SLA Breaches

In the SRE distinction between indicators, objectives and agreements, an SLA breach is a failure to meet a service level specified in a contract. A late event and a breach of an aggregate target are distinct counts; the contract must state which triggers a remedy.

One breach or eighty

Take a hypothetical agreement: the contract promises 95% of orders within 48 hours, and this month 920 of 1,000 arrived on time.

The on-time share is 920/1,000 = 92%, below 95%. There are 80 late orders, one failed monthly target, and 30 late orders beyond the allowance of 50. All are useful operational counts, but only the contract determines which is a breach and how a credit is calculated.

For the stated monthly 95% target, up to 50 of 1,000 orders may be late without missing the target. A separate per-order remedy would need its own clause. Multiplying a credit by all 80 late orders cannot be inferred from the monthly target alone.

The window and the exclusions decide the verdict

For an illustrative time-based availability commitment, assume one unplanned 110-minute outage and no other unplanned outage across a 90-day quarter. Over the affected 30-day month, availability is 1 − 110/43,200 ≈ 99.7454%; over the quarter, it is 1 − 110/129,600 ≈ 99.9151%. The first misses a 99.9% target and the second meets it, because they assess different periods.

Now suppose the quarter also contains 180 minutes of planned maintenance, separate from the outage. If maintenance counts as downtime, availability is 1 − 290/129,600 ≈ 99.7762%. If it is non-penalized while the full denominator is retained, the result remains 99.9151%. If those minutes are removed from eligible time, the denominator becomes 129,420 and the result is 1 − 110/129,420 ≈ 99.9150%. An exclusion must specify both numerator and denominator treatment; the word “excluded” alone is not enough.

Which is why a breach rate is only interpretable next to its window, and why moving from monthly to quarterly reporting is a change to the assessed obligation if the agreement’s measurement window is changed.

A credit and business loss are different quantities

Assume a hypothetical clause grants a credit of 10% of a 40,000 monthly fee when the monthly target is missed. That credit is 4,000. If each of 80 late orders were canceled, and each cancellation lost 180 × 35% = 63 of contribution margin, the modeled loss would be 5,040. The credit would cover about 79.4% of that assumed loss. At 400 such cancellations it would cover about 15.9%. These assumptions illustrate the distinction; delay alone does not establish cancellation or a particular loss.

The credit amount is fixed under this example’s clause, while customer losses can exceed it. That does not cap the provider’s total incentive: churn, reputation, termination rights and other liabilities can also matter. Read the remedy and its limits separately from the performance target.

Operationally, preserve the measured events, applicable agreement version, eligible population, exclusions, calculation, and incident timeline. Investigate and restore service while the contractual assessment proceeds; an urgent service problem need not wait for the reporting window to close. Check remedy conditions and any repeat-failure terms rather than assuming a second miss always triggers the same consequence. The dashboard reports the obligation unit, measurement window, exclusion treatment, and remedy conditions defined by the agreement.

References

• Google SRE: Service Level Objectives
• Numerical scenarios, chosen rates and simulations in this entry are illustrative calculations by the author, not measurements reported by these sources.