Home/Insights/Infrastructure
Infrastructure · 15 Jul 2025

Designing managed service SLAs that mean something

Most support agreements specify response time and nothing that matters. Four clauses separate a real service level from a statement of intent.

Support agreements are among the easiest commercial documents to sign without reading carefully, because they all look similar and the differences only surface during an incident. A small number of clauses determine whether the agreement is a service level or a marketing statement.

Response time is the least important number

Almost every agreement specifies response time, and it is the metric least connected to outcome. A response is an acknowledgement. It does not mean anyone is working on the problem, and it certainly does not mean the problem will be fixed.

What matters is resolution — or where resolution cannot be guaranteed, a committed progression: escalation at defined intervals, with named seniority attached, until the issue is closed. An agreement with a two hour response and no resolution commitment permits a ticket to be acknowledged promptly and then sit.

The four clauses that matter

Severity definitions, written by the client. If the provider defines severity, the provider decides what is urgent. Definitions should describe business impact — this system down stops invoicing, this one inconveniences one team — and should be agreed before signature.

Resolution or progression targets per severity, not just response. Including what happens when the target is missed.

Service credits with actual weight. A credit regime is the only mechanism that makes a target consequential. Credits set low enough to be cheaper than performing are a cost of doing business, not an incentive.

Scope boundaries in writing. Which systems, which hours, which locations, and specifically what is excluded. Most disputes are scope disputes, and they are all preventable at signature.

What to ask for beyond the document

Reporting that shows performance against target per period, including misses. A provider unwilling to report their own misses is telling you something.

A named service manager and a defined escalation path with actual names, so escalation does not mean raising a second ticket.

A quarterly review with the authority to change the agreement. Estates change, criticality changes, and an SLA fixed for three years will be wrong within one.

The uncomfortable question

For your current support arrangement: what happens, contractually, if the provider misses their target on your most critical system? If the answer is nothing, the target is aspirational — which may be acceptable, but should be a decision rather than a discovery.

Is this a live question for you?

We are happy to talk it through — no proposal attached.