Every provider advertises uptime. Almost nobody explains what the number permits, or what happens when they miss it. Here is how to read the claim.

The nines, in minutes

Uptime percentages describe allowed downtime over a period. Per month:

  • 99% (“two nines”) allows about 7 hours 18 minutes of downtime
  • 99.9% (“three nines”) allows about 43 minutes
  • 99.99% (“four nines”) allows about 4 minutes 20 seconds
  • 99.999% (“five nines”) allows about 26 seconds

Three nines is the practical standard for business voice. Five nines is a claim worth scrutinizing, because meeting it requires redundancy at every layer including the parts most providers do not control.

What the number usually excludes

This is where the real differences hide. Read the definition of “downtime” in the actual agreement, because most exclude:

  • Scheduled maintenance, sometimes unlimited and sometimes capped
  • Problems in your own network, including your internet connection
  • Upstream carrier failures, which is a meaningful carve-out for voice
  • Force majeure events

A provider offering 99.99% while excluding upstream carriers and unlimited maintenance windows may deliver less real availability than one offering 99.9% with narrow exclusions.

Redundancy versus failover

These are not the same thing and both matter.

Redundancy means more than one of something exists: two data centers, two carriers, two power feeds. It describes capacity.

Failover is the mechanism that moves traffic when one of them dies, and how fast it does so. Redundancy without automatic failover means someone has to notice a failure and act, which converts a two-minute event into a forty-minute one.

For voice specifically, ask about failover at three layers:

  1. Carrier level. If one upstream carrier degrades, do calls reroute automatically to another? Quality-based routing that reacts to answer-seizure ratio and call quality catches problems before they become outages.
  2. Region level. If an entire data center region goes offline, does registration and call control move to another region, and do in-progress calls survive?
  3. Endpoint level. Do your phones and softphones have secondary registration targets configured, or do they simply fail?

What happens when they miss

Most SLAs pay out service credits, typically a percentage of that month’s fee. Credits are modest by design; they are not compensation for lost business, and almost every agreement caps them.

The more useful question is not the credit but the process. Ask whether you get a written post-incident report, how quickly, and whether it names the root cause. A provider willing to publish that is one that investigates. A provider offering only credits is offering you a discount on the problem.

Questions worth asking

Before signing anything, ask for specifics:

  • What was your actual measured uptime in the last twelve months?
  • How many upstream carriers do you route through, and how does traffic move between them?
  • Where are your data centers, and do you keep call data inside the US?
  • What does your maintenance window policy allow?
  • Can I see a sample post-incident report?

Vague answers to these are the answer. A provider that runs redundant infrastructure knows exactly how it is built and will happily walk you through it.