SLA Breach Rate

SLA breach rate is the share of service commitments that were missed.

\[ \text{breach rate} = \frac{\text{breaches}}{\text{obligations}} \]

As the SRE discussion of service objectives explains, the indicator and target must be defined. The denominator and measurement window determine which commitment this fraction describes.

What counts as one obligation

The denominator is described as total SLA obligations, which sounds like a count and is actually a contract reading.

Suppose two customers have 900 and 100 eligible tickets in a month, with 36 and 20 late tickets. Across tickets, the miss rate is 56/1,000 = 5.6%. If each customer-month instead has a requirement of at least 95% on-time tickets, the first customer passes at 96% and the second fails at 80%: one of two customer-month obligations is missed, or 50%. A single overall monthly 95% requirement would also fail, giving one failed aggregate obligation out of one. These are different definitions, not conflicting calculations.

A breach rate without its denominator named is not checkable, and the denominator belongs in the agreement rather than in the reporting pipeline. What counts as a breach is the same question one level down.

For a response-time commitment, define when the clock starts and stops, whether business hours or pauses count, and which window owns a ticket crossing month-end. A still-open ticket whose deadline has passed must not vanish merely because a query counts only completed tickets. Deduplicate obligations and distinguish exclusions from missing records. With no eligible obligations, the rate is undefined or not applicable, rather than zero by default. If logs are incomplete, report that limitation instead of treating unobserved obligations as successful.

Different incident patterns, one monthly rate

Breaches can cluster around outages, releases or staffing gaps. An independent-draw model is useful for a sampling baseline, but does not describe all operational incidents.

Consider a month with 30 days and 2,000 eligible tickets each day. Three thousand late tickets produce a 5% monthly rate. They could occur as 100 late tickets every day, or as 1,000 late tickets on each of three days and none on the other days. The latter has 50% misses on those days, despite the identical monthly figure. A daily alert above 10% catches the burst but not the steady pattern. This illustrative alert is separate from the monthly contractual rule.

A monthly rate cannot distinguish these patterns. Report important sub-windows and trends beside it when the timing of failures matters.

Sampling variation at low volume

Supplier scorecards rank vendors on breach rate, and vendors differ in size by orders of magnitude.

With 25 eligible orders, one late order changes the reported rate by four percentage points; with 2,000 orders it changes it by 0.05 percentage points. Under an illustrative independent Bernoulli model with true miss probability 5%, the standard deviation of an observed proportion is about 4.36 percentage points at n = 25 and 0.49 at n = 2,000. Identical underlying probabilities can produce noticeably different observed rates. This sampling comparison does not cancel or replace a contract’s rule for actual observed events.

A guardrail threshold needs an uncertainty model as well as a denominator. Under independent Bernoulli events with probability p, the sampling standard deviation of a proportion is sqrt(p(1−p)/n). Real outages may violate that independence assumption; longer pooling also risks hiding a changing rate.

A usable report states the obligation, window, eligible volume, exclusions, and relevant segment or sub-window results. It preserves source timestamps and the rule used to classify each obligation, so disagreements can be investigated. Precise reporting improves traceability; it cannot make a disputed contract interpretation disappear.

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.