunities..."
Read the quote again, because it changes what good cost work looks like.
The big rocks were centralised levers. Commit to reserved capacity. Rightsize the twenty largest instances. Move cold storage down a tier. One person with a dashboard and a mandate pulled those levers, and the savings were huge.
Those levers are pulled. What is left is gravel. Thousands of tiny items scattered across every team:
- A debug log shipping 40GB a day into an indexed store, switched on for an incident in 2023
- A nightly job scanning a whole table because an index was never added
- A staging cluster spun up for a launch, still running eighteen months later
- A bucket of build artefacts from 2021 with no lifecycle rule
- Three copies of the same dataset because two teams did not know about the third
No dashboard finds those. A dashboard shows you a number is high. It never tells you the number is pointless. Only someone holding the context knows the difference between $9,000 a month of necessary spend and $9,000 a month of leftovers.
You have hundreds of people holding the context. You pay them well. You never ask them.
Ask the specific question
"How do we save money" is not a question. It is a mood. People hear it as "layoffs are coming" and go quiet.
Here is what works instead.
Ask something small and specific. "Name one thing we pay for every week which delivers nothing to a customer. Twenty minutes, go." People answer in ten. They have been carrying the answer around for months.
Give it a slot in the calendar. Half a day, whole team, no other agenda. Waste-hunting squeezed into someone's evenings tells everyone what you think it is worth.
Publish the entire list. Every item, with a name next to it and a rough number. Not a filtered version for the leadership deck. The full list, visible to everyone who contributed.
Fund the fixes. An idea with no engineering time attached is a complaint. If item four saves $60,000 a year and takes six days of work, put the six days in the sprint. Out loud. In front of the team.
Report the number back. Three months later, say what it saved and who found it. Silence after a suggestion round teaches people never to bother again.
Run it again next quarter. The gravel regenerates. New services, new logs, new forgotten environments.
Pay the finders
The FinOps 2026 report flags something uncomfortable. Teams are pushing cost awareness earlier into engineering, and the incentive structures have not caught up. One practitioner asks how you give developers credit for work which prevents a cost from ever appearing.
Good question. An engineer who deletes $40,000 of annual waste gets nothing. The saving shows up as an absence. Meanwhile the engineer who shipped a feature nobody uses gets a mention in the all-hands.
So change what you count. My preference is a straight split: a fixed share of verified annual savings goes back to the team who found it, as budget they control. Tools, conference tickets, training, a decent laptop. Nothing motivates a second round of waste-hunting like the first round paying for something the team wanted.
If your finance function refuses a split, then at minimum put names and numbers in the same sentence in public. "Priya found the log pipeline. $71,000 a year." It costs nothing and buys years of goodwill.
Why most managers skip this
They are afraid of what comes back.
Ask a team where the money leaks and you get an answer about a system somebody senior chose. The duplicated data platform. The vendor contract signed at a golf day. The migration halfway done since 2022, paying for both sides.
Crowdsourcing savings works only if you are willing to hear a finding aimed upward. My own survey research found 99.5% of respondents have experienced one or more types of bad boss, and the most common variety is the one who asks for input and then bins it. Do it twice and your team stops answering. You have taught them their knowledge is decoration.
I write about the leadership side of this over at Step It Up HR, because it is the same muscle: asking, then acting on the answer.
Where this goes wrong
Leaderboards. Turn savings into a ranking and you get gaming. Someone deletes a test environment another team needed.
Cost police. The moment a savings drive becomes a stick, engineers start over-provisioning quietly to avoid the conversation.
Chasing gravel with an expensive process. Some savings cost more to capture than they return. A $300-a-year item is not worth a change request, two reviews and a rollback plan. Say so out loud, and drop it.
Cutting the load-bearing walls. Test environments, observability, backups. Every one of them looks like waste on a spreadsheet and behaves like insurance during an incident. Cheap systems which fall over are the most expensive systems you own.
Try this before your next cost review
Invite the three people who deploy most often. Ask them one question: what do we pay for weekly which delivers nothing? Write down every answer. Fund the top three within a fortnight. Then tell your board exactly where the money came from, and who found it.
Your cloud bill is not a finance problem. It is a listening problem with an invoice attached.
Who in your organisation already knows the answer... and when did you last ask them?