An SLA is not a promise that software will never fail. It is an agreement about what happens when it does: who responds, how quickly, what counts as resolved, and what the vendor owes you if the commitment is missed. Two contracts can both quote the same availability percentage and give you very different protection.
This comparison sets out five SLA structures a business is likely to encounter. The software SLA cost guide covers pricing; this page covers what each structure actually commits the vendor to.
The five models at a glance
| Model | What it commits to | Best fit | Strength | Weakness |
|---|---|---|---|---|
| Tiered support | Support hours and response targets by package | Business software with predictable support needs | Simple to compare and budget | Response is not resolution |
| Uptime SLA with credits | A monthly availability percentage and service credits | Hosted platforms where availability is the main risk | Measurable and automatic | Credits are usually small; exclusions are broad |
| Severity matrix | Response and resolution clocks by severity level | Operations with a mix of minor and critical issues | Forces prioritisation | Severity is often decided by the vendor |
| Outcome commitment | A business outcome such as order processing availability | Revenue-critical workflows | Closer to what the business cares about | Hard to measure and easy to dispute |
| Managed retainer | Support plus a monthly improvement capacity | Businesses that need ongoing change, not only fixes | Prevents the system from standing still | Capacity can be consumed by small requests |
Criteria that separate a real SLA from decoration
Coverage is the first test. Does the SLA cover the application, the hosting, the integrations and the third-party services the application depends on, or only the vendor's own code? Clock definitions are the second. When does the clock start: when the ticket is opened, when it is acknowledged, or when the vendor begins work? Does it pause while waiting for your team? Severity ownership is the third. If the vendor alone decides whether an issue is critical, the response target is not a commitment.
Then look at the remedy. A service credit is a refund on fees, not compensation for lost sales. Check the cap, the claim process and whether credits are automatic or require a written request within a short window. Finally, look at reporting. A monthly report with ticket volumes, response times, resolution times and breaches is the difference between an SLA and a clause nobody reads.
Response, restoration and resolution
These are three different clocks and contracts often blur them. Response is the time to a human acknowledgement. Restoration is the time to bring the service back, sometimes through a workaround. Resolution is the time to fix the underlying defect. A vendor can be excellent at response and poor at resolution. For each severity level, write down which clock applies and what evidence proves it stopped.
Consider a checkout outage on a Friday evening. A 30-minute response target is useful only if the person who responds can actually restart the service. A four-hour resolution target may be unrealistic if the cause sits with a payment provider. The SLA should distinguish what the vendor controls from what it can only escalate.
Service credits: what they do and do not cover
Credits are usually calculated as a percentage of the monthly fee and capped at that fee. For a business losing orders during an outage, the credit rarely matches the loss. That does not make credits useless: they create an incentive and a visible breach record. But they are not insurance. If downtime is genuinely existential, ask about business interruption cover or a contractual remedy that reflects the impact, and expect to pay for it.
Exclusions to read closely
- Scheduled maintenance and emergency maintenance windows, including how much notice is given.
- Third-party failures such as hosting, payment, messaging or identity providers.
- Customer changes to configuration, infrastructure or integrations outside the agreed change process.
- Unsupported versions and end-of-life components the vendor has asked you to upgrade.
- Force majeure, which should be narrow enough to mean something.
- Beta features and customisations excluded from the standard commitment.
Choose the structure that matches the risk
- Choose tiered support when the main need is access to help during business hours and the system is not revenue-critical.
- Choose an uptime SLA when a hosted platform carries the availability risk and you can verify the measurement source.
- Choose a severity matrix when the mix of issues is wide and prioritisation is the real dispute.
- Choose an outcome commitment only when both sides can measure the outcome from the same data.
- Choose a managed retainer when you need continuous improvement and want support and change capacity in one agreement.
Testing whether the SLA is real
Ask for three months of reports before signing, not a sample template. Pick two tickets and trace them from open to close against the clock definitions. Ask what happened the last time a breach occurred and who received the credit. During the first quarter, review every breach with the vendor and record the cause. An SLA that is never reviewed is a filing cabinet item; one that is reviewed monthly changes behaviour.
Where this fits in our work
Support and maintenance arrangements sit inside Custom Software & Integrations, alongside application ownership and handover. The pricing page explains how support hours and change capacity are separated. For the cost side, read the software SLA cost guide for business owners.
Sources
- Google, Site Reliability Engineering: Service Level Objectives, accessed 17 November 2026.
- Atlassian, SLA vs SLO vs SLI, accessed 17 November 2026.
- AXELOS, ITIL Best Practice, accessed 17 November 2026.
- Gartner Glossary, Service-Level Agreement, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.