Every hosting sales page has an uptime number on it, usually 99.9% or 99.99%, sometimes dressed up as “five nines” marketing even when the actual contract says something less impressive. The gap between those numbers looks small until you do the math in minutes. This is a guide to that math, how service credits work, the ways uptime gets measured (and gamed), and what to actually ask for before you sign anything.
The math: how much downtime each tier allows
Uptime percentages are just a ratio of time up to total time. Flip the percentage around and you get your allowed downtime budget.
- 99.9% uptime allows about 43 minutes of downtime per month, or roughly 8.76 hours per year.
- 99.99% uptime allows about 4.3 minutes of downtime per month, or roughly 52 minutes per year.
That is a 10x difference in allowed downtime for a change of one digit. It is also why 99.99% is meaningfully harder to deliver than 99.9%. Getting from three nines to four nines usually requires redundant infrastructure, automated failover, and monitoring tight enough to catch problems in seconds rather than minutes. If a host promises 99.99% without describing any of that underlying architecture, ask how they get there.
It is worth noting that for most WordPress sites, the practical experience of “uptime” is shaped by more than the origin server. If you run Cloudflare in front of your site, edge caching can keep pages serving from the CDN’s cache even during a brief origin hiccup, which is one reason a well-cached WordPress site often feels more available than the raw server-level SLA would suggest. That is not a substitute for a real SLA, but it does mean the two numbers, contractual and experienced, are not always the same thing.
How service credits actually work
An SLA is a promise with a penalty attached, and the penalty is almost always a credit against future hosting costs, not a refund and not compensation for lost business. Read the credit schedule carefully. Most tiered SLAs look something like this:
- A small credit (often a percentage of one month’s fee) for dropping below the promised uptime but staying above a lower threshold.
- A larger credit for falling into a worse bracket, say below 99% or below 95%.
- A cap, almost always, on the total credit you can claim in a billing period, frequently one full month’s fee and never more.
Two things matter here that are easy to miss. First, credits are usually not automatic. Many hosts require you to file a claim within a specific window, often 30 days, with documentation of the outage. If you do not ask, you do not get credited, even if the outage clearly happened. Second, the credit is proportional to your hosting bill, not to the outage’s actual cost to your business. If your site went down during a product launch and cost you real revenue, the SLA credit will not come close to covering it. SLAs are a floor on accountability, not insurance.
The measurement games
Uptime percentages are meaningless without a clear definition of what counts as “down.” This is where SLAs get slippery, and where you should read closely.
Who is measuring, and how often
Some hosts measure uptime using their own internal monitoring, checked from inside their own network. That setup has an obvious conflict of interest: the party grading the exam wrote the exam. Better SLAs specify third-party or external monitoring, checking the site from outside the host’s infrastructure at a defined interval, commonly every one to five minutes. A monitor that only checks once every 30 minutes can miss short outages entirely, since a five-minute blip that starts and ends between checks never gets recorded.
What counts as “unavailable”
Does a 500 error count? A timeout? A page that loads but with broken assets? Some SLAs only count total unresponsiveness, meaning degraded performance, partial failures, or a database error on some pages but not others may never trigger a credit at all. Ask for a precise definition of a qualifying outage, not a vague phrase like “service interruption.”
Exclusions
Nearly every SLA excludes scheduled maintenance, and that is reasonable within limits, but check how much notice is required and how long a maintenance window can be before it stops being reasonable. Also watch for broad exclusions covering “factors outside our control,” which can be stretched to cover almost anything, including issues that were arguably within the host’s control, like a botched update to their own platform.
Measurement window
A 99.9% SLA measured monthly is stricter than one measured annually, because a single bad day can blow through a whole month’s downtime budget but gets diluted across a year. Confirm the measurement period before comparing two hosts’ promised numbers directly.
What to actually demand in an SLA
Before signing on the promise of a number, look for these specifics in the actual contract, not the marketing page:
- An explicit definition of “downtime” or “unavailable,” ideally covering both full outages and severe degradation.
- Independent, external monitoring with a stated check interval, not self-reported internal metrics.
- A clear, short claims process, or better, automatic credits with no claim required.
- Narrow, specific exclusions for maintenance, with advance notice requirements.
- A monthly (not annual) measurement window, since it is the stricter and more actionable standard.
None of this replaces good operational practice on your own site. Keeping WordPress core, PHP, and plugins current, running a persistent object cache like Redis so a database hiccup does not cascade into full downtime, and keeping backups current all reduce your actual risk regardless of what the SLA says. A managed host that handles the server-side redundancy, patching, and monitoring for you (this is the kind of thing ServerBorn’s NVMe and Redis-backed stack is built around) removes one layer of risk, but the SLA is still the paper trail that tells you what happens when something goes wrong anyway.
Takeaway
99.9% and 99.99% are not just marketing digits, they represent a real 10x difference in allowed downtime, roughly 43 minutes versus 4 minutes a month. But the number on the sales page is only useful once you know how it is measured, what counts as an outage, and how credits are actually paid out. Read the SLA itself, not the badge, and you will know exactly what you are buying.