Understanding
your Internet SLA.

01 / Availability

What does the percentage mean?

Availability is generally the proportion of a defined period when a service meets the agreement’s definition of available. The exact calculation varies: check the measurement period, what counts as an outage, when the clock starts and stops, and which events are excluded.

The figures below are illustrative arithmetic for a 30-day month measured continuously, with every minute included. They show the time corresponding to the unavailable portion of each percentage. They are not a planned outage allowance or a prediction of actual service.

For example, a 30-day month contains 43,200 minutes. At 99.9%, 0.1% of that period is 43.2 minutes. A different month length, a minimum outage duration, excluded maintenance, or another method of counting changes the result under the agreement.

02 / Scope

Where does the commitment begin and end?

A provider may measure availability at a service handoff, across part of its network, or between named locations. The service boundary is the portion covered by the commitment. Ask the provider to show it on a diagram and explain which equipment and connections are included.

Your experience also depends on the office network, firewall, Wi-Fi, other networks, and the application itself. A working handoff does not establish that every destination is reachable or that an application is responding normally. Equally, a problem outside the covered boundary may still affect your work.

Read the exclusions as well as the target. Agreements may address planned maintenance, customer equipment, third-party networks, and events outside a provider’s control differently. Ask how the provider investigates the cause and how it treats uncertain or missing measurements.

03 / Performance

A connection can be up and still be slow.

Availability describes one part of the experience. Depending on the agreement, an SLA may also address latency, packet loss, or jitter. These can matter to calls, interactive applications, remote desktops, and transfers even when a link remains connected.

For each target, ask which endpoints and protocol are used, how often measurements are taken, and whether the result is an average, percentile, or another statistic. A 95th percentile describes the upper end of a set of readings: roughly 95% are at or below that value. It is not the maximum, and the measurement period matters.

Also ask how long performance must exceed a threshold before it qualifies, how missing samples are handled, and what evidence is used to determine responsibility. A public dashboard can be useful context without being the contractual measurement.

04 / When something goes wrong

Response time and repair time are different.

A response commitment describes when support will acknowledge or begin working on a report. A restoration commitment describes when service is expected or required to be restored, subject to the agreement. Some documents state targets rather than guaranteed deadlines. Read the exact language and qualifying conditions.

Ask how to reach support, whether it is available around the clock, what information starts an incident, and how updates and escalation work. For critical services, ask which failures are covered and whether a temporary workaround counts as restoration.

05 / Remedies

What happens if the target is missed?

Many agreements offer a service credit for a qualifying incident. A credit may be calculated against a particular circuit’s monthly charge rather than the whole invoice, and it may be subject to thresholds, caps, or other conditions. Other remedies, including cancellation rights, depend on the agreement.

Ask whether a credit is automatic or must be requested; who must be notified; the deadline; what logs, ticket numbers, or timestamps are needed; and whether the credit is the sole remedy. Keep a record of the incident, including when it began, what was affected, and the support case.

A credit can recognize a missed commitment, but it does not restore lost work or necessarily cover the business impact. Evaluate the operational response and resilience design alongside the financial remedy.

06 / Specialized commitments

What about an AI connectivity SLA?

Some agreements add commitments for specific destinations or workloads. For an AI service, network latency and reachability can be useful measures, but the time an employee waits for a result also depends on the application, model processing, and the customer’s own network. Ask exactly what is measured and which destinations are covered.

The published addendum uses IPv4 ICMP round-trip latency at the 95th percentile in defined five-minute intervals and separate TCP port 443 reachability checks. A qualifying incident must meet the stated duration and attribution conditions. It also specifies notice requirements and remedies; consult the addendum and signed agreement for the full terms.

Those measurements do not measure AI inference time, guarantee application uptime, or include the customer’s premises network. Public endpoints can resolve to nearby CDN locations. The addendum does not create a packet-loss or jitter guarantee. A public graph’s averages do not by themselves establish a contractual breach.

07 / Comparing proposals

Questions to take to each provider.

• What service, locations, equipment, and hours are covered? Where is the handoff?
• What counts as an outage, and when does the measurement clock start and stop?
• Which events are excluded, and how are maintenance and partial failures treated?
• Are latency, loss, jitter, or throughput covered? Between which points, and using what method?
• How are missing readings, short interruptions, and attribution handled?
• What are the support response and restoration commitments, and how are incidents escalated?
• What remedy applies, what charges is it based on, and are there caps or cancellation rights?
• Who must report an incident or request a credit, by when, and with what evidence?
• What services must remain operational during an outage, and has the backup been tested?

Use the same questions for each proposal. A higher percentage is useful only when you understand what it covers and how it relates to the work your business needs to do.