The cloud bill doubled over three months with no traffic growth. Where do you look?

What the interviewer is testing

Whether you have a systematic method rather than a hunch, and whether you know the categories that grow without traffic.

The method

Start from the bill itself, decomposed by service, then by account, then by tag, over time. The step change or the gradient tells you where and when. Most organisations skip straight to guessing; the bill answers the question directly if the tagging is in place.

The usual causes, in rough order

Data transfer, particularly cross-zone and cross-region. Grows with chattiness and with instance count rather than with user traffic, and appears as a line item nobody owns.

Storage accumulation. Nothing deletes: logs, backups, snapshots, old artifact versions, non-current object versions, orphaned volumes from terminated instances. This is slow, compounding and entirely invisible until someone looks.

Observability spend. Log volume rises with instance count and with any debug logging left on. A single verbose deployment can double it.

Environments that were never destroyed — preview environments, a migration's parallel infrastructure, a proof of concept.

Autoscaling floors raised during an incident and never lowered.

A managed service tier upgrade made for one requirement and applied broadly.

Expired commitments. Reserved capacity or savings plans lapsed, so the same usage now bills at on-demand. This produces exactly the "no change, double cost" pattern and is very commonly the answer.

The fixes beyond the immediate cause

Cost anomaly detection with alerting, so a step change surfaces in days rather than at the quarter.

Mandatory tagging enforced at resource creation, so attribution is possible at all.

Showback per team, published — the single most effective behavioural control.

Lifecycle policies on storage by default, and TTLs on ephemeral environments.

What a strong answer adds

Expressing the result as unit cost — cost per customer, per order, per request — rather than as total spend. That distinguishes "we grew" from "we got less efficient", which is the question leadership is actually asking.

Common weak answers

Rightsizing instances as the first move, which is rarely where a doubling comes from. Asking teams to be careful.