9.2 Cost visibility, tagging and budgets
Describes cloud cost engineering as of August 2026; no pricing figures, which go stale
What this is and why it exists
A bill you cannot explain is a bill you cannot reduce. Before any optimisation is possible, somebody has to be able to say which team, product or feature caused each line — and that is a property you build in advance, not a report you request afterwards. This lesson is the small amount of structure that makes a cloud bill answerable, and the method for the morning it has grown by a third.
The vocabulary
- Tag (or label) — a key-and-value pair attached to a resource, carried through to the bill.
- Tagging standard — the agreed set of tags every resource must have, and their permitted values.
- Cost allocation — dividing the bill along those tags.
- Showback — telling a team what it spends; chargeback is billing it internally.
- Untagged (or unallocated) — the share of the bill nothing accounts for; the number that matters most.
- Budget — a threshold with a notification attached.
- Anomaly detection — alerting on an unusual change in spend rather than on an absolute figure.
- Amortised cost — a commitment's cost spread across the period it covers, rather than landing in one month.
The mental model
Tagging is not administration, it is the schema of your bill. Decide four or five tags and no more — owner, environment, product or service, and cost centre are enough for almost everybody — fix the permitted values, and write them down. Then enforce it: a rule at the organisation level that refuses to create untagged resources works, and a request that people remember does not. The measure of success is the unallocated share, and it is the first number to ask for: if a third of the bill is untagged, every conclusion drawn from the other two thirds is a guess.
Showback and chargeback are the same data with different force. Showback tells a team what it costs and relies on people caring, which is usually enough in a small organisation and has the advantage of not turning cost into an argument. Chargeback puts the amount on that team's own budget and creates real pressure — along with real incentives to game the tags. Start with showback, make the numbers visible and regular, and note that most of the benefit arrives with visibility alone: engineers who can see what their environment costs turn things off without being asked.
Budgets and anomaly alerts do different jobs and you want both. A budget compares the month against a figure you set and is good at "we are on course to overspend" — but a monthly threshold notices a runaway several days late, which for something spending quickly is far too slow. Anomaly detection compares today with the recent pattern and catches the sudden change: a forgotten cluster, a recursive function, a job stuck in a loop. Set both, set them low, and route them somewhere a person actually reads — an alert into an unread mailbox is a slow-motion incident with a witness.
Then the practical exercise this topic exists for: the bill has grown by a third and somebody needs an explanation today. Work in order rather than by hunch.
First, compare like with like: this month against the same period last month, not against a partial month. Second, group by service and find where the increase lives — almost always one or two lines, not a general drift. Third, group that service by tag: which environment, which product, which owner. Fourth, look at the shape — a step change on a particular day means something was created or deployed; a gradual ramp means something is accumulating, usually storage or logs. Fifth, correlate that day with what happened: a release, a new environment, a traffic event, or a commitment expiring, which looks exactly like a price rise because it is one. And finally, remember to check whether something has merely grown as intended — a bill that rises with successful usage is not a problem to be solved, and mistaking it for one wastes a week.
What you should now be able to explain or do
Name four tags worth standardising and say what makes a standard hold. Say which single number tells you whether your cost data is trustworthy. Distinguish showback from chargeback and recommend one for a small team, with a reason. Explain why a budget alone misses a runaway, and what to add. Work through a bill that grew by a third in the five-step order, and name what the shape of the increase tells you.
Check yourself
What is the first number to ask for when you inherit a cloud account?
The unallocated share — how much of the bill has no tags. If it is large, every other cost conclusion is a guess dressed as a finding.
Why does a tagging standard need enforcement rather than agreement?
Because tagging is remembered for the first fortnight and forgotten afterwards. A rule that refuses to create untagged resources holds; a convention does not.
Why is a monthly budget alert not enough?
It measures against a monthly figure, so a fast runaway is noticed days after it started. Anomaly detection compares today against the recent pattern and catches the sudden change.
The bill rose by a third. What is your first step, and what is the wrong first step?
First, compare the same period month on month and group by service to find where the increase lives. The wrong first step is guessing at a cause and looking for evidence for it.
The increase is a step change on the fourteenth. What does the shape tell you?
That something was created or deployed that day — a new environment, a release, a commitment expiring back to on-demand rates. A gradual ramp instead would point at accumulation, usually storage or logs.
Go deeper
We haven't checked most of these for screen reader use yet.
Back to Cost visibility, tagging and budgets: work through the checklist