A service level agreement exists for one thing: so that client and provider know, without arguing, when the service is fine and when it is not. Most of them fail even at that. They get copied from a template, filled with numbers that sound demanding, signed, and three months later nobody knows whether the committed 99% is being measured on the same thing both sides thought it was.
A badly written SLA is not neutral. It is worse than having none, because it creates a false sense of control and turns every month into an argument about the wording instead of about the service.
The original mistake: signing a target with no baseline
The question almost never asked before signing is simple: is this number achievable with the operation we are contracting? A 30-second response-time SLA sounds fine until someone works out how many agents it takes to sustain it at peak hour. If the number was set because it sounds serious rather than because it came out of a sizing exercise, it is a promise that will be broken with a signature attached.
An achievable SLA is calculated, not wished for. It comes from the real demand curve, from team sizing and from a measured baseline, not from what the client would want in an ideal world. That is why SLAs that work are agreed after the diagnosis, when there is volume data, and not in the first sales meeting.
The five gaps that make an SLA useless
1. Not defining the unit or the moment of measurement
"Response time under two minutes" says nothing until it is clarified: response to what, measured from when, until what event? Does the automated bot count as a response? Does the clock run outside service hours? Two identical operations can report opposite figures on the same service purely because of where the stopwatch starts and stops.
2. Measuring on averages instead of percentiles
An average hides the tail. An average response time of one minute is compatible with one in ten users waiting half an hour, and those are exactly the ones who complain. Committing to the p90 —the value 90% of cases do not exceed— describes the real experience far better than the mean. It is one of the principles in what to measure in a BPO operation from month one.
3. Not excluding what the provider does not control
No provider can answer for the time the client's CRM is down, for an unannounced volume spike, or for cases that depend on a third party. An SLA without well-defined exclusion clauses breaks for reasons outside the service and contaminates the measurement. Exclusions are not an escape hatch: they are the condition for the committed number to mean anything.
4. Confusing a metric with a business objective
A 99.9% availability SLA is worthless if what the client cares about is resolving cases, not keeping the line open. Every metric in the agreement should be able to answer the question "and what does this decide?". If you commit ten metrics and eight change no decision, you have a decorative dashboard with contractual force.
5. Not saying what happens on a miss
A target with no consequence is an aspiration. The SLA has to say what happens in a month of non-compliance: remediation plan, escalation, penalty, whatever the mechanism is. But the penalty is not the point; the recovery plan is. A contract that only punishes and does not compel a fix aligns both sides on arguing about the fine, not on fixing the service.
Few metrics, well chosen
The temptation is to commit everything that can be measured. The result is a twenty-line annex nobody reviews. A good SLA commits three or four metrics that genuinely describe the service, each with its definition, its measurement method, its window and its exclusion. It is harder to draft and far easier to manage.
The rule of thumb: if you cannot explain in one sentence what decision a metric changes when it is missed, it does not belong in the SLA.
Who measures is part of the agreement
An SLA where the provider reports its own numbers with no sample audit does not measure performance: it measures trust. It does not mean the provider is lying; it means the definition and the instrumentation can drift without anyone noticing. The agreement should say which system the data comes from, how often it is reported, and how a discrepancy is resolved when the client's number and the provider's do not match. That clause looks bureaucratic until the first month the two numbers disagree.
An SLA is a management instrument, not a weapon
The best sign of a healthy SLA is that the penalty is almost never invoked, because the remediation plan kicks in before the miss escalates. An agreement used to fight every month has already failed as a management tool, even if it is technically in force. The goal is not to have something to punish the provider with; it is for both sides to look at the same dashboard and know what to do when a number goes out of range.
At smartBPO we agree SLAs after the diagnosis, once we know the demand curve and can calculate what is achievable instead of promising what sounds good. We would rather commit few metrics we can sustain than sign many that break within the quarter. An SLA both sides knew was unreachable on signing day protects no one; it just postpones the hard conversation by a few months.