Erlang C Explained Without the Math Headache: A Staffing Guide
What Erlang C is, what it assumes, what inputs it needs and how to read the result, with a worked example and the limits to keep in mind.
Anyone who has looked into call centre staffing has met the name Erlang C. It is the classic method for turning a forecast into a number of agents, and it sounds more frightening than it is. You do not need to do the maths by hand, but you do need to understand what goes in, what comes out and where the method stops being reliable. This article explains it in plain language.

Erlang C starts from a forecast; see Forecasting Call Volume for Small Teams. The target you give it is usually a Service Level, and the output feeds your schedule in Shift Planning Basics.
What Erlang C is
Agner Krarup Erlang was a Danish mathematician who studied telephone exchanges. His work on queues gives us a formula for the chance that a caller has to wait when calls arrive randomly and a fixed number of agents serve them. Erlang C is the version for queues where callers wait rather than give up. Given calls, handle time and agents, it estimates the probability of waiting and from that, the average wait and the share of calls answered within a target time.
In practice, you reverse it: you set the service target and ask how many agents are needed to meet it.
The inputs, in plain words
| Input | What it means | Where to get it |
|---|---|---|
| Call volume | Calls arriving in an interval, such as 30 minutes | Your forecast |
| Average handle time | Talk time plus after-call work | Your call history |
| Service level target | A share of calls answered within a set number of seconds | Your own decision |
| Shrinkage | Time agents are paid but unavailable | Your schedule and absence data |
Choose the interval carefully. Half-hours are common, because demand changes during the day. A daily total tells you nothing about who should be on the phones at ten o'clock.
The service level target
A service level target states how fast you aim to answer. A well-known example is "eight in ten calls answered within twenty seconds". It is a business decision about customer experience and cost, not a law of nature. A tighter target needs more people; a looser one needs fewer. See Service Level and the 80/20 Rule for how to choose one.
Workload in hours
The first step is workload: calls multiplied by handle time. If you expect 60 calls in a half-hour and each takes 5 minutes, the workload is 300 minutes, which is 5 hours of work to fit into half an hour. That alone tells you at least ten agents' worth of effort are needed, because ten agents give five hours in thirty minutes. The extra agents Erlang C adds above that figure are what keep waits short when calls arrive in clumps.
How to read the result
From inputs to agents
- Calculate workloadCalls multiplied by handle time, expressed as agent-hours of work in the interval.
- Run the formulaA calculator finds the smallest number of agents whose service level meets the target.
- Add shrinkageRaise the number to allow for breaks, meetings, leave and sickness.
- Round upYou cannot schedule part of a person.
- Compare with the scheduleLook for intervals that are over or under.
The extra agents above the bare workload give you safety margin. A small queue needs proportionally more margin than a big one, because randomness matters more with few agents. This is the "economy of scale" in contact centres: a larger pooled team can reach the same service level with less spare capacity per agent.
Things the result will show
- Required agentsThe number on duty in the interval to meet the target.
- OccupancyThe share of paid time agents spend handling calls; if too high, agents are overworked.
- Expected waitThe average time a caller waits.
- Service level reachedCalls answered within the target time.
What Erlang C assumes, and where it fails
Erlang C is a model. A model is useful because it is simple, but its assumptions are not exactly true.
- Callers never give up. In reality some hang up, which makes waits shorter for others. Erlang C therefore tends to overstate the staff you need at busy times.
- Arrivals are random and steady within the interval. Real calls bunch after a promotion or an outage.
- All agents handle all calls. Skills-based routing and specialist groups change the picture.
- Handle time is steady. Long calls cluster in real life.
- No retries. Abandoned callers often call back, which adds load.
- No other work. If agents also answer email or chat, they are not free for calls.
There are refinements, such as Erlang A, which allows for abandonment, and simulation tools. You do not need them to start. Most small teams do better by using Erlang C as a guide, adding a sensible margin and checking results weekly.
A worked example, in words
Suppose a small clinic's phone team expects 40 calls in the 10:00 half-hour, with a typical call of 4 minutes. The workload is 160 minutes, close to 2.7 hours of work in half an hour, so the team needs at least 6 agents working fully just to keep up, and that would leave no time for random clumps. A calculator set to answer eight in ten calls within twenty seconds would suggest a few more than that, say 8. After allowing for breaks and leave, the schedule might need 9 or 10 people on the roster for that slot.
Notice that these numbers are hypothetical and only show the logic: workload sets the floor, the service target adds a margin and shrinkage adds the rest. If the team cannot afford that, the right response is to change something you control, such as handle time, the service target, callbacks or the call mix, not to hope.
When a simple rule of thumb is enough
For very small teams, a formal calculation can be more precision than you need. If you have three or four agents and your calls are short, a simple rule works: staff to the workload, add one person's worth of margin for the busiest interval, and use callbacks or voicemail when the line is full. Check the result by looking at abandoned calls and waiting times every week.
The calculator becomes more valuable as the queue grows, as you add several queues or skills, or when you must show the business why a shift needs another person. Even then, the lesson of Erlang C is simple: random arrivals mean you need slack, and the smaller the team the more slack you need. Give the model honest inputs and keep checking.
Using the numbers wisely
Good practice with Erlang C
- Use half-hour intervals from your own forecast
- Base handle time on recent data, by call type
- Agree the service level target with the business
- Add shrinkage from your measured history
- Check the occupancy for each interval
- Compare to actual results weekly and adjust
- Use callbacks or queue messages to relieve peaks
Occupancy deserves attention. If the model says you can run at a very high occupancy, think about what that feels like: no breathing space between calls. See Occupancy and Utilization and Preventing Agent Burnout.
Comparing needs and schedules in Call
Call's Workforce area gives you a "Required vs scheduled" view for the day after you press "Rebuild forecast", which builds a half-hour forecast for each inbound queue from its call history. It compares the staffing required with what has been scheduled, so you can see the gaps by interval. Queue service level appears in the Supervisor view against its target, with the longest wait and the number of agents available, and queue thresholds can raise alerts when waits grow.
These views help you apply the thinking in this article without building your own model. Use them alongside your own checks, because any model, built-in or not, depends on its inputs. Which features are available depends on your workspace.
Frequently asked questions
Do I need to know the Erlang C formula?
No. You need to understand the inputs and how to read the result. Use a calculator or spreadsheet for the arithmetic.
Is Erlang C accurate?
It is a useful approximation for planning, but it assumes patient callers, steady arrivals and equal agents. Real centres differ, so adjust with experience and compare results with actual performance.
What service level should we choose?
It depends on your customers, your cost and your channel. A common example is eight in ten calls within twenty seconds, but you should choose a target your business can sustain.
What is shrinkage?
The proportion of paid time when agents are not available to take calls, due to breaks, training, meetings, sickness and leave. Measure it from your own data.
Can small teams use it?
Yes, but small queues are very random, so keep a margin and review often. A simple spreadsheet calculator is enough.
What is the difference between Erlang B and Erlang C?
Erlang B models lines with no waiting, where blocked callers are lost. Erlang C models queues where callers wait for an agent. Contact centre staffing uses Erlang C.