"99.99% SLA" in one line reassures leadership. Nobody translates that into 52 minutes per year — or reads page 14 of the terms where planned maintenance, "exceptional" DDoS attacks, and client-side DNS failures do not count. The day the shop stays down two hours on a Friday, support points to an exclusion. Promise and remedy were never the same thing.
A hosting SLA (Service Level Agreement) sets service levels and sometimes credits if the host misses them. Reading an SLA means separating what is measured, what is excluded, and what you actually recover when things fail. The homepage percentage is only an entry point; the full contract is what matters on outage day.
Breaking down the uptime number
On a civil year — 525,600 minutes — theoretical orders of magnitude are:
| Advertised SLA | Theoretical max downtime / year |
|---|---|
| 99% | ~87 h 36 min |
| 99.9% | ~8 h 45 min |
| 99.99% | ~52 min |
| 99.999% | ~5 min |
Three questions must follow immediately. What is measured: datacenter ping, hypervisor, load balancer, client URL? Where: one region or the whole network? Who measures: host internal tool or contractually accepted external monitoring?
Typical exclusions to track
"Exclusions" or "Force majeure" sections deserve more time than the marketing badge. Announced maintenance is often removed from the count even when your site is offline. Client factor — full disk quota, expired certificate, bad configuration — excludes the outage from host liability. Third-party services (registrar, external CDN, payment), attacks beyond a "reasonable" threshold, and off-datacenter causes appear regularly.
Two hosts showing 99.9% can have incomparable exclusion universes. The number alone creates false precision. Read the SLA like a lawyer, not like a buyer reassured by one more nine.
SLA credits: real compensation
When breach is proven, the host often grants a credit on a future invoice: typically 5–10% of the affected month per downtime band, capped at 100% of the month — rarely more. Claims may must arrive within 7–30 days with tickets and proof. An €8 credit on an €80 VPS does not compensate €40,000 in abandoned carts.
For real continuity, cross the SLA with a disaster recovery plan and a reliable status page. Compare offers via the comparator by reading contractual PDFs, not pricing pages.
The climax: the SLA protects the host more than your revenue
Serious hosts publish incident history; others publish a percentage with no documented method.
Decide and move forward without blind spots
Download the full SLA — uptime and support — before signing and highlight definitions and exclusions in a document shared with the technical team. Convert the advertised percentage into theoretical lost minutes, then list what is removed from the calculation. Verify scope: hypervisor only, HTTP service, or your entire application. Require a public status page and proactive notifications rather than a badge with no history. Finally, compare providers on evidence via the directory — not on the number alone.
Frequently asked questions
What does 99.9% availability mean?
About 8 h 45 min of downtime per year if measured across the full year without exclusions. Many SLAs measure only datacenter network or hypervisor — not your site or database.
Do SLA credits compensate for an outage?
Rarely proportionally. A 5–10% monthly credit after hours of downtime does not cover lost revenue or reputation. The SLA mostly caps the host's liability.
Does planned maintenance count in the SLA?
Often not if announced 24–72 hours ahead. Read the exact definition of "unavailability" and allowed maintenance windows.
Support SLA vs uptime SLA?
Two separate documents. Uptime covers reachable infrastructure. Support sets first human response time. Good uptime with bad support still leaves your team alone — see our host support guide.
Next time someone shows you 99.99%, ask what counts as an outage — and the maximum credit in euros. Those two lines are worth the percentage.