Call center occupancy is the share of agents' available time that went to handling contacts. Take the time spent on calls, hold and after-call work, divide it by that same time plus the time agents sat ready with nothing to answer, and multiply by 100. It is measured per half hour and leaves out breaks, training and any other time agents were logged in but unavailable.
No single occupancy rate is right for every center, because occupancy comes out of three inputs: call volume, handle time and the service level you staff to. Hold every queue to 80 percent of calls answered within 20 seconds with 5 minute calls, and the occupancy that target allows runs from 50 percent on a queue that needs 4 agents to 92.6 percent on one that needs 108. The first table below shows where a queue of your size lands, and the second shows what a 5 or 10 percent forecast miss does at each of those levels.
What is occupancy in a call center?
Occupancy measures how busy agents were during the time they were free to take calls. In the half-hour reports an automatic call distributor produces, it equals total active time divided by total active plus total available time, times 100. Time agents were logged in but unavailable is left out on purpose.
That definition comes from the ACD half-hour report Gans, Koole and Mandelbaum walk through in Telephone Call Centers: Tutorial, Review, and Research Prospects (Manufacturing and Service Operations Management, 2003). The report sorts each logged-in agent's time into three states: handling a call, available and waiting for one, or unavailable. Occupancy only covers the gap between the first two, which is idle time left by the natural variation in when calls arrive and how long they take. Time agents spent logged in but unavailable, and time they were scheduled but not logged in, are separate losses with separate figures, so a center can post high occupancy while losing a lot of paid time elsewhere. Schedule adherence is where those losses get measured.
Occupancy is also the quantity staffing models work with. Offered load, in Erlangs, is the calls in an interval times average handle time, divided by the length of the interval. Divide offered load by the agents on the phones and you get the average occupancy those agents will run at, which the paper calls system utilization or occupancy.
How do you calculate call center occupancy?
Divide handling time by handling time plus available idle time for the same interval, then multiply by 100. Handling time is talk, hold and after-call work. For planning, use the equivalent form: offered load in Erlangs divided by agents on the phones. Work it out per half hour and per queue before you average anything.
A worked example for one half hour, with 25 agents on the phones and 120 calls answered at a 5 minute average handle time:
- Handling time is 120 calls × 5 minutes = 600 agent minutes.
- Time available to handle calls is 25 agents × 30 minutes = 750 agent minutes, assuming all 25 stayed in a handling or available state for the whole interval.
- Occupancy is 600 ÷ 750 = 80 percent. The other 150 agent minutes went to waiting for the next call.
The planning form gives the same answer. 120 calls × 300 seconds ÷ 1,800 seconds is 20 Erlangs of offered load, and 20 ÷ 25 is 80 percent.
In a spreadsheet with one row per agent for the interval, handling seconds in column B and available idle seconds in column C, the interval's occupancy is =SUM(B2:B26)/(SUM(B2:B26)+SUM(C2:C26)). Sum first and divide once. Averaging each agent's own percentage gives an agent who was logged in for ten minutes the same weight as one who was there all half hour.
What your platform puts in each side
Platforms draw the line in slightly different places, so read the definition before you compare two reports. Amazon Connect calculates occupancy as contact time divided by contact time plus idle time, and its documentation says the metric does not account for concurrency: an agent handling at least one contact for the whole interval counts as 100 percent occupied (Amazon Connect metric definitions). Its idle time includes the seconds from when a contact starts routing to an agent until the agent picks up or declines. NICE notes that interaction handling typically includes talk, hold and after-call work, but that some calculations leave after-call work out (NICE glossary). Two teams with different settings for after-call work can report different occupancy for identical days.
Call center occupancy vs utilization: what is the difference?
Occupancy compares handling time with the time agents were available to handle contacts. Utilization usually measures agent time against something larger, often the whole shift, so it also takes in meetings, coaching and training. Published definitions of utilization disagree about what sits on each side of the fraction, so pin yours down in writing before comparing teams.
| Term | Numerator | Denominator | Where this version is published |
|---|---|---|---|
| Occupancy | Time handling contacts | Handling time plus available idle time | ACD half-hour reports, per Gans, Koole and Mandelbaum (2003) |
| Utilization, queueing sense | Offered load in Erlangs | Agents on the phones | Gans, Koole and Mandelbaum (2003), as another name for occupancy |
| Utilization, workforce sense | Total logged-in time | Total shift time | Call Centre Helper |
The third version doesn't measure call work at all. An agent logged in for the whole shift and idle for most of it scores 100 percent. Use occupancy to judge the staffing model in an interval, and use utilization, in whichever form you settle on, to see how much paid time reached the phones.
What is a good occupancy rate for a call center?
A good occupancy rate is whatever your service level target produces for your queue's size, not a figure borrowed from another center. The range you will usually hear is 80 to 85 percent, but with the same 80/20 target and the same 5 minute calls, the right number runs from 50 percent on a small queue to more than 90 percent on a large one.
Call Centre Helper calls 80 to 85 percent typical, and in the same piece says there is no recommended best-practice benchmark for occupancy. The table shows why a single figure can't hold. Each row is one half hour with a 5 minute average handle time, and the agent counts are the fewest that meet each target under the Erlang C formula.
| Calls in the half hour | Offered load | Agents for 80/20 | Occupancy at 80/20 | Agents for 90/20 | Occupancy at 90/20 |
|---|---|---|---|---|---|
| 12 | 2 Erlangs | 4 | 50.0 percent | 5 | 40.0 percent |
| 30 | 5 Erlangs | 8 | 62.5 percent | 9 | 55.6 percent |
| 60 | 10 Erlangs | 14 | 71.4 percent | 15 | 66.7 percent |
| 120 | 20 Erlangs | 25 | 80.0 percent | 26 | 76.9 percent |
| 300 | 50 Erlangs | 57 | 87.7 percent | 59 | 84.7 percent |
| 600 | 100 Erlangs | 108 | 92.6 percent | 111 | 90.1 percent |
Two things drive the spread. A bigger queue pools more random arrivals, so the same service level needs proportionally fewer spare agents, and a stricter target pulls occupancy down at every size. Staff the 60-call queue to an occupancy of about 83 percent and the plan comes out at 12 agents instead of 14. Occupancy then reads 83.3 percent, while service level falls to 60.7 percent and the average caller waits 67 seconds. One contributor to a Call Centre Helper reader roundup put it plainly: "I can easily ensure a 90%+ occupancy by simply understaffing."
The figures use Erlang C as Gans, Koole and Mandelbaum write it, the same model behind our service level guide, and it assumes no caller ever hangs up. Real queues lose some callers instead, so read the tables as the shape of the trade rather than a forecast for your floor. To pick a ceiling from the operating style you run, see the occupancy section of our workforce management guide.
Why does high occupancy backfire?
High occupancy leaves no spare capacity, so any half hour that brings more calls than the forecast turns into a long queue. Waits climb slowly at moderate occupancy and steeply as it nears 100 percent, because the idle minutes that soak up random bunches of calls are already gone. The same forecast miss does far more damage to a center running hot.
Each queue below is staffed for its forecast at 80/20, using the agent counts from the table above, and then more calls arrive than planned with nobody added.
| Forecast calls in the half hour | Agents | Planned occupancy | Service level, 5 percent more calls | Average wait | Service level, 10 percent more calls | Average wait |
|---|---|---|---|---|---|---|
| 12 | 4 | 50.0 percent | 82.4 percent | 31 seconds | 79.9 percent | 38 seconds |
| 30 | 8 | 62.5 percent | 82.8 percent | 23 seconds | 78.7 percent | 30 seconds |
| 60 | 14 | 71.4 percent | 81.6 percent | 20 seconds | 75.2 percent | 30 seconds |
| 120 | 25 | 80.0 percent | 76.5 percent | 23 seconds | 64.5 percent | 43 seconds |
| 300 | 57 | 87.7 percent | 67.7 percent | 29 seconds | 37.7 percent | 107 seconds |
| 600 | 108 | 92.6 percent | 43.7 percent | 69 seconds | No steady state | No steady state |
On a normal day the largest queue runs the leanest staff in the table, and a bad forecast hurts it most. Five percent more calls take its service level from 80.7 to 43.7 percent. Ten percent more puts more work in the half hour than 108 agents can do, which the formula can't resolve and a real floor resolves through abandoned calls and callers who try again later. The smallest queue loses about 5 points on the same 10 percent miss.
The curve shows up in operating data too. Plotting average speed of answer against occupancy for each of 3,867 hourly intervals a small call center was open in 1999, the data "clearly parallel the theoretical relationship," according to Gans, Koole and Mandelbaum (2003, citing Brown and colleagues). At the top of the range, the no-hang-up assumption matters most. In one heavily loaded half hour the same paper analyzed, a model without abandonment would have needed agents busy more than 100 percent of the time, and the authors call ignoring abandonment there an order-of-magnitude error. They recommend Erlang A, which models callers hanging up, and note a common complaint from managers that workforce systems consistently recommend overstaffing. The workforce management guide works through what a 5 percent miss does to a center planned at 95 percent.
High occupancy can also hide in other numbers. Another contributor to the Call Centre Helper roundup said "the higher occupancy rates are often hidden in a higher AHT," as tired agents take longer on calls. That is one manager's observation rather than a measured result, and you can test it on your own floor: plot each interval's average handle time against its occupancy and see whether handle time rises in the hot ones.
How to set an occupancy ceiling for each queue
Set the ceiling per queue from your own numbers: start at the occupancy your service level target produces at forecast volume, then lower it by whatever headroom your worst forecast misses require. Treat it as a planning limit inside the staffing model, report occupancy per interval and keep it off individual scorecards.
- Choose the definition. Decide whether after-call work counts as handling time and whether ring time counts as idle, then write the choice down so every report uses it.
- For each queue's busiest intervals, work out offered load from forecast calls and handle time, and find the agents your service level needs. Occupancy at that agent count is the queue's natural level.
- From recent interval history, find how far actual volume ran above forecast in your worst intervals.
- Rerun the staffing at that higher volume. Where service level collapses, either lower the ceiling and staff more agents for those intervals, or plan overflow coverage for them.
- Report occupancy per interval and per queue, next to service level and average speed of answer. A shift or daily figure averages hot half hours into calm ones, a trap the occupancy entry in our call center metrics guide walks through.
- Keep it off agent scorecards. Agents don't decide when calls arrive, and Call Centre Helper's guidance is that "occupancy must only be measured on a contact center-wide level; it cannot be used as an advisor target."
Our staffing calculator runs step 2 for your busiest hour and takes a maximum occupancy as one of its inputs, so you can see when the ceiling, rather than service level, sets the agent count.
Where an AI agent changes the occupancy math
The usual answers to a hot interval are more people or longer waits. An AI agent changes the offered load instead. Telvana answers immediately at any hour, so a phone-menu option offered when the wait climbs, or a forwarding rule for overflow, moves calls out of the queue in exactly the half hours that run hottest (hold-queue relief). Callers who would rather wait for a person keep their place in line. A call the agent transfers to your team lands back in that queue's volume, so count it there.
Taking load off a queue also lowers the occupancy its remaining agents can run at the same service level, as the first table shows. Plan the roster from the new volume rather than cutting it by the share of calls that moved.
On outbound, a predictive dialer exists to keep agents busy by dialing before they are free, which only works on campaigns with enough agents to forecast from (predictive dialer vs AI agents). When an AI agent runs the campaign, the thing that places the call is the thing that talks, so there is no idle gap between calls to fill. Our team builds each agent from your scripts and transfer rules and tests it with you before it takes a live call. Pricing starts at $399 a month based on call volume, plus a one-time setup fee (pricing).
Frequently asked questions
Is occupancy the same as utilization?
Sometimes. In queueing models the two words name the same quantity, offered load divided by agents. In workforce reporting, utilization often compares logged-in time with total shift time, which shows how much paid time reached the phones rather than how busy the phones kept people. Check which one a report uses before you compare it with another.
Does after-call work count toward occupancy?
In most calculations it does, because an agent finishing notes can't take the next call. NICE notes that some calculations leave it out. If yours does, a team with long wrap-up reads as less busy than it is. Our after-call work guide covers how wrap-up time is counted.
How many calls does an agent take per hour?
Divide the minutes an agent spends handling calls in an hour by the average handle time. At 80 percent occupancy and a 5 minute handle time, that is 48 minutes of handling, or about 9.6 calls an hour. At 92.6 percent occupancy, the large queue's level in the first table, it is about 11.1.
Should occupancy be a KPI for individual agents?
No. Agents don't choose when calls arrive, so an individual's occupancy mostly reflects the forecast and the schedule. Call Centre Helper's guidance is to measure it across the center only. Score agents on schedule adherence and call quality, and use occupancy to judge the staffing plan.
How is occupancy measured for chat?
Chat bends the arithmetic, because one agent can hold several conversations at once. Amazon Connect's occupancy metric does not account for concurrency, so an agent with at least one chat open for the whole interval reads 100 percent however many they handle. Track concurrent conversations per agent next to occupancy for chat queues.
Try the numbers on your own queue
Put your busiest hour into the staffing calculator with your own maximum occupancy, then see how the agent count changes when AI agents take part of the calls. To hear one take a call from your queue, book a demo and bring your interval report.