Erlang C staffing calculator.

Worked example

100 calls arrive in a 30-minute interval and each takes 180 seconds, so the workload is 100 x 180 / 1,800 = 10 erlangs. To answer 80% of calls within 20 seconds you need 14 agents on the phones (service level 88.8%, average speed of answer 7.8 seconds, occupancy 71.4%). With 30% shrinkage you schedule 14 / 0.70 = 20.00, rounded up to 20 agents.

How it works

Workload A = calls x handle time / interval length, in erlangs. The Erlang C formula gives Pw, the probability an arriving call has to wait, for N agents. Service level is 1 - Pw x e^(-(N - A) x T / AHT) and the average speed of answer (ASA) is Pw x AHT / (N - A). The tool tries N = A+1, A+2 and so on until both your service level and your optional occupancy cap are met.

Shrinkage is applied after the Erlang step: scheduled agents = ceil(agents / (1 - shrinkage)). It is a planning convention, not part of the Erlang C maths.

Limits. Erlang C assumes Poisson arrivals, exponential handle times, an unlimited queue and callers who never hang up. Real queues with abandonment, skills routing or bursts need more or fewer agents. Check interval-by-interval, because workload changes through the day.

FAQ

What is Erlang C used for?

It estimates how many agents a queue needs so that a target share of calls is answered within a target time, assuming callers wait in an unlimited queue.

How do I add shrinkage to Erlang C?

Run Erlang C to get the agents needed on the phones, then divide by (1 - shrinkage) and round up. With 30% shrinkage, 14 agents on the phones becomes 20 scheduled.

What is a good service level target?

There is no single standard. The defaults here (80% answered within 20 seconds) are only an illustration. Set the target from your own customer expectations and budget.

Why does Erlang C sometimes overstate staffing?

It assumes nobody abandons the queue. Callers who hang up shorten the queue for others, so the real service level is often better than the model says.