A mathematical expression that, for an M/M/c queue with Poisson arrivals, exponential service times, c identical servers, an infinite waiting room and no customer abandonment, gives the steady‑state probability that an arriving customer must wait (the delay probability) and related performance measures (e.g., average waiting time, queue length) as functions of offered load, number of servers and service rate.

Principle

Under the M/M/c assumptions, the probability of delay and waiting‑time metrics are determined by the offered load (a = λ/μ) and the number of servers; per‑server utilization is ρ = a/c = λ/(c·μ). Erlang C computes delay probabilities by conditioning on the state in which all servers are busy and applying the birth–death steady‑state solution with an infinite queue.

Demonstration

Illustrative scenario → Situation: A call center receives Poisson calls at rate λ=120 calls/hour; each agent serves with exponential service mean 1/μ=6 minutes (μ=10/hr) and there are c=14 agents. → Recognition: The system satisfies M/M/c assumptions and no abandonment is assumed. → Action: Apply Erlang C to compute delay probability P(wait) and Wq. → Consequence: Management obtains the fraction of callers who will wait and the expected queueing delay to size staff or evaluate service targets.

Misapplication

Treating Erlang C results as accurate when arrivals are non‑Poisson, service times are not memoryless, customers abandon, the queue discipline is not FIFO, or capacity limits exist; the semantic error is assuming the model’s structural stochastic assumptions need not hold for its outputs to be valid.

Consequence

When validly applied, Erlang C yields closed‑form staffing and service metrics that inform capacity decisions; when misapplied, it systematically underestimates waiting and overstates service levels because it ignores abandonment, service variability beyond exponential, or arrival burstiness.

Reversal

If customers may abandon (impatience), or arrival/service processes deviate substantially from Poisson/exponential, the appropriate model becomes M/M/c+M (Erlang A) or more general queueing approximations; in heavy‑traffic limits diffusion approximations can replace Erlang C for large systems.

Boundary

Clearly within: Poisson arrivals, exponential independent service times, c identical servers, FIFO, infinite buffer, no abandonment. Boundary case: Near‑Poisson arrivals or modest non‑exponential service times—Erlang C may approximate but requires validation. Clearly outside: Finite capacity systems, explicit abandonment, prioritized queues, time‑varying arrival rates without modification.

Semantic Tension

Analytical simplicity and closed‑form staffing guidance ↔ fidelity to real customer behavior (abandonment, time‑varying demand, non‑exponential service); using Erlang C trades model tractability for realism.

Synthesis

Erlang C is a compact tool for translating offered load and server count into delay probabilities under strict stochastic assumptions; its practical utility depends on validating those assumptions or replacing the formula with models (Erlang A, time‑dependent or simulation models) when assumptions fail.