Skip to content
Tutorials Center
My learning

Topics

Lesson 35 of 38 · Simulation and Applications

Staffing a Call Center with Erlang C

Use Erlang C to compute the probability of waiting, average speed of answer and service level, then find the number of agents that meets a target.

A support line receives calls, and agents answer them. Too few agents and callers wait on hold, hang up and complain. Too many and you pay for idle time. Staffing is the application where queueing theory earns its keep, and the standard tool is the Erlang C model, an M/M/c queue with unlimited waiting room.

The model and its inputs

  • λ: call arrival rate (for example 120 calls per hour).
  • 1/μ: average handling time (talk plus wrap-up), so μ is the number of calls one agent handles per hour.
  • c: number of agents.
  • Offered load a = λ/μ (in erlangs): the average number of agents busy with work. Occupancy is ρ = a/c and we need c > a.

Callers who find all agents busy wait in a single FIFO line and nobody hangs up (a simplification we revisit below).

The formulas

The probability that a caller has to wait is:

Text
C(c, a) = [ a^c / (c! (1 - ρ)) ] / [ Σ_{k=0}^{c-1} a^k/k!  +  a^c / (c!(1 - ρ)) ]

Given that the caller waits, the wait is exponential with rate cμ - λ. So, for a time t:

Text
P(Wait > t) = C · e^{-(cμ - λ) t}
Service level SL(t) = 1 - C · e^{-(cμ - λ) t}
Average speed of answer ASA = C / (cμ - λ)

A typical target is "80 percent of calls answered within 20 seconds", written 80/20.

Python Erlang C calculator

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The offered load here is a = (120/3600) × 180 = 6 erlangs. With only 7 agents the line is nearly saturated and the waits are long; each extra agent cuts them sharply.

Finding the staffing level

Staffing is the inverse problem: find the smallest c meeting the service goal.

Python Find the minimum number of agents for an 80/20 target

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Notice that occupancy rises as the center grows. Large centers pool their agents, so they can run busier at the same service level. This is the same economy of scale seen in loss systems.

What the model leaves out

Erlang C is a good first estimate but real centers differ in important ways:

  • Abandonment: callers hang up after waiting. Erlang C ignores this and so overestimates waits. The Erlang A model adds an abandonment rate.
  • Non-exponential handling times: if handling times are more variable, waits are longer than predicted (compare with the P-K lesson).
  • Time-varying arrival rates: staff for each half-hour using that period's forecast rate; averaging over the day hides peaks.
  • Breaks and shrinkage: schedule more people than c to cover breaks, training and absence.

Use Erlang C to get the order of magnitude, then check the candidate staffing with a simulation that includes abandonment and realistic handling-time variability.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. A center receives 90 calls per hour with a 4-minute (240 s) average handling time. What is the offered load in erlangs?

    a = λ/μ = 90 calls/h × (240/3600) h = 6 erlangs.
  2. 2. Given that a caller waits, how is the wait distributed in the Erlang C model?

    The conditional wait is exponential with rate cμ - λ, which gives SL(t) = 1 - C·e^{-(cμ-λ)t}.
  3. 3. Why does Erlang C overestimate waits in many real centers?

    Real callers hang up, which shortens the line for the others.
  4. 4. Occupancy at the staffing level typically rises when the center gets larger because

    Larger pooled systems need proportionally less spare capacity for the same service level.

Summary

  • Erlang C: M/M/c with unlimited waiting; inputs are λ, μ, c.
  • P(wait) = C(c, a), SL(t) = 1 - C e^{-(cμ-λ)t}, ASA = C/(cμ - λ).
  • Staff by searching for the smallest c meeting the target.
  • Abandonment, variable handling times and time-varying demand call for refinements or simulation.

Next: Practice Set: Finite Capacity and Priorities