Blog · November 12, 2026 · 12happy team
The nines of availability, and what you can prove
“Three nines”, “four nines”, “five nines”: availability targets are usually counted in nines, the number of 9s at the start of the percentage. It’s a convenient shorthand, but it hides how different the levels are. Each extra nine divides the downtime you’re allowed by ten, and multiplies the effort needed to stay inside it by roughly the same.
This post turns the nines into minutes, explains what each level takes, and covers a part most SLA discussions skip: whether your monitoring can even tell if you met the target.
The nines, in minutes
Using an average month of 30.44 days and a year of 365.24 days:
| Availability | Name | Downtime per month | Downtime per year |
|---|---|---|---|
| 99% | two nines | 7h 18m 17s | 3d 15h 39m |
| 99.5% | 3h 39m 9s | 1d 19h 49m | |
| 99.9% | three nines | 43m 50s | 8h 45m 57s |
| 99.95% | 21m 55s | 4h 22m 58s | |
| 99.99% | four nines | 4m 23s | 52m 36s |
| 99.999% | five nines | 26s | 5m 16s |
For other levels, business hours, or maintenance windows, use the SLA & uptime calculator.
Why calculators disagree by a minute or two
Search for “99.9% uptime per month” and you’ll find 43m 12s, 43m 50s and 44m 38s. All three are right. They assume a 30-day month, an average month, and a 31-day month. The same goes for years of 365, 365.25 or 365.2425 days.
The difference doesn’t matter much in a conversation, but it does in a contract. If an SLA says “99.9% per calendar month”, February allows 40m 19s and March allows 44m 38s. Write down which month you mean.
What each level takes
Two nines and 99.5% leave hours of slack each month. A single server, backups, and someone who looks into problems the same day will usually stay inside them. That’s fine for an internal tool. For a shop or a service customers log into, an outage of several hours is something they notice and remember.
Three nines is where most hosting and software contracts sit, and where most small teams can honestly aim. With 44 minutes a month, the main risk isn’t the outage itself but noticing it late. An outage found 20 minutes in, because a customer emailed, uses half the month’s budget before anyone starts fixing it. Fast alerts and a rehearsed rollback matter more here than extra servers.
99.95% halves that to 22 minutes, and a single half-hour incident misses the month. This is where redundancy starts to pay: a second server or zone that takes over, and deployments that don’t take the site down.
Four nines allows about four and a half minutes a month. That’s less time than it takes to read an alert, open a laptop and log in, so a human can’t be part of the recovery. Failover has to be automatic, and every dependency, from DNS to the database to the payment API, has to be at least as available as the promise. If your app relies on three services at 99.99% each, it can reach about 99.97% at best. The calculator’s composite mode works this out.
Five nines allows 26 seconds a month. It belongs to telephone networks and core infrastructure. For a website it almost never pays off: your visitors’ own connections and devices are often down for longer than that.
The part most SLA discussions skip: can you prove it?
An uptime figure is only as good as the monitoring behind it. A monitor that checks your site every few minutes sees an outage only if a check happens to run during it. So the downtime it reports is only accurate to about one check interval, and an outage shorter than the interval can be missed entirely.
That gives a simple rule of thumb: your monthly downtime budget should be at least three check intervals long for the figure to mean something.
| Checks | 99.9% (43m 50s) | 99.95% (21m 55s) | 99.99% (4m 23s) | 99.999% (26s) |
|---|---|---|---|---|
| Every 5 minutes | Yes | Yes | No | No |
| Every minute | Yes | Yes | Yes | No |
| Every 30 seconds | Yes | Yes | Yes | No |
| Every 15 seconds | Yes | Yes | Yes | Borderline |
The row that matters most is the first. Many free monitoring plans check every 5 minutes, and at that interval 99.99% and 99.9% look the same: a 4-minute outage can fall between two checks. If you promise four nines to a client, you need checks every minute or faster to back it up.
Checking often isn’t the only thing that keeps the figure honest. A check that fails because of one congested network path isn’t an outage of your site. Confirming a failure from a second location before counting it, as 12happy does on every plan, keeps those blips out of your numbers and out of your notifications.
Picking a target
- Start from what you can measure. Your check interval sets a ceiling on the SLA you can report credibly.
- Then look at what you depend on. Your availability can’t beat the product of your dependencies’. For services like Stripe and Cloudflare, live status pages show their recent uptime from our own checks.
- Then decide what an outage costs. Each nine costs roughly ten times more than the last. Spend that only where an outage costs more than the redundancy would.
For most websites and web applications, that lands on 99.9%, measured by checks every minute or faster, with failures confirmed from more than one place. It’s a promise a small team can keep and prove.