You didn't lose customers to downtime. You lost them to a quota nobody was watching.
Cloud providers meter everything - bandwidth, API calls, database reads - and when a number crosses its line, nothing crashes. The service just starts saying no, quietly, while the bill keeps climbing in the background.
Nothing crashed. The service just started saying no.
Cloud providers meter almost everything now: bandwidth out of a database, rows read per month, API calls per minute. None of it is a hard wall with a warning light. It's a sliding scale - fast and free until a threshold passes, then throttled, then degraded, then a bill that jumps without anyone changing a line of code. A support ticket that used to load in a second takes four. A sync job that ran every five minutes starts silently skipping runs. Nobody shipped a bug. The business just used the product more than the plan assumed it would.
The bill is cheaper than the distraction it causes
37signals famously left the public cloud in 2022 after watching their AWS bill climb for years on usage that had become routine, not exceptional - they projected over $7 million in savings across five years once they owned the hardware instead of renting it by the gigabyte. Most businesses never get a number that clean, because most never measure it until the invoice forces them to. Cloud waste studies have put unused or over-provisioned spend at roughly a third of total cloud budgets for years running - not fraud, just nobody assigned to watch the meter.
The warning email exists. It's just not addressed to anyone in particular
Every major provider sends a usage-threshold email before the throttling starts - 80% of quota, then 100%, then a grace period before Fair Use kicks in. The email is real and specific. It goes to whichever address signed up for the account, which by the time it matters is often a founder's personal inbox, a shared ops alias nobody checks weekly, or someone who left the project eight months ago. The warning wasn't skipped. It just wasn't anyone's job to read it.
The ACS approach
We set a budget alert on every service before it goes live, not after a surprise invoice teaches us to. We know in advance what happens at 80%, 100%, and 120% of whatever we're metered on, and who gets paged at each step - not just a noreply address nobody owns. Growth should force a conversation about scaling up, not a forensic one about what happened last month.