Articles

Service contracts and SLAs: what is actually promised

A support contract promises who answers, how fast, and what counts as fixed. The hours, exclusions and escalation it omits are where money goes.

By Anes Hamdaoui6 min read
  • support
  • contracts
  • operations
All articles

A service contract promises three things whether or not it says so: who answers, how fast, and what counts as working again. The parts that decide whether it survives a bad week — the calendar it is measured against, the failures it excludes, and who to call when the named contact does not pick up — are usually left out.

What does a clause need before it means anything?

Four things: a trigger, a clock, evidence, and a remedy. Without all four it is a sentence of intent.

The trigger says what starts the clock, and it must be something the customer can start — a message to a monitored address, not a call to a number that rings in an empty office. The clock says which hours count and when it stops. Evidence is the timestamped record both sides can read. The remedy is what happens when the clock runs out, and "we will try harder" is not one.

Clause What it must state The argument it prevents
Response Who acknowledges, in which hours "Someone replied"
Restoration What counts as working again Fix versus workaround
Exclusion Who proves it, and by when A week of silence
Remedy What the missed clock buys A discount called a remedy

Do support hours follow the calendar or the working week?

The working week, unless the contract says otherwise, and this is where most promises quietly shrink. Eight hours of cover inside a six-hour working day is a promise that expires before lunch, and a Friday afternoon incident waits until Monday.

Name the calendar, not the duration: working days and their exact hours, the time zone, public holidays in both countries when the team is elsewhere, and reduced hours during Ramadan if that is how the office actually runs. Then say what happens outside it — nothing, a best-effort number, or a paid out-of-hours rate. All three are acceptable; silence is read as twenty-four hours by the buyer and as nothing by the vendor.

What is a response supposed to produce?

A named person who owns the problem, and a next action with a time on it. Not an automated ticket number, and not "we are looking into it".

The distinction matters more as the system gets older and the queue gets longer. A reply that takes four hours but says "Yacine has this, the fault is in the invoice import, we call you at 14:00 with a fix or a workaround" is worth more than a five-minute acknowledgement followed by two silent days. Where the contract covers a public site, the same rule applies to speed: the number that means something is measured on real visits, not on a laptop in the office, which is why Core Web Vitals belong in the clause instead of the word "fast".

Which failures are excluded, and who proves it?

Exclusions are legitimate — a customer's own network, a third-party service that went down, a change made by another supplier, a machine that was switched off. What is not legitimate is an exclusion invoked after four days of silence.

Every exclusion needs an owner and a deadline: who checks it, within how many hours, and what the customer is told while the check runs. Otherwise the exclusion becomes a way to stop the clock rather than describe reality, and the buyer learns to read the contract as a list of excuses they will hear later.

What happens when the escalation path is not written?

Nothing happens, which is the problem. Every contract names an account manager; almost none names the person who can authorise an emergency change at 22:00, or the person above the account manager who answers when the first two are unreachable.

Write two levels at least: who takes the call, and who can decide — release an urgent fix, call in a second engineer, accept a workaround in writing. Put a number and an out-of-hours rule next to each name. The escalation nobody wrote is the one everyone improvises, and improvisation at that hour is expensive and usually wrong.

Are service credits a remedy?

They are a discount, applied after the fact, and measured against an invoice rather than a business. A credit does not reopen a shop that lost a day of sales, and it is often the only remedy the contract offers — which caps what the failure costs the vendor and leaves the loss with you.

Read the remedy clause as a whole: whether credits are the sole remedy, whether they are capped, and whether claiming one needs a report within a stated period. Then ask what the vendor does instead when the same failure happens three months running.

How do you negotiate one?

  1. List the three failures that would stop your business that day. Those are critical; nothing else is.
  2. For each level, write the trigger, the response, the restoration target, and who is told.
  3. Fix the calendar in writing: working days, hours, time zone, holidays, and the out-of-hours rule.
  4. Name the exclusions, each with the person who checks it and the hours allowed to say so.
  5. Write the escalation ladder with names and out-of-hours numbers, and update it when people leave.
  6. Decide what a missed target actually changes, and whether credits are the only answer.
  7. Get the monthly report in the contract: incidents, clock times, open items, hours used.
  8. Review after three months against the timestamps, not against impressions, and renegotiate once.

Where this goes wrong

  • The clock starts when the vendor reads the message. A monitored inbox is a clause; a phone that rings out is not.
  • Uptime is promised and response is not. The system is up and nobody can log in, so the promise is kept and useless.
  • Holidays are counted differently on each side. The vendor's team is off on a day the buyer works, and the gap is nobody's fault and everybody's cost.
  • The clock pauses silently. "Waiting on the customer" becomes the default state, entered by the vendor with no timestamp.
  • Escalation goes to the salesperson. The account manager is on leave, the number is a personal mobile, and the incident waits.
  • The credit is claimed and the cause is not fixed. The invoice is adjusted, the outage returns, and the second credit is smaller.

What this does not cover

It quotes no durations and no rates: those follow the market, the system and how well it is documented. It does not cover the legal enforceability of a clause, which depends on your jurisdiction and your lawyer. And it assumes a vendor who wants the system to work — a contract is a poor defence against a supplier who has already decided to leave.

We build support contracts with the clock, the calendar, the exclusions and the escalation ladder in one document, because a promise nobody timestamps is a discount waiting to be argued.

Services in this article

Start a project

Have something to build?

One short brief is enough. You get a written scope and a fixed quote before anything starts.