Call center SLAs: writing service levels that mean something

An SLA is not a KPI list

The most common structural mistake is putting every metric you track into the contract. A KPI is something you measure to manage the operation. An SLA is a contractual commitment with a consequence attached to missing it. They should not be the same list, and the SLA list should be much shorter.

Over-stuffing an SLA causes real damage. It dilutes attention across a dozen commitments so the partner optimizes for whichever carries the largest penalty rather than for the outcome you actually care about. It creates administrative overhead on both sides that gets billed back to you eventually. And it makes disputes more likely, because more definitions mean more edges to argue about.

The useful discipline: three to six contractual SLAs, chosen because a miss on any of them genuinely damages your business. Everything else goes in a reporting schedule — measured, reviewed monthly, and managed through the relationship rather than through remedies.

The service level convention: 80/20 and why it isn't a standard

'80/20' — answering 80% of calls within 20 seconds — is the most cited target in the industry and it has no empirical basis. It's a convention that hardened into a default decades ago, and it gets copied into contracts by people who assume someone once proved it.

Nobody did. The right target depends on what the call is about and what the alternative costs. An emergency line, a fraud hotline, or a medical triage number may warrant 90/10 or better, and the cost is justified. A general billing enquiry line can often run 70/60 or 80/30 with no measurable satisfaction difference and materially lower staffing cost, because callers with non-urgent issues have a much longer tolerance than the convention assumes.

The economics deserve a moment of attention: because staffing curves are non-linear, moving from 80/20 to 90/10 costs far more than the ten-point difference suggests. Buyers routinely specify aggressive service levels reflexively, pay for them, and would not be able to detect the difference in customer outcomes. Before setting the number, ask what happens to your customer at 30 seconds versus 20 — for most queues, nothing.

Two guardrails belong alongside whatever target you pick. Specify the measurement interval, because a target met daily but missed in six consecutive half-hours is a failure customers experienced and the report didn't show. And pair service level with abandonment rate, since a service level can be met while abandonment climbs — the callers who gave up before the threshold left the denominator on their way out.

The metrics that belong in the contract

A defensible SLA usually draws from a small set, weighted toward outcomes rather than activity.

• Speed of answer — service level (X% within Y seconds) by queue, with the measurement interval named. Include equivalent targets for chat first-response and email or ticket response.
• Abandonment rate — the essential companion to service level. Define whether calls abandoned within the first few seconds — usually misdials — are excluded, and at what threshold.
• Quality score — against a rubric attached as an exhibit, with a defined sample size per agent per month and a calibration process both parties join. A quality SLA without an agreed rubric and sample size is unenforceable.
• Resolution — first-contact resolution or, better, repeat-contact rate within a defined window. Harder to define than speed metrics and worth the effort, because this is the one that actually describes whether customers were helped.
• Customer satisfaction — CSAT on a defined survey instrument, with response-rate floors so the number can't be produced from a handful of responses.
• Compliance and accuracy — for regulated work — disclosure adherence, consent handling, data-handling errors. Usually a zero-tolerance or near-zero threshold rather than a percentage target.
• Availability and continuity — systems uptime during contracted hours, and defined recovery times for a site or connectivity failure.

The metrics that shouldn't be SLAs

Some numbers are useful to watch and destructive to contract.

Average handle time is the clearest example. Making AHT a contractual commitment gives your partner a direct financial incentive to end conversations early, which raises repeat contacts and lowers resolution — you pay for the same customer twice and the SLA report looks excellent. Watch it; don't contract on it.

Occupancy and utilization belong to the partner's staffing model, not to you. Contracting them either constrains their ability to manage the operation or, if set high, incentivizes agent burnout that shows up later as attrition and quality decline.

Attrition rate is a legitimate reporting item and a poor SLA, because it's affected by labor market conditions neither party controls; it's better handled as a review trigger than a penalty. And raw contact volume handled is an activity measure that rewards taking more contacts rather than needing fewer.

Definitions are where SLAs actually get decided

The metric name is the easy part. Nearly every dispute turns on a measurement detail that seemed too obvious to write down.

Nail these before signature. What starts the clock — when the call arrives, when the IVR completes, or when it enters the queue? IVR time excluded from speed-of-answer is the most common way a service level flatters itself. What counts as answered — an agent connection, or a greeting? How are transfers counted: one contact or two? What's the exclusion list, and it should be genuinely narrow: force majeure, client-caused system outages, volume above a stated forecast tolerance, and client-requested changes. A partner asking for broad exclusions is asking to be paid for a commitment that evaporates when it matters.

Then the arithmetic. Is performance measured monthly, weekly, or by interval — and if monthly, is it a simple average across intervals or weighted by volume? Volume weighting is usually right and usually favors the partner, so decide deliberately. What's the minimum sample size below which a metric isn't scored? A quality score on four sampled calls is noise with a decimal point.

Finally, who measures. Reports produced solely by the partner from their own systems, with no audit right and no shared data access, means you are marking someone else's homework from a summary they wrote. Insist on source-data access or a defined audit right, and specify what happens when the two parties' numbers disagree.

Remedies that change behavior

Service credits are the standard remedy and, at typical magnitudes, they mostly function as a signal rather than a deterrent — a small percentage of monthly fees rarely outweighs the cost of the staffing that would have prevented the miss.

Structure them to escalate, so a chronic miss becomes materially more expensive than an isolated one. Distinguish acute failures (one bad month) from chronic ones (three consecutive months, or four in six), and attach different consequences: a credit for the first, a mandatory remediation plan with named actions and dates for the second, and termination rights without penalty for continued failure. The termination right is usually the term with real force — it changes the negotiation more than any credit percentage.

Two additions worth pushing for. A cure period with a written remediation plan is more useful than a credit, because it produces action rather than a refund. And an earn-back or bonus structure for sustained over-performance turns the relationship from purely punitive into something a partner will actively invest in — which, over a multi-year contract, tends to matter more than the penalty schedule.

Be realistic about magnitude. Credits large enough to make an account unprofitable produce a partner who assigns their weakest team to it or exits at renewal. The goal is a partner who wants to keep your business and finds it worth resourcing properly, not one who is trapped and resentful.

Reporting, governance, and the ramp

An SLA without a governance rhythm is a document nobody reads until there's a problem. Specify the cadence in the contract: a monthly operational review with a defined report package, a quarterly business review with commercial and strategic scope, and a named escalation path with roles and response times on both sides.

Define the report contents explicitly — performance against each SLA, the underlying volume and staffing data, root-cause analysis for any miss, and the status of open remediation items. Ad hoc reporting formats drift toward whatever makes the numbers look best.

Handle the ramp deliberately. New programs miss service levels during transition for legitimate reasons — agents are learning, forecasts are unproven, and knowledge transfer is incomplete. A contract that applies full SLAs and penalties from day one either gets quietly waived, which teaches both sides the terms are soft, or it poisons the relationship in month one. Write an explicit ramp period with reduced or reporting-only targets, a defined end date, and a stabilization review before full SLAs engage.

Then keep the agreement current. Contact mix changes, products launch, volumes shift, and an SLA written against a two-year-old operation gradually stops describing the business. Build in an annual review of targets and definitions, with a change mechanism that doesn't require renegotiating the whole contract.