SLA · Service Level Agreement
Service Level Agreement
What is an SLA?
- A contract clause that guarantees service quality in numbers
- 99.9% allows about 43 minutes of downtime a month, and each extra 9 adds major cost
- The measurement method, exclusions, and claim process matter more than the headline number
Once an SLA is in the contract, the conversation about outages changes character. Instead of 'it was force majeure this time' traded against 'we still took a loss,' you get a settlement conversation: cumulative downtime last month, minutes over the baseline, credit percentage. Because it is a device that turns an emotional issue into a calculation, the vendor’s sales rep arrives with an incident report and numbers instead of an apology, and the procurement manager can explain to internal audit and executives, 'this is how much we are protected.' Quality that isn’t written into the contract is a favor, not a service, and favors are the first thing to disappear as an outage drags on.
In practice, you need to keep three layers apart. An SLI is what gets measured (uptime, first-response time for P1 incidents), an SLO is the internal target the provider aims to hit, and an SLA is the contract under which money changes hands when the target is missed. Mature providers set the SLO one notch tighter than the SLA and use that margin to avoid paying out credits. The terms are also granular. First-response and final-resolution times are written separately for each severity level (P1–P4); the same uptime figure can produce different results depending on whether it is calculated as a monthly total or as cumulative 5-minute samples; and credits are typically capped at 10–30% of the monthly fee. If the provider subcontracts its infrastructure, whether the same standard is passed down back-to-back in the subcontract determines whether the commitment can actually be met.
The most common misconception is treating credits as a penalty. Many standard contracts make credits the 'sole and exclusive remedy,' so even if an 8-hour outage causes hundreds of millions of won in lost business, all you get back is a portion of the monthly fee. Add a claim-based clause such as 'written request within 30 days of the event,' and missed filing deadlines are common. A definition of uptime that counts only 'complete outages' is also risky: if every query takes 10 seconds and the business is effectively paralyzed, reported uptime still shows 100%. The risk runs the other way for providers, too. If sales promises 99.99% in its eagerness to win the deal while operations has no overnight on-call staff, credits start eating into margin from the first quarter.
John KimAuthor of “B2B Marketing Design” and CEO of Nanum Consulting
Sangmi LeeInstructor specializing in CS, leadership, and communicationIt began in the 1980s, when telecom carriers wrote quality guarantees for enterprise circuit services into their contracts. With the spread of IT outsourcing and ITIL (the IT service management standard), it became a standard contract element in the cloud and SaaS era.
Consider a hypothetical case. Company A makes electronic components, with 1,200 employees and annual revenue of about KRW 320 billion (≈ US$239,413,400). Two months before renewing a production-management SaaS adopted three years earlier, the company flagged a problem: four outages in the previous year had stopped its lines for a total of 11 hours. Yet the vendor's annual report stated '99.8% uptime, no SLA violations.'
- When Mr. Kim from procurement cross-checked the outage logs against the vendor report, he found that six hours of scheduled overnight maintenance had been left out of the measurement entirely. The measurement exclusion clause was quietly eroding the coverage they thought they had.
- Ms. Lee from IT Operations said, "If the line stops for one hour, we scrap KRW 40 million (≈ US$29,900) in materials alone," and asked for severity levels to be broken out, with P1 incidents requiring a first response within 15 minutes and recovery within 4 hours.
- The vendor's sales director held out, saying 99.95% was not viable at their pricing. The two sides settled on a 99.9% target, changing the measurement basis from a monthly total to cumulative 5-minute samples and capping excluded scheduled maintenance at 2 hours a month.
- Legal review caught wording that made credits the sole remedy. The team added a clause allowing penalty-free termination after two consecutive months, or three misses in a year, plus conditions for applying credits automatically.
- Six months after renewal, actual downtime fell from 11 hours to 2 hours 40 minutes and average P1 first response dropped from 52 minutes to 12 minutes. The KRW 19 million (≈ US$14,200) in credits for the two months that fell short was deducted from the invoice automatically, with no claim process.