Setting up SLAs

Measure before you promise

Look at what you actually do now – median and worst case, over a normal month. Almost every team guesses too optimistically, because they remember the good days and the bad ones happened to somebody else.

Set the first target at roughly your current worst case. You can tighten it later; you cannot un-promise it.

Two or three classes, not seven

A workable starting point:

• Standard – everyone without a specific agreement.
• Priority – customers whose contract says so.
• Critical – production down, defined narrowly enough that it stays rare.

Every additional class is a decision somebody has to make on every ticket.

Response first, resolution later

Start with a first-response target only. It is entirely within your control – it depends on whether someone looked, not on how hard the problem is.

Resolution targets should come later, tied to priority, and only once you know from data what is achievable. The distinction is explained in what is an SLA.

Business hours, written down

Define the hours the clock runs and make sure the system counts the same way. Otherwise every Friday evening produces a breach nobody could have prevented, and the numbers stop meaning anything.

What happens on a breach

Decide this in advance: who is notified, what changes, and whether the customer is told. An SLA without an agreed consequence is a number in a document. See escalation management.

Reviewing them

Once a quarter, look at the rate rather than individual breaches. A rate of 100 per cent usually means the target is too loose to be informative; below 90 means either the target or the staffing needs changing, and it is worth knowing which.