Call Center Operational Metrics Count Agent Time Four Ways

Updated 18 min read How we research

TL;DR: Call center operational metrics are mostly ratios with agent time in the denominator, and the denominators do not match each other. Asterisk, whose queue documentation is public, returns four different counts for one team: count is every member configured on the queue, logged is every member signed in, free is every signed-in member who can take a call or is wrapping up the last one, and ready is only the members available to answer right now. Occupancy, utilization, schedule adherence and average handle time each divide by a different slice of that, so one team produces four different productivity numbers and nobody is lying. Average handle time is three clocks, talk plus hold plus after-call work, and squeezing the third one pushes the work into first contact resolution. Occupancy and service level pull against each other because idle time is what absorbs arrival variance, and under the federal hours-worked rules that idle time is paid either way: an agent engaged to wait is working, rest periods running from five minutes to about twenty minutes must be counted as hours worked and may not be offset against compensable waiting time, meal periods of ordinarily thirty minutes or more are unpaid only when the agent is completely relieved from duty, and employer-required job-related training is work time unless all four exemption criteria are met. So shrinkage is a scheduling divisor, not a cost you can delete. Abandonment means two different things on the same phone system, a caller hanging up in queue and a dialer dropping a live answer, and only one of them carries a federal ceiling. Write down the numerator, the denominator, the exclusions and the interval for every metric before you set a target on any of them.

Here is a dashboard you have probably seen. Occupancy 85 percent, adherence 92 percent, service level missed three days running. Those figures are invented for the example. What is not invented is that all three came out of the same eight hours.

Nobody is lying. Each number divided by a different slice of the day, and the slice is not printed on the tile. That is the recurring problem with call center operational metrics. They look like independent readings of separate things. They are mostly one quantity, agent time, cut up by four or five different rules, and a target on any one of them moves the others whether you meant it to or not.

So before you argue about what good looks like, find out what your numbers are dividing by.

Call center operational metrics are ratios with agent time underneath

Strip the vocabulary away and most of the operational set has the same shape. Something that happened on the phone goes on top. Some amount of agent time goes on the bottom. Occupancy puts handling time over available time. Utilization puts productive time over paid time. Schedule adherence puts time in the right state over time you were scheduled to be in that state. Average handle time puts total handling effort over handled contacts, which is agent time again wearing a different hat.

Four ratios. One underlying quantity. So which minutes count? Four different rules, and nobody publishes theirs.

That is why two managers can quote two numbers for the same team, on the same day, and both be right. It is also why an industry benchmark you found in a listicle cannot grade you. The benchmark came from someone else’s definition of the denominator, and the definition is almost never published next to the number.

A metric is not a target either. A metric describes what the phone system recorded. A target says what you want it to record next month, and a target is only meaningful once the calculation rule, the exclusions and the interval are written down. Document those three things first. Then argue about the number.

Your call center software already counts the same team four ways

This is not a theoretical complaint, and you do not have to take it on faith. Asterisk is an open-source telephony platform, its queue documentation is public, and one function in it exposes the whole problem in four lines.

The QUEUE_MEMBER function returns a count of queue members based on the criteria you pass it. Pass it count and it returns the total number of members for the queue. Pass it logged and it returns the number of logged-in members. Pass it free and it returns the number of logged-in members that either can take calls or are currently wrapping up after a previous call. Pass it ready and it returns only the logged-in members that are immediately available to answer a call.

Read those back slowly. How many agents do you have right now? There are four correct answers. They are correct because they are four different questions.

There is a fifth state underneath them. A queue member can be paused, and the Asterisk application that does it describes pausing as blocking calls for that member, which prevents any calls from being sent from the queue to that interface until it is unpaused. A paused agent is still logged in. Whether that agent lands in your occupancy denominator is a configuration decision somebody made once, probably years ago, possibly by accident.

So when a vendor quotes you an occupancy figure, or a peer tells you theirs, the first question is not what the number is. It is which of those five populations went on the bottom of the fraction. If the answer is a shrug, the comparison is worthless. That is not a knock on the vendor. It is arithmetic.

The average handle time metric is three clocks, not one

Average handle time gets treated as a single stopwatch. It is three, running in sequence, and they respond to completely different interventions.

One long horizontal bar of frosted translucent white glass built from three abutting segments of unequal length, floating on a pale violet background.

Talk time is the conversation. Hold time is the customer waiting while the agent goes to find something. After-call work is everything the agent does once the call ends: the disposition, the notes, the CRM update, the follow-up task. Add the three, divide by handled contacts, and you get the number on the report.

Now watch what happens when a manager is told to bring handle time down. Which clock gives first? Talk time is hard to cut without cutting the conversation. Hold time is mostly a knowledge and tooling problem and takes a quarter to fix. After-call work is sitting right there, visible, and feels like overhead. So after-call work gets squeezed.

Where does the squeezed work go? It does not evaporate. The disposition gets picked at random, the notes get thin, and the next agent who picks up that customer starts from nothing. Handle time drops. Repeat contacts rise. First contact resolution falls, usually a reporting cycle later, in a different section of the dashboard owned by a different person.

Decompose the number before you set a target on it. If talk time is flat and after-call work is climbing, that is an admin problem and a tooling fix. If talk time is climbing and after-call work is flat, that is a conversation problem and a coaching fix. The aggregate tells you neither. Pair handle time with resolution and call quality scoring every single time you report it, so the tradeoff stays visible instead of settling somewhere nobody is watching.

Occupancy and service level are metrics that pull against each other

Here is the coupling that burns the most teams, and it is not subtle once you see it.

A frosted translucent white glass beam tilted on a small fulcrum, with a large glass sphere settled on the low end and a small glass sphere raised on the high end, on a pale violet background.

Occupancy measures the share of an agent’s available time spent handling contacts. Raising it means shrinking the gaps between calls. Service level measures the share of contacts answered inside a threshold, and answering a contact quickly requires somebody to be sitting there, unoccupied, at the moment it arrives.

The gaps are not waste. The gaps are the buffer.

Contacts do not arrive on a metronome. They cluster. The idle minutes between calls are what absorb a cluster without pushing callers into a queue, and an occupancy target is a direct instruction to remove them. Push occupancy high enough and the first bad fifteen-minute interval blows through your service level, because there was no slack left to spend. You did not get more efficient. You spent your margin for error. Then you booked it as a win.

Report the two together, on the same row, for the same interval. If occupancy went up and service level went down, that is one event, not two. And look at the distribution across intervals rather than the daily figure, because a day that averages fine can contain two intervals where everyone was underwater. The daily number hides exactly the thing you needed to see.

Inbound teams have a second lever worth checking before they staff up. Queue behavior changes how much of a cluster you have to absorb live, and what a call queue actually measures decides whether your demand number is the real one or a filtered one.

The idle time in your occupancy metric is already paid

There is a second reason an occupancy target does less for your cost line than it looks like it should. For covered, nonexempt employees in the United States, the federal hours-worked rules have already decided who pays for the gaps, and the answer is you.

The rule on waiting time draws the line between an employee engaged to wait and one waiting to be engaged. An employee who is inactive because the work has not arrived yet, who cannot use the period effectively for personal purposes, is working. The regulation says the time belongs to and is controlled by the employer, and it lists the examples: a stenographer reading a book while waiting for dictation, a messenger doing a crossword while awaiting assignments. Waiting is an integral part of the job. An agent sitting in ready state with no call in the queue is that example, in 2026, wearing a headset.

Off duty is the opposite case, and the bar is specific. The employee has to be completely relieved from duty, for a period long enough to use effectively for personal purposes, and has to be told in advance that they may leave and will not have to start work again until a definitely specified hour. Stay logged in, we might need you, does not clear that bar.

So the idle minutes an occupancy target is designed to eliminate are minutes you are paying for in either state. Raising occupancy does not recover that money. It converts a staffing buffer into queue time for the customer. That may still be the right call during a budget crunch. Make it knowing what you are actually trading, because the dashboard presents it as free.

None of this is legal advice and your own staffing model needs its own review. The arithmetic point survives the caveat.

Shrinkage is a divisor, and the law decides most of it

Shrinkage is the gap between the hours you scheduled and the hours actually available to take contacts. Breaks, meals, training, meetings, coaching, system outages, absence. Most teams treat it as a line item to attack. It is a divisor, and treating it as a line item produces the single most common staffing error there is.

The arithmetic is short. If the forecast says you need twenty agent-hours on the phone and shrinkage runs at 30 percent, you schedule twenty divided by seventy percent, which is just under twenty-nine hours. Schedule twenty and you are not break-even. You are about 30 percent short, every interval, and it shows up as queue time and abandons rather than as an obviously missing person.

Then there is the question of which parts of shrinkage you are even allowed to stop paying for, and the hours-worked rules answer most of it directly.

  • Short breaks are paid. Rest periods of short duration, running from five minutes to about twenty minutes, are common in industry, promote employee efficiency, and must be counted as hours worked. The regulation adds a line people miss: compensable rest time may not be offset against other working time such as compensable waiting time or on-call time. You cannot net a break against idle time.
  • Meals are unpaid only when they are real. Bona fide meal periods are not work time, ordinarily thirty minutes or more is long enough, and the employee must be completely relieved from duty for the purpose of eating a regular meal. Coffee breaks and snack time are not meal periods, they are rest periods. An agent required to stay at the desk and stay reachable while eating is working.
  • Training usually counts. Attendance at lectures, meetings and training programs needs to be counted as work time unless all four exemption criteria are met at once: outside regular hours, voluntary in fact, not directly related to the job, and no productive work performed. Attendance is not voluntary if the employer requires it, or if the employee is led to believe that nonattendance would hurt their standing. Training designed to make someone better at their present job is directly related to it.

Read that list as a staffing instruction rather than a compliance memo. Most of shrinkage is paid time that you are required to schedule around. So the useful management question is never how to cut shrinkage to zero. It is whether your measured shrinkage rate matches what you actually schedule, and whether you are dividing by it at all.

Measure it from the real record, by interval, over at least a quarter. One bad week of absence is not a planning number.

Abandonment rate is one metric name for two different things

Worth a short detour. A blended team reports one abandonment number, and it is secretly two.

On the inbound side, an abandoned contact is a caller who hung up while waiting. It is a service problem, it is yours to fix with staffing and queue design, and the threshold is whatever you decide it is. On the outbound side, an abandoned call is one the dialer placed and failed to connect to a live representative fast enough, and that one is not a matter of taste. It carries a federal ceiling, a specific measurement window, and a denominator written by a regulator. The detail is worked through in how the rules define their own denominators, and it is worth reading before you put a single abandonment tile on a blended dashboard.

Split the tile. One number for queue abandons, one for dialer abandons, labeled, never summed.

Which call center operational metrics belong on a dashboard

Start from the question you need answered, not from the list of what the platform can render. What do you actually need to decide this week? Most operating questions fall into four buckets. One metric from each beats twelve metrics from one.

  • How much work arrived? Contact volume by interval, by queue or campaign, by direction. Never the monthly total on its own, because the total hides the clusters you actually staff for.
  • How fast did people reach us? Service level with its threshold printed next to it, plus average speed of answer. These are not interchangeable. Service level says what share got in under the line. Average speed of answer says what the average wait was and nothing about the tail, and it usually excludes abandons, so it describes the experience of people who stayed.
  • What did each contact cost in time? Average handle time, broken into talk, hold and after-call work, and never reported alone.
  • Did the work actually land? First contact resolution against a stated window and a stated definition of resolved, plus a quality score from a rubric you wrote down.

Then add the pairings, because the pairings are the dashboard’s real job. Handle time next to resolution. Occupancy next to service level and adherence. Average speed of answer next to abandonment. Cost per contact next to contact mix, since a quarter where simple contacts moved to self-service makes cost per contact rise while the operation got better.

Three views cover most teams. A real-time view for the floor, showing only what somebody can act on in the next fifteen minutes. A weekly view for managers, showing distributions and outliers rather than averages. A monthly view for leadership, showing trend and the one or two decisions it implies.

And a rule that saves a lot of arguments later. No metric goes on a tile without an owner, a source system, a calculation rule and a stated exclusion list. If nobody will sign their name to the definition, the tile is decoration.

Where the raw numbers for these call center metrics come from

Every ratio above needs two things from the phone system: per-call timestamps and a state history for the agent. Reporting tiles are a rendering of those. If you cannot get at the underlying events, you cannot audit a definition you disagree with, and you certainly cannot rebuild one.

So check the data path first. The feature list can wait. Kixie’s call center analytics reporting covers the dashboard layer, and underneath it the developer API is included on every plan, with webhook events that fire on Start Call, Answered Call, End Call, Disposition, Call Outcome and Voicemail. Those events are the numerators. Pulling them into your own warehouse is how you get a definition you control rather than one you inherit.

Resolution metrics need a disposition that reps actually set, which is a workflow problem long before it is a reporting problem. Call disposition logging writes the outcome back without a second round of data entry, and bi-directional CRM sync is included on every plan, so the disposition and the customer record stay in one place. Quality scoring needs the conversation itself, and call recording is included on every plan as well. Conversation Intelligence, the AI analysis layer on top of those recordings, is a premium add-on rather than something in the base plan, so price it separately when you scope a QA program.

One more practical note. Agent state history is the hardest piece to reconstruct after the fact, because most platforms keep it for a shorter window than call records. Find out what your retention actually is before you promise anyone a six-month occupancy trend.

Call center operational metrics questions managers ask

What is a good occupancy rate for a call center?

Any number you are quoted is unusable until you know its denominator, and the four counts your own queue software returns for one team should tell you how wide that range is. Set your own baseline instead. Measure occupancy by interval for a quarter, find the level where service level stays inside its threshold on your worst day rather than your average one, and treat that as your ceiling. Then watch it alongside adherence and attrition, because sustained high occupancy with no slack is a scheduling decision that shows up later in turnover.

Is average handle time a useful call center metric?

Yes, for staffing math, which is what it was built for. It is a poor agent scorecard. The fastest route to a lower number is thinner after-call work, and thinner after-call work raises repeat contacts, so an agent target on handle time buys you a worse operation and a better chart. Use it to size the team. Use resolution and quality to coach the person.

How do I calculate shrinkage for scheduling?

Take the hours your agents were scheduled, subtract the hours they were actually available for contacts, and divide by the scheduled hours. Then use it as a divisor, not a subtraction: required on-phone hours divided by one minus the shrinkage rate gives you hours to schedule. Measure it by interval rather than as a single annual figure, because Monday mornings and the week of a product release do not share a shrinkage rate.

Why do service level and average speed of answer disagree?

Because they summarize different parts of the same distribution. Service level counts how many callers got in under a threshold and says nothing about how long the rest waited. Average speed of answer averages the wait for callers who were answered, which usually means abandons are excluded entirely. A queue can hit its service level target while a small group of callers waits a very long time, and the average will not show it. Report both, and look at the distribution when they disagree.

Do outbound sales teams need the same metrics as a service center?

Some of them, with different meanings. Handle time, occupancy and adherence translate directly, because they are agent-time ratios and an outbound rep has the same eight hours. Service level and average speed of answer are inbound concepts and do not. Abandonment exists on both sides and means two different things, which is the one that catches blended teams. If you run both motions on one phone system, label every metric by direction before anyone builds a report on it.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Ready to close more deals with Kixie?

See how Kixie's AI-powered tools can transform your sales and support operations.

Start Free Trial

Kixie Research Team is the organizational byline for content created and maintained by Kixie’s staff writers and editors through a shared research and editorial process. The team covers sales technology, outbound calling workflows, CRM integrations, and revenue operations. This shared byline reflects organizational accountability; it does not mean every article reflects hands-on testing. Individual pages identify their sources, methods, and limitations.

How we research, source, and update this article

How Kixie researches, sources, and updates articles like this one is described in our editorial standards.

If you spot an error or an outdated detail, request a correction.

Third-party sources linked in this article

This list is generated automatically from the third-party links referenced in the article above.