Service Level Agreement (SLA): Definition and Examples

Most SLAs share the same core parts:

• Service description. What service is covered, and what is not.
• Targets. The measurable levels the vendor commits to, such as 99.9% monthly uptime or a 30 minute first response.
• Measurement. How each target is measured, over what period, and from whose data.
• Responsibilities. What the vendor must do and what the customer must do, such as reporting problems through the right channel.
• Exceptions. Events or conditions the targets do not cover. Read this list closely, because it shapes every number above it.
• Remedies. What the customer gets when a target is missed, and how to claim it.
• Review and exit. How often the SLA is reviewed and when repeated misses allow the customer to leave.

An SLA is often an exhibit to a larger contract. Our guide to the master service agreement explains how those documents fit together.

Sources: Desk365, "What are key components of an SLA?"; Alloy Software, "The components of an SLA"; BMC Helix, SLA template section "Exceptions and limitations"; Dropbox Sign, "Where do SLAs, NDAs, and purchase orders fit?"

Engineers at Google split the idea into three terms. The split is useful for buyers too.

Google's site reliability book gives a simple test. Ask "what happens if the SLO is not met?" If nothing happens, it is an objective, not an agreement. The consequence is often money, such as a rebate or penalty, but it can take other forms.

Sources: Google, Site Reliability Engineering, chapter "Service Level Objectives"

There are three common types:

• Customer SLA. Between a vendor and its customer. A software vendor that sells online access will usually have one.
• Internal SLA. Between teams in the same company, such as IT and sales.
• Multilevel SLA. Combines levels. For example, one set of terms for every customer, and extra terms for one service or one customer group.

Sources: Vendr, "The 3 types of SLAs"; Alloy Software, "Types of SLAs"

Definitions vary between providers. Read how each one is measured before comparing two vendors.

Sources: Kelser, "What Is Response Time?" and "What Is Resolution Time?"; Desk365, "Measuring SLA Performance"

Uptime is often written as a number of nines. Google's SRE book notes that 99% is called "two nines" and 99.999% is "five nines." Small changes in the percentage make a big difference in allowed downtime.

The table below shows the downtime each level allows in a 30 day month. A 30 day month has 43,200 minutes.

Sources: Google, Site Reliability Engineering, "Service Level Objectives". The minutes are our own arithmetic: 43,200 minutes multiplied by the share of time allowed down.

Amazon Web Services publishes an SLA for its EC2 compute service. For instances deployed across two or more availability zones in a region, AWS commits to a monthly uptime of at least 99.99%. If it misses, the customer can claim a credit:

Three details in the fine print matter more than the headline number:

• You must claim it. The customer opens a support case to request the credit. The request must arrive by the end of the second billing cycle after the incident.
• Credits are not cash. The SLA says credits do not entitle the customer to a refund or other payment.
• Small credits are dropped. A credit is issued only if it is more than $1 for the billing cycle.

The same SLA also sets a separate, lower target of 99.5% for a single instance.

Sources: Amazon Web Services, "Amazon Compute Service Level Agreement"

Say a company spends $8,000 a month on cloud compute in one region. One month, uptime is 99.5%. That falls in the 10% band, because it is below 99.99% but at least 99.0%.

• Credit: 10% of $8,000 = $800, applied to a future bill.
• Downtime: about 3.6 hours in a 30 day month.

Now ask what 3.6 hours of outage cost the business. If the outage stopped online orders worth $5,000 an hour, the loss is about $18,000. The credit covers less than 5% of it.

So an SLA credit works like a discount, not insurance. Buyers who depend on a service should plan for outages with their own backups, and price that work into the deal. Count that backup work as a line in your total cost of ownership.

Sources: Amazon Web Services, "Amazon Compute Service Level Agreement". The $8,000 bill, the outage and the order values are invented for teaching.

1. Is each target measured over a clear period, such as a calendar month?
2. Whose monitoring data decides if a target was missed?
3. What is excluded from the targets, and who decides?
4. Do response and resolution times run in business hours or around the clock?
5. Is the credit automatic, or must you claim it? By when?
6. Is there a cap on credits, and are they cash or account credit?
7. Do repeated misses give you the right to end the contract?
8. Does the vendor report its own SLA results to you?

If you also need proof that a vendor's controls work, ask for its audit report. Our guide to the SOC 2 report explains what it shows. For a wider view of vendor checks, see third party risk management.

Sources: Desk365, "Business Hours vs 24x7 SLAs"; Amazon Web Services, "Amazon Compute Service Level Agreement"; DiliTrust, "Common Pitfalls in SLA Management"

• Enterprise software providers listed in our B2B platform register.
• Support and retention teams covered by the customer success category.
• The evidence standard this publication applies, set out in the methodology.

1. What is a service level agreement?

   A service level agreement is a promise, usually in a contract, about how well a service will perform. It sets measurable targets such as uptime and response time, and names a remedy, often a service credit, if a target is missed.

2. What is the difference between an SLA and an SLO?

   An SLO is an internal target for a measured number, such as 99.9% of requests succeeding. An SLA is an agreement with a customer that attaches a consequence, such as a credit, when a target is missed.

3. What is an SLA in simple terms?

   An SLA is a written promise of service quality with a penalty if the promise is broken. For example, a vendor may promise 99.9% uptime and give a 10% credit if it falls short.

4. What are the three types of SLAs?

   The three common types are customer SLAs, internal SLAs and multilevel SLAs. A customer SLA covers a vendor and its customer, an internal SLA covers teams in one company, and a multilevel SLA combines several levels.

5. What happens if an SLA is breached?

   Usually the customer can claim a service credit. Many SLAs require the customer to request it within a set time, and repeated breaches may allow the customer to end the contract.

6. How much downtime does 99.9% uptime allow?

   99.9% uptime allows about 43 minutes of downtime in a 30 day month. That is 0.1% of 43,200 minutes.

1. Google, Site Reliability Engineering, "Service Level Objectives." sre.google
2. Amazon Web Services, "Amazon Compute Service Level Agreement." aws.amazon.com/compute/sla
3. Vendr, "How to write a service-level agreement." vendr.com
4. Desk365, "The Ultimate Guide to Service Level Agreements." desk365.io
5. Alloy Software, "What is SLA (Service Level Agreement)?" alloysoftware.com
6. BMC Helix, "SLA template and examples." blogs.helixops.ai
7. Kelser, "What Is An IT Service Level Agreement (SLA)?" kelsercorp.com
8. DiliTrust, "Understanding SLA Agreements." dilitrust.com
9. Dropbox Sign, "Master service agreements." sign.dropbox.com