How to evaluate a managed-application SLA
Compare application SLAs with an editable worksheet for availability, response clocks, recovery objectives, exclusions and service credits.
Last updated
In short
- Availability, first response and recovery time are different commitments.
- Compare the measured service boundary before comparing percentages.
- Credits and exclusions belong in the evaluation alongside the headline target.
On this page (7)
Procurement and evaluation · Part of Sla
Availability boundaries, response clocks, exclusions, remedies and supplier commitment fields.
Download the SLA evaluation worksheetCompare what the promise measures
A managed-application SLA should be evaluated against the workflow the organization needs. A virtual machine can remain available while an application cannot authenticate users. An engineer can respond quickly while recovery takes much longer. A backup can exist while the acceptable data-loss objective is missed.
The downloadable worksheet separates these promises. It includes an arithmetic example and blank supplier fields, not a legal opinion or a claim about actual outages. Have procurement and the operational owner review the applicable contract together; ask qualified counsel to resolve interpretation questions.
Read five commitments separately
| Commitment | Question to ask | Evidence or clause to locate |
|---|---|---|
| Availability | Which operation must succeed, from where and how often? | Measurement definition and service boundary |
| Initial response | Does the clock end at acknowledgement or qualified engineer action? | Severity table and support hours |
| Restoration | Is there an objective for usable service after an incident? | Recovery procedure and agreed start/end events |
| Recoverable data | How much recent work may be lost? | RPO, backup cadence and consistency scope |
| Remedy | What happens if the commitment is missed? | Credits, claim process, caps and exclusions |
Pilae’s SLA is the primary source for its standard commitments. AWS’s compute SLA is a useful example of an infrastructure-level agreement with its own definitions and claim conditions. They cover different services, so their headline numbers should not be placed side by side as if they promise the same application outcome.
Translate percentages without changing their meaning
For a hypothetical 30-day month, there are 43,200 minutes. If every minute counts and the target is 99.9%, the arithmetic allowance is 43,200 × 0.001 = 43.2 minutes. At 99.5%, it is 216 minutes. These are calculations, not predictions of outage duration or statements that a provider may freely cause that downtime.
The actual contract can define its denominator, exclusions, rounding and measurement intervals differently. A planned maintenance exclusion may materially change the experience of users who work during that window. Ask suppliers to calculate the same hypothetical incident using their own terms, showing the eligible minutes and remedy.
Test the clock with a scenario
Describe an application outage beginning outside normal office hours. A customer opens a ticket, the monitoring system raises an alert, an engineer investigates, and the service becomes usable later. Ask which timestamp starts each commitment and whether the selected plan covers that period.
Then introduce a missing customer approval or an unavailable third-party identity provider. Determine whether the clock pauses, the incident is excluded or another team takes responsibility. A well-defined handover is more valuable than an apparently strict response target that no one can interpret during an incident.
Record customer obligations: supported configurations, access for the operator, incident reporting channels, maintenance approval and responsibility for plugins. These conditions should be feasible for the customer’s staffing and change process.
Assess recovery independently
Availability monitoring does not prove restoration from corrupted or deleted data. Ask for the recovery scope, dependencies, backup consistency and a representative exercise. NIST’s contingency-planning guidance provides background for structuring this assessment.
A recovery test should include identity, attachments, secrets and a business acceptance step. If only a database was restored, record that limitation. Confirm who decides to restore, what happens to transactions after the selected recovery point and how users are informed.
Examine remedies and operational incentives
Identify the credit base: the affected service fee, a larger subscription or another amount. Record claim deadlines, evidence required, maximum credits and whether the customer must request them. Service credits generally do not make a business whole for lost productivity; their exact legal effect belongs to the contract.
Also evaluate reporting and review. Will the provider explain causes, corrective actions and recurring failures? Can the customer export availability history and incident records? Does the agreement support a practical change if repeated incidents make the service unsuitable?
Choose the commitment that fits the workload
A small internal tool may be well served by business-hours support with a documented recovery procedure. A critical multi-site operation may require different staffing, architecture and negotiated objectives. Buying a stricter number without funding the dependencies can create false confidence.
Use the worksheet to compare an internal operating team, another managed provider and Pilae on the same workload. Include infrastructure, retained customer duties and the cost of the required support scope. Bring the unresolved clauses and scenario calculations to a service review before selecting a plan from current pricing.
Sources and further reading
Use the linked documentation to confirm the version, edition and contract that apply to your project.
