Service level agreements: the clauses that actually get tested.

← All insights

A service level agreement is read properly twice: once when it is signed, and once when something has gone wrong. Most SLAs are drafted for the first reading — optimistic, relational, full of goodwill. The second reading is the one that matters, because by then the goodwill is spent and the document is all that is left.

SLAs sit everywhere in modern organisations: IT and managed services, facilities and maintenance, outsourced payroll and administration, shared services between group companies, and the service arms of public and statutory bodies. Their job is simple to state and hard to draft: to convert a promise of service into measurable obligations with defined consequences. Five disciplines do most of the work.

1. Define the service before you define the level.

No metric can rescue a vague scope. Before a single service level is written, the agreement must say — usually in a schedule, and in operational detail — what the service actually is: the deliverables, the boundaries, the exclusions, the assumptions, and the things the provider is entitled to expect from the client. The schedule is not an annexure to the deal; for most practical purposes it is the deal.

Just as important is a working change-control clause. Services drift — volumes grow, systems change, small favours become standing expectations. Change control turns that drift into something recorded and agreed rather than something argued about eighteen months later, when nobody can say when the extra work became “part of the service”.

2. A level only counts if it can be measured — and someone owns the measuring.

A service level is a promise expressed as a number: availability percentages, response and resolution times, turnaround periods, accuracy rates. For each one, the agreement should answer four questions. How is it measured? From what data source? Over what window — per incident, per month, per quarter? And who reports it, to whom, how often? A metric without a measurement method is not a service level; it is a hope.

Two traps recur. The first is the pseudo-metric — “promptly”, “as soon as reasonably possible”, “best endeavours” — which reads like a standard but measures nothing. The second is conflating response with resolution: a provider who answers the phone in fifteen minutes has met a response target while the system stays down for a week. Draft both, and be honest about which one the client actually cares about.

A metric without a measurement method is not a service level; it is a hope.

3. Say — in advance — what happens when a level is missed.

An SLA that specifies targets but not consequences merely relocates the dispute. The usual instruments are an escalation ladder (who is engaged, at what seniority, how quickly), service credits (pre-agreed reductions of the periodic charge for defined failures), and a persistent-failure trigger that converts repeated or sustained breaches into a right to terminate.

Service credits deserve careful calibration. South African law will generally enforce agreed penalty and liquidated-damages stipulations under the Conventional Penalties Act 15 of 1962 — but a court may reduce a penalty that is out of proportion to the prejudice actually suffered. Credits that are calibrated to the real impact of a failure are therefore not just fairer; they are more enforceable. The agreement should also say expressly whether credits are the sole remedy for a service failure or cumulative with other remedies — silence on that point is a dispute waiting for a venue.

4. Disputes are decided in the machinery, not the metrics.

When SLA disputes reach advisers, they rarely turn on whether 99.5% availability was achieved. They turn on the unglamorous clauses around the metrics. Client dependencies and relief events: a provider cannot perform if the client withholds access, information or decisions — and the agreement should say what happens to the service levels when that occurs. Exclusions: planned maintenance, force majeure and third-party failures need clear boundaries, or every outage becomes an argument about categorisation. Liability: the caps and carve-outs in the master agreement must be read against the credits regime, so the two do not contradict each other. And where the service involves processing personal information, the operator provisions of POPIA require written terms and appropriate security safeguards — an SLA that is silent on this is a compliance gap, not a drafting style.

Finally, the exit. Termination assistance and transition clauses — handover of data, knowledge and in-flight work — are negotiated cheaply at the start of a relationship and expensively at its end. Draft them at the start.

5. Keep the agreement alive.

An SLA is a living instrument or a dead letter; there is little in between. The mechanisms that keep it alive are ordinary and decisive: a governance forum that actually meets and minutes; reporting that is read, not filed; a scheduled review of the service levels against changing needs; and disciplined version control of the schedules, so that everyone is performing against the same document. An agreement no one revisits will quietly part company with the service being delivered — and the gap between them is where disputes are born.

The test of a good SLA

The best compliment a service level agreement can receive is that a crisis was boring: everyone knew what had been promised, how the shortfall was measured, what consequence followed, and who had to do what next. That is what the second reading should find — and it is entirely a product of how the document was drafted at the first.

This piece is general commentary on drafting practice. It is not legal advice on any specific matter, and no professional relationship arises from reading it.