Uptime, Downtime, and What 99.9% Means
Every hosting company advertises an uptime number, and nearly every customer reads it wrong. "99.9% uptime" sounds like "essentially always up." It actually means your site can be down for almost nine hours a year — and the guarantee behind it usually promises not that this won't happen, but that you'll get a small service credit if it does. Neither fact is a scandal; both are just arithmetic and contract language. But an owner who understands the arithmetic and the contract makes much better hosting decisions than one comparing decimal points on marketing pages.
This post covers the math of the nines, what an SLA does and doesn't promise, why when and how often matter more than how much, and what a small business should actually do about all of it.
The nines, converted to human time
Uptime percentages compress badly into intuition, so here's the table worth bookmarking. Percent of a year, converted:
| Uptime | Downtime per year | Per month (avg) | Per week (avg) |
|---|---|---|---|
| 99% | ~3.65 days | ~7.3 hours | ~1.7 hours |
| 99.9% | ~8.77 hours | ~43.8 minutes | ~10 minutes |
| 99.95% | ~4.38 hours | ~21.9 minutes | ~5 minutes |
| 99.99% | ~52.6 minutes | ~4.4 minutes | ~1 minute |
| 99.999% | ~5.3 minutes | ~26 seconds | ~6 seconds |
Three observations fall out of this table:
- The gap between nines is 10x, not incremental. Three nines and four nines look like cousins on a pricing page; they are an order of magnitude apart in reality. Each extra nine typically costs disproportionately more to deliver, because it demands redundancy at layers that were previously single points of failure.
- Two nines is not a good number. It reads as "almost perfect" and permits three and a half days of annual outage. Any host advertising two nines is telling you something.
- Five nines is not a hosting product. ~5 minutes a year requires engineering (multi-region failover, redundant everything) that no shared or ordinary managed hosting provides, whatever the sales page implies. When you see five nines advertised on commodity hosting, read it as decoration.
For calibration: quality managed WordPress hosting realistically delivers between three and four nines in practice — minutes to a few hours per year — and that is genuinely sufficient for almost every small business, for reasons we'll get to.
What an SLA actually promises (read the exclusions)
An uptime SLA — service level agreement — is not a promise your site won't go down. It is a formula for compensation when measured downtime exceeds a threshold. Three parts matter, and the fine print is where each one lives:
1. The remedy is credits, not damages
The standard remedy is a service credit: some share of your monthly hosting fee, often on request rather than automatically. Concretely: if the host misses its target badly, your compensation might be a fraction of one month's fee in hosting credit — for an outage that possibly cost you a day of orders. An SLA transfers a trivial fraction of your downtime risk. Its real value is as a signal: a host that commits to a high target with automatic credits has aligned incentives and confidence in its infrastructure. The credit itself will never make you whole, and no SLA pretends otherwise.
2. The exclusions define the number
Every SLA excludes categories of downtime from the calculation. Typical exclusions:
- Scheduled maintenance — announced windows don't count as downtime. Fine if windows are rare and off-peak; check how much notice is promised.
- Your fault — a plugin you installed crashing PHP, exhausting your plan's resources, a broken deploy. Legitimate exclusion, and worth internalizing: on WordPress, a meaningful share of real-world downtime is application-level (a bad update, a conflicting plugin), which no host SLA covers but a good managed host actively prevents.
- Force majeure and upstream failures — DDoS attacks, DNS provider outages, cloud-region incidents are often excluded or contested.
- Measurement fine print — some SLAs only count outages longer than N minutes, or measure per-month (a 43-minute outage in a three-nines monthly SLA is within target every single month — about 8.7 hours a year while never breaching).
3. Who does the measuring
If the host's own monitoring is the referee, disputes default in their favor. This is why you run your own monitor — covered below. It costs nothing and converts "I think it was down a lot" into a timestamped record.
Frequency and timing beat totals
Annual percentages hide the two variables that determine actual business damage:
Shape. Eight hours of downtime as one overnight incident during your dead hours is an inconvenience. The same eight hours as ninety 5-minute outages sprinkled across business days is a catastrophe the percentage can't see: every customer who hits one of those windows experiences "this business's website doesn't work," your checkout abandonment climbs, and search crawlers repeatedly find errors. Frequent short outages also indicate an unstable platform — overloaded shared servers behave exactly this way — whereas one long outage can happen to anyone.
Timing. Downtime is worth what the lost minutes were worth. A florist down on the Tuesday before Valentine's Day loses more than in the whole rest of the quarter. When you evaluate an incident — yours or a host's public postmortem — ask when it happened and what it collided with, not just how long it lasted.
A practical scoring rule when comparing hosts or judging your current one: prefer the host with rarer, well-explained incidents over the one with a slightly better percentage and chronic micro-outages. A public status page with honest incident history is worth more than a fourth nine on the pricing page.
What downtime actually costs a small business
Worth estimating once, roughly, because it sets your budget for caring:
- Direct: (revenue per hour through the site) × (hours down during business hours). An e-commerce shop's revenue per hour averages out low — but sales concentrate, so peak-hour downtime costs several multiples of that average.
- Leads: a service business missing two inquiries during an outage may have lost jobs worth hundreds or thousands — the highest-stakes case per minute, because leads don't retry.
- Trust and search: a visitor who hits an error rarely reports it; they just leave. Search engines tolerate brief outages gracefully — a site down for minutes loses nothing; a site down for days starts seeing pages drop from the index. Chronic instability is the pattern that does SEO damage.
The conclusion of this arithmetic for most small businesses: the difference between three-nines and four-nines hosting is worth something, but far less than the difference between knowing and not knowing when you're down — which brings us to the practical program.
What to actually do: monitor, then decide
Run your own uptime monitor (non-negotiable, free)
A free tier from any uptime service (UptimeRobot, Better Stack, Pulsetic, many others) will ping your site every 1–5 minutes and alert you by email or phone when it fails. Ten minutes of setup buys you three things: you learn about outages before customers tell you, you accumulate an independent record for SLA claims and host evaluations, and after a few months you know your actual uptime instead of the advertised one. Two configuration tips:
- Monitor a real page, not just the server. Check for a keyword on your homepage — a server can respond 200 OK with a database error page. Keyword checks catch "up but broken."
- Add a second check for your most valuable path if the tool allows — the checkout, the booking page.
- Route alerts somewhere you'll actually see them. An email alert that lands in a folder you check weekly defeats the purpose. Use push notifications or SMS for the primary check, and set a sensible confirmation rule (alert after two consecutive failures) so a single network blip doesn't page you at 3 a.m. for nothing.
Judge your host on evidence
After 3–6 months of monitoring data, the decision rule is simple:
- Measured uptime at or above three nines with incidents that were rare, brief, explained, and off-peak: you're fine. Spend your energy elsewhere.
- Repeated short outages during business hours, or any pattern of unexplained downtime and evasive support: move hosts. No SLA credit compensates for a platform you can't trust, and migration is cheaper than the anxiety.
Reduce the downtime that's yours
Remember that application-level failure — the bad plugin update, the theme conflict, the expired certificate — is downtime your visitors can't distinguish from a server fire, and it's the kind you control. Staged updates (test before production, or a managed host that does safe updates for you), automatic SSL renewal, and tested backups convert the most common self-inflicted outages into non-events. For many WordPress sites, disciplined updates improve real availability more than upgrading to a fancier hosting tier would.
Conclusion
Three nines means about nine hours a year, an SLA means a modest credit and a set of exclusions worth reading, and the numbers that actually predict your experience are frequency, timing, and how honestly your host communicates when things break. The right posture for a small business is neither obsession nor faith: set up an independent monitor this week, estimate what an hour of downtime costs you, hold your host to the standard that math justifies — and remember that some of the most damaging downtime is the kind you cause yourself, which no one's SLA covers but your own maintenance habits prevent.