Erlang C calculator
Work out how many agents a queue needs to answer a target share of contacts within a target time. The calculation runs in your browser; nothing you type is sent anywhere unless you ask for the spreadsheet.
Agents required on the queue
14
Set by the service level
Agents to schedule
20
After 30% shrinkage
Service level achieved
88.8%
Answered within 20 sec
Average speed of answer
7.8 sec
Occupancy
71.4%
Workload
10 Erlangs
17.4% of contacts wait
What Erlang C calculates
Erlang C is a queueing formula for contact centers. Given how much work arrives in an interval and how many agents are on the queue, it gives the probability that a contact has to wait. From that probability you can work out the service level (the share of contacts answered within a target time), the average speed of answer (ASA) and the occupancy of the agents.
A planner usually runs it the other way round: start from the service level you promise, such as 80% of contacts answered within 20 seconds, and find the smallest number of agents that meets it. That is what this calculator does. It steps up the agent count from the minimum needed to carry the workload until the service level target, and your occupancy cap if you set one, are both met.
The formula and the inputs
Workload is measured in Erlangs: the number of agents' worth of work arriving at once. With N agents on the queue and A Erlangs of workload, the probability that a contact waits is:
A = (contacts x AHT in seconds) / (interval in seconds)
C(N, A) = [ A^N / N! x N / (N - A) ] / [ sum(k = 0..N-1) A^k / k! + A^N / N! x N / (N - A) ]
Service level = 1 - C(N, A) x e^( -(N - A) x T / AHT )
ASA = C(N, A) x AHT / (N - A)
Occupancy = A / N- Contacts per interval: calls or chats arriving in the interval you are planning, usually 15 or 30 minutes.
- Average handle time (AHT): talk time, hold time and after-contact work, in seconds.
- Service level target and answer time: for example 80% within 20 seconds. T in the formula is the answer time.
- Maximum occupancy: an optional cap on how busy agents are allowed to be. If the cap needs more agents than the service level does, the result says so.
- Shrinkage: time agents are paid but not available. It does not change the Erlang C result; it converts agents on the queue into agents to schedule.
Worked example
Suppose 100 calls arrive in a 30-minute interval and the average handle time is 180 seconds. The workload is 100 x 180 / 1,800 = 10 Erlangs. The target is 80% of calls answered within 20 seconds.
With 13 agents the service level is about 79.6%, just under the target. With 14 agents it is about 88.8%, the average speed of answer is about 8 seconds and occupancy is about 71%. So 14 agents must be on the queue. If shrinkage is 30%, you schedule 14 / (1 - 0.30) = 20 people for that interval.
Notice how sharply the result moves around the target: one agent more or fewer changes the service level by roughly ten points here. That is why planners rarely staff exactly to the number and review the plan interval by interval.
Assumptions and where Erlang C is wrong
Erlang C is a model, and its assumptions are simple on purpose. Knowing them tells you when to trust the output and when to add judgement.
- Contacts arrive at random at a steady average rate within the interval. Real traffic has spikes, so the model is usually a little optimistic.
- Nobody abandons. Real customers hang up, which shortens the queue and makes real service levels look better than the model, while hiding lost contacts.
- Every agent can handle every contact. If skills differ, model each skill group separately or use a skills-aware tool.
- Handle times are exponentially distributed. Handle times with a long tail behave a little differently.
- One contact at a time. For chat with several conversations per agent, divide the workload by concurrency, as a first approximation, and verify against your own data.
Common mistakes
- Using a daily average instead of the interval. Staffing for the average hides the busy half hour where the service level is lost.
- Leaving after-contact work out of handle time. Wrap-up is workload.
- Forgetting shrinkage, or applying it twice. Shrinkage is applied once, to turn agents on the queue into agents scheduled.
- Planning for 100% occupancy. Agents who are always busy burn out and have no time for coaching or breaks; many operations cap occupancy well below 100%.
- Treating the output as a promise. Compare the model with your actual service level for several weeks and calibrate.
Typical ranges to check against your own operation
Service level targets are a business decision, and common examples include 80% within 20 seconds for voice, though many operations choose something different. Occupancy targets for voice queues are often discussed in the range of roughly 80% to 90%, and lower for work that needs concentration. These are typical ranges, verify them for your operation, your channel and your contract, since a client agreement or regulation may fix the numbers for you.

