What is call center service level?

Call center service level is the percentage of incoming contacts answered within a defined time threshold, written as two numbers such as 80/20, meaning 80% of contacts answered within 20 seconds. It is the standard way contact centers set and report a responsiveness commitment.

Service level exists because an average is not a promise. Telling customers the typical wait is 24 seconds says nothing about how many of them waited three minutes. Service level fixes a threshold and reports how many contacts cleared it, which makes it the number most centers actually manage to.

This page covers how service level is expressed and calculated, where the 80/20 convention came from, the counting decisions that quietly change the result, what a good target looks like, how service level differs from a service level agreement, and how to hold it.

Call center service level in one sentence

Service level is the share of contacts answered inside the time you said you would answer them.

How service level is expressed

Service level is written as a pair: target percentage / threshold in seconds.

• 80/20 means 80% of contacts answered within 20 seconds

• 90/30 means 90% answered within 30 seconds

• 70/60 means 70% answered within 60 seconds

The two numbers move together. Loosening the threshold from 20 to 30 seconds makes the same operation look better without anything changing, which is why quoting a service level percentage without the threshold attached is meaningless. "We hit 85% service level" is not a statement anyone can evaluate.

How to calculate call center service level

Service level = contacts answered within the threshold ÷ total contacts offered × 100

A worked example. In a given period, 1,000 contacts are offered. 900 of them are answered within 20 seconds.

Service level = 900 ÷ 1,000 × 100 = 90%, reported as 90/20.

The counting decisions that change the answer

This is where service level gets slippery, and where two centers reporting the same figure can be describing very different operations.

How abandoned contacts are treated. There are at least four common conventions, and they produce materially different numbers from the same data:

The second convention is the dangerous one. Excluding abandons means the worst hours produce the best-looking numbers, exactly as with average speed of answer.

What counts as "offered". Contacts that hit an engaged tone, get blocked by a capacity limit, or are diverted by an overflow rule may or may not enter the denominator depending on the platform.

The reporting period. Service level calculated across a month smooths over a disastrous Monday morning. Calculated per half-hour interval, it shows exactly when the operation failed. The monthly number is for reporting; the interval number is for managing.

The channel. A seconds-based threshold makes sense for voice and chat. For email or messaging the equivalent is usually expressed in hours, and blending the two into one figure produces a number that describes nothing.

Write the convention down. Service level is only comparable against itself, and only if the definition holds still.

Where 80/20 came from

80/20 is treated across the industry as the default, and it is worth knowing that nobody has ever justified the numbers.

Operations researchers Ger Koole and Avishai Mandelbaum put it plainly in the Annals of Operations Research in 2002. The 80/20 rule is an apparent industry standard, they wrote, and they were "aware of no rationalized justification for these parameter values." Two decades later the position has not moved. A forecast analyst writing in Call Centre Helper in 2019 observed that after eleven years in resource planning, "no one seems to know where 80/20 came from."

That does not make it a bad target. It has the practical virtue of being widely understood, and a target everyone recognizes is easier to plan and staff against than a bespoke one. But it should be chosen rather than inherited, because the cost of the last few points is steep. Moving from 80/20 to 90/20 does not cost 12.5% more staff; because of how queueing behaves, it costs considerably more, and the customer-visible difference is smaller than the number suggests.

Some centers set their target lower on purpose. A 70/60 target with strong resolution and well-set expectations can produce happier customers at lower cost than 90/20 with rushed conversations. The right target is a deliberate trade, not a default.

Service level vs. average speed of answer

Both are calculated from queue data and they answer different questions.

Take a queue where half of contacts are answered instantly and half wait 48 seconds. Average speed of answer reports 24 seconds and looks fine. Service level at an 80/20 target reports 50% and correctly shows a problem.

Now take the reverse case. A center hitting 80/20 might be leaving the other 20% waiting eleven minutes, and service level alone will not show it. The two metrics cover each other's blind spots, which is why most operations report both.

Service level vs. service level agreement

These are different things with confusingly similar names.

Service level is an internal operating target. The center chooses it, staffs to it, and manages against it. Nobody is contractually bound by it and it can be changed at the next planning cycle.

A service level agreement (SLA) is a contract between two parties that specifies committed performance, how it is measured, and what happens when it is missed. An SLA frequently contains a service level target, along with remedies and penalties that a service level on its own never has.

The short version: a service level is a goal, an SLA is a promise with consequences. Outsourced operations usually have both, and the SLA target and the internal target are often deliberately different, with the internal one set tighter to leave headroom.

Why service level matters

It is a commitment rather than a description. Service level is the only standard queue metric expressed as a threshold customers could in principle be told about. That makes it the honest basis for a stated wait-time expectation.

It exposes the tail that averages hide. The customers who wait longest are the ones most likely to abandon, complain, and remember. Service level counts them; the average does not.

It is the anchor for capacity planning. Erlang-based staffing models take service level as the input. Change the target and the required headcount changes with it, which makes service level the point where a customer-experience decision becomes a budget decision.

Where service level misleads

It says nothing about the misses. A center at exactly 80/20 could be answering the remaining 20% at 25 seconds or at nine minutes. Track the abandonment rate and the longest wait alongside it.

The threshold can be moved. Loosening from 20 to 30 seconds improves the figure without improving anything. Any change to the threshold should be flagged loudly on the trend line, because otherwise it reads as performance.

Abandon handling can be gamed. See the table above. Choosing the convention that excludes abandons makes a failing queue look adequate.

Answering is not resolving. Service level measures the speed of the pickup and nothing after it. A center can hit 90/20 and resolve very little. It belongs next to first contact resolution and customer satisfaction, not on its own.

Chasing the last points is expensive. The staffing cost of each additional point rises steeply near the top of the range. Past a point, the money buys almost no customer-visible improvement and would do more good spent on resolution quality.

How to hold service level

Plan at the interval, not the day. Service level fails in specific half-hour windows. A daily figure will not locate them.

Get routing right on the first attempt. Transferred contacts consume capacity twice and the second wait is usually the one that breaks the target. Routing on customer history and topic rather than on a menu selection removes a large share of the rework.

Reduce the volume that should not exist. Repeat contacts about the same unresolved issue, and contacts chasing information the customer was never given, are the cheapest volume to remove. Every one taken out of the queue improves service level without a single extra hour of staffing.

Offer a callback. It converts a wait into a scheduled contact, reduces abandonment, and smooths the peak into the trough. One of very few options that improves the metric and the experience at the same time.

Set the target deliberately and defend it. If 80/20 was inherited rather than chosen, price the alternatives. The conversation about what the target should be is usually more valuable than any tactic for hitting the current one.

Service level when AI answers first

AI at the front of the queue breaks the standard service level calculation in a way that is easy to miss and easy to misreport.

If an AI assistant responds to every arriving contact within a second, service level measured across all contacts approaches 100%. That figure is arithmetically correct and operationally useless. It says the automation responded, not that anybody was served.

Two adjustments make the metric mean something again:

Measure service level on escalated contacts separately, with the clock starting at the point of escalation. This is the queue that still has people waiting in it, and it is the one where the target belongs.

Set that target tighter than the general one. A customer who has already spent time with an assistant before asking for a person has less patience remaining, not more. Centers often do the opposite, treating the escalation queue as a lower priority because overall volume has dropped, and the customers who most need help end up waiting longest.

The reporting discipline: publish the escalation service level, not the blended one. A blended figure near 100% will be quoted in places it should not be, and it will be wrong.