A service-level agreement (SLA) is the part of a contract, or a separate document attached to it, that sets measurable performance targets for a service and states what happens if they are missed. Typical targets cover availability, response time and resolution time; typical remedies are service credits, escalation and, for serious or repeated failures, termination rights.
An SLA may be a standalone document, a schedule to a master agreement, or a supplier’s published terms incorporated by reference.
Why service-level agreements matter
A contract that says a supplier will provide “reliable” service is hard to enforce. An SLA turns that promise into numbers that can be measured, reported and acted on. It also sets the price of failure, and often limits it: many SLAs make service credits the customer’s only remedy for missing a target.
Which services, regions, environments and hours the targets apply to.
Metrics and targets
Availability, response time, resolution time, error rate or throughput, each with a number.
Measurement
How each metric is measured, over what period, and who measures it.
Exclusions
Scheduled maintenance, customer-caused issues, force majeure and third-party failures.
Remedies
The credit schedule, any caps, and whether credits are the sole remedy.
Claim process
Deadline, required evidence and how credits are applied.
Reporting and review
Regular performance reports, review meetings and how the SLA can change.
How availability is calculated
The most common SLA metric is availability, usually expressed as a monthly percentage:
Availability = (total minutes − qualifying downtime minutes) ÷ total minutes
A 30-day month has 43,200 minutes, so a 99.9% target allows 43.2 minutes of downtime. What counts as downtime is defined by each agreement. For example, Google Cloud’s Compute Engine SLA counts only periods of one or more consecutive minutes and excludes shorter, intermittent interruptions.
Service-level agreement example
An illustrative software service costs $20,000 a month and promises 99.9% monthly availability. The SLA gives a 10% credit for availability from 99.0% to under 99.9%, 25% from 95.0% to under 99.0%, and 100% below 95.0%. Scheduled maintenance announced in advance is excluded.
In a 30-day month there are two outages of 95 and 35 minutes, plus 60 minutes of announced maintenance.
Qualifying downtime is 95 + 35 = 130 minutes. Availability is (43,200 − 130) ÷ 43,200 = 99.70%.
That falls in the 10% tier: 10% × $20,000 = a $2,000 credit, if the customer claims it within the deadline and with the evidence the SLA requires.
Convention used: a 30-day month, with announced maintenance excluded. Counting the maintenance would give 99.56%, still in the 10% tier here, but the difference can change the tier in other cases.
An illustrative month of SLA performance, the credit schedule and the resulting claim.
A real SLA: what the fine print looks like
Google Cloud’s Compute Engine SLA (as last modified 14 October 2026) shows how much the details matter. Its credits depend on the service tier and configuration; the customer must notify support within 60 days of becoming eligible and provide log files showing the downtime, or forfeit the credit; and the SLA states that it is the customer’s “sole and exclusive remedy” for missing the target. Other providers’ terms differ, and negotiated contracts may override published ones.
SLA vs. related terms
Term
How it differs
Service-level objective (SLO)
An internal or published target, such as 99.9% availability. An SLA adds contractual commitments and remedies.
Service-level indicator (SLI)
The measurement itself, such as the availability percentage actually recorded.
Service credit
The usual remedy under an SLA: a reduction in fees when a target is missed.
Key performance indicator (KPI)
A broader business measure; it may or may not be contractual.
Common misunderstandings
An outage means a credit. Only qualifying downtime, measured the SLA’s way and claimed in time, earns one.
Credits compensate for losses. They are usually a small share of fees and may be the only remedy.
Higher targets are always better. Each extra “nine” costs more; set targets the business actually needs.
How teams use service-level agreements
Procurement: negotiates targets, credits, caps and exit rights.
IT and operations: monitor performance and collect incident evidence.
Finance: confirms credits are applied to invoices.
Legal: reviews exclusions, sole-remedy language and termination rights.
Frequently asked questions
Is an SLA legally binding?
Yes, when it forms part of the contract, directly or by reference. A published target that is not incorporated may be only a statement of intent.
How much downtime does 99.9% allow?
About 43 minutes in a 30-day month, or roughly 8.8 hours over a 365-day year.
Are SLA credits paid automatically?
Often not. Many SLAs require the customer to claim, with evidence, within a set deadline.
What is the difference between an SLA and an SLO?
An SLO is the target; an SLA is the contractual commitment to it, including what happens if it is missed.
Where ObligaAI fits
ObligaAI describes its platform as helping teams search key terms across contracts, track obligations and key dates, and spot missed value and recovery opportunities.