Writing a service level you can enforce

Most service levels in outsourcing contracts are unenforceable. Not because anyone acted in bad faith, but because the clause names a metric without pinning down how it is measured, over what window, with what exclusions, and what happens when it is missed. When performance slips, both parties then discover they had been reading the same sentence differently for a year.

An enforceable service level needs four things per metric: a definition, a measurement window, an exclusion list, and a remedy. Miss any one and you have an aspiration.

Answer time

Say what you mean by answered. The usual definition is the percentage of contacts answered by a live agent within N seconds, measured from the moment the contact enters the queue — not from the end of the IVR, and not from the start of the routing attempt. Write down which of those it is.

• Name the window. "80% in 20 seconds" measured monthly is a much weaker promise than the same figure measured per half-hour interval, because a catastrophic Monday morning disappears into a good month.
• State whether abandoned contacts count against it, and after how many seconds a short abandon is excluded. Five seconds is a misdial; ninety seconds is a failure.
• Decide whether out-of-hours and holiday intervals are excluded, and list the holidays. Two countries, two calendars.

Abandon rate

Abandon rate is the honest check on answer time, which is why it belongs in the contract alongside it and not instead of it. A queue can hit its answer-time target beautifully by letting the hard half of the volume hang up.

Define the short-abandon threshold explicitly, use the same measurement window as answer time, and require both to be reported on the same view. Reported separately, they can each look fine.

First-contact resolution

The hardest of the four to write and the most valuable to get right. You need to answer three questions in the clause itself:

• What counts as a repeat contact — same customer, same issue, within how many days? Seven days is common; whatever you choose, choose it in writing.
• Who decides whether the issue was the same? If it is the agent disposition alone, the metric measures disposition discipline rather than resolution.
• Which contact types are out of scope? Multi-step cases that legitimately require a follow-up should not be scored as failures.

Sample-audit it. FCR derived purely from self-reported dispositions drifts upward over time in every operation, including honest ones.

CSAT

Specify the question, the scale, the trigger, and the response-rate floor. A CSAT clause without a response-rate floor can be met by surveying only the customers most likely to be happy.

State who owns the survey tool. If the vendor sends the survey, sets the trigger, and reports the result, you are grading their homework with their pen.

The remedy clause

This is the part that gives the rest teeth, and the part most often left as "the parties will discuss".

• Tie a defined percentage of the monthly fee to defined misses. The figure matters less than its existence; a service credit changes what gets escalated internally on the vendor's side.
• Distinguish a miss from a breach. One bad month triggers a credit and a written remediation plan. Three consecutive months triggers a termination right. Without the second, the first is just a discount for poor service.
• Require root-cause analysis in writing within a fixed number of days of any miss, and make the remediation plan reviewable. The analysis is worth more than the credit.
• Write the excusable-event list narrowly and symmetrically. Your own outage, your own volume forecast being wrong by 40%, and a systems change you made without notice are all legitimate exclusions — and all should have to be evidenced, not asserted.

A last, practical rule

Do not sign a service level nobody has measured yet. Run the pilot, measure the metric as defined, and set the target from observed performance plus a deliberate stretch. Targets invented at contract stage are either too soft to matter or too hard to hit, and both outcomes end with a metric everyone quietly stops reading.