Your contract promises "99.99% guaranteed availability." Sales insists it is enterprise-grade. Then one Tuesday morning your shop stays down for forty minutes — unannounced network maintenance, according to the ticket. The SLA does not apply: the outage is "out of scope." You discover the number you bought never measured your database, your CDN or your TLS certificate.
The question is not "should we aim for the highest percentage?" It is: what exactly is measured, who is exempt, and what happens when the guarantee fails?
Two numbers, two orders of magnitude
On paper, the gap between 99.9% and 99.99% is a decimal place. In annual downtime, it is brutal.
| Advertised SLA | Theoretical downtime / year | Monthly equivalent |
|---|---|---|
| 99.9% | ~8 h 45 min | ~43 min |
| 99.99% | ~52 min | ~4 min |
| 99.999% | ~5 min | ~26 s |
These figures assume continuous measurement with no exclusions. Most hosting contracts define "available" as an HTTP response from the hypervisor, the control panel or a generic endpoint — not your application's ability to process a payment.
An SLA is a contractual promise on a defined scope. It is not a promise that your site will work.
What the contract measures — and what it excludes
Before comparing two offers, ask for the full SLA document, not the quote line. Four areas decide real value.
The definition of "up". Returning HTTP 200 on a static status page is not the same as a working checkout. Check whether measurement includes the database, object storage, private network or only shared infrastructure.
Exclusions. Planned maintenance, force majeure, DDoS above a threshold, misconfiguration on your side, external DNS, certificates you manage yourself: the list is long. Every exclusion is free downtime for the provider.
Credit caps. Many SLAs limit refunds to 10–30% of the monthly fee for the affected service. A four-hour outage on a sale day can cost thousands; the credit, a few dozen.
Claim deadlines. Some contracts require opening a ticket within 24 hours to claim a credit. Without an internal process, you lose the commercial gesture — without recovering lost availability.
Scenario A vs B: same number, different risk
Scenario A — SMB e-commerce, managed 99.99%. The host measures availability of the application cluster including MySQL and load balancer. Maintenance announced 72 hours ahead, outside peak hours. 24/7 incident support with documented escalation. The 40% premium over 99.9% pays off if one hour of downtime costs more than the annual difference.
Scenario B — Corporate blog, 99.99% "infrastructure" VPS. Measurement covers the hypervisor. Your WordPress, plugins and manual backups: outside the SLA. A plugin update that breaks the site triggers no credit. Here, 99.9% vs 99.99% matters little: the main risk is application-level, not infrastructure.
| Criterion | Scenario A (managed) | Scenario B (VPS only) |
|---|---|---|
| Measured scope | App + DB + LB | Hypervisor |
| Main risk | Infra + load | Code + plugins |
| SLA useful? | Yes, if credits + process | Low |
| Alternative | Multi-region, runbook | Staging, tested backups |
The peak: an SLA protects the contract, not your business
Here is what pricing pages avoid stating clearly.
Teams that hold under load do not stop at the quote number. They simulate failure: node loss, database corruption, certificate expiry. They time recovery. That duration — not the advertised SLA — is what matters on the day it breaks.
Decide and move forward without blind spots
To choose between 99.9% and 99.99% without blind spots:
- Calculate the cost of one hour down for your business (lost revenue, support, reputation).
- Read the full SLA: scope, exclusions, credit cap, claim deadline.
- Compare the annual premium for the higher tier to the maximum possible credit.
- Invest the difference in external monitoring, tested backups and an incident runbook.
If the premium exceeds your plausible maximum loss, the lower tier plus application redundancy often wins. Filter hosts by SLA transparency in our directory or compare tool.
Frequently asked questions
How much downtime does a 99.9% SLA allow?
In theory, about 8 hours 45 minutes per year. In practice, maintenance, DNS exclusions and client-side failures often fall outside the calculation.
Is a 99.99% SLA worth the premium?
Only if the scope covers your real service and your team cannot compensate alone. Otherwise you are funding a marketing number.
Do SLA credits make up for an outage?
Rarely. A partial credit does not replace lost revenue or trust. Treat it as a commercial gesture.
What should you ask before signing an SLA?
The definition of "available", exclusions, measurement window, notification timeline and uptime history on the exact scope.
Next time a quote pits 99.9% against 99.99%, ask one thing: show me what is measured and what is not. The percentage only makes sense after that answer.
