MMOOZZGG.com

Glossary

What is a service level agreement (SLA)

In short

A service level agreement (SLA) is an agreement between a supplier and a customer about how fast and how reliably a service must work, and what happens if the promise is broken.

Business and adoption · service level agreement

MMOOZZGG.com editorsChecked: 1 min read

For a child

You made a deal with a friend: every morning at eight they come by, and you walk to school together. If your friend is more than five minutes late, they carry your school bag all the way to class. This is your small rule: you both know what to expect and what happens if the promise is broken. Companies have the same kind of deals.

In plain words

When a company buys a service, such as a cloud service or technical support, it cares not only about what is done but also how. A service level agreement turns this into numbers: the service is available a set share of time each month, an urgent request gets an answer within an hour and a fix within a day. Both sides measure these numbers the same way. If the supplier misses them, the client gets a discount or compensation. So the promise “we work reliably” becomes a contract term you can check.

For business

For AI products, this kind of agreement is needed on two sides. Outside, the cloud supplier of the model promises API availability. Check whether the promise covers response time and quality, and whether a failure is compensated with money or credits for future requests. Inside, the contractor who supports the agent is responsible for incident response time, recovery and knowledge base updates. Look at what exactly is measured, how downtime is counted, what exceptions sit in the small print, and whether there is a backup model if the main one fails. The clients the agent serves also need their own number, such as the share of requests answered within a minute. The stricter the promises, the more you pay for backups and on-call duty. The risks of depending on one supplier are covered separately.

For a specialist

A service level agreement (SLA) is an explicit or implicit contract with users that sets the consequences of meeting or missing service level targets. Most often the consequences are financial: a discount or a penalty. The Google book “Site Reliability Engineering” (chapter “Service Level Objectives”, Jones, Wilkes, Murphy, Smith) defines a service level indicator (SLI) as a quantitative measure, such as the share of successful requests or latency. A service level objective (SLO) is a target value or range for an SLI. The internal goal is set stricter than the promise to users to leave room for a response, and the allowed share of misses against the SLO is the error budget. For LLM services, availability is joined by time to first token, generation speed, the share of refusals caused by rate limits, and answer quality metrics that are hard to put into a contract. For models inside the company’s own network, the same numbers go into an internal agreement between IT and the business.

Example

A company connected an AI support agent through a cloud model. The contract with the contractor says: the agent is available around the clock, a failure is fixed within four hours, and if the model fails, requests go to a backup model automatically.

Related terms

Questions and answers

How is an SLA different from an SLO? An SLO is the team’s internal goal, such as the share of successful requests in a month. An SLA is a promise to the client in the contract, with consequences for a breach, so it is usually set softer than the internal goal.
What should you check in an AI supplier’s SLA? What counts as downtime, how availability and latency are measured, what exceptions the contract has, and what compensation you get. Also ask about request limits and how the supplier announces that a model will be switched off.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us