How Erlang-C Pacing Reduces Abandon Rates by 60%

The Problem with Fixed-Ratio Dialing

Traditional predictive dialers use a fixed ratio of calls to available agents. If you have 10 agents, you might dial 15 numbers simultaneously, hoping that by the time a human answers, an agent will be free. This approach has a fatal flaw: it treats call arrival as deterministic when it is fundamentally stochastic.

The result? Abandon rates that routinely exceed the FCC's 3% threshold, exposing contact centers to TCPA liability of $500--$1,500 per violation.

Enter Erlang-C

Erlang-C is a queuing model developed by Agner Krarup Erlang in 1917 for telephone traffic engineering. It calculates the probability that an incoming call must wait in queue given:

c -- the number of available agents

λ (lambda) -- the arrival rate of answered calls

μ (mu) -- the average service rate per agent

The key formula gives us P(wait > 0), the probability that a call will not be immediately served. From this, we derive the expected wait time and, critically, the probability of abandonment.

How MobDial Implements Erlang-C Pacing

Our call-pacer module in apps/dialer-service recalculates pacing every 2 seconds using real-time metrics:

Agent pool state -- how many agents are idle, in wrap-up, or about to become available

Historical answer rate -- the rolling 15-minute percentage of calls answered by humans (vs. voicemail, busy, no-answer)

Average handle time -- the rolling mean of talk time plus after-call work

The pacer feeds these into the Erlang-C model and outputs an optimal number of simultaneous dials that keeps the predicted abandon rate below 2% -- well under the 3% regulatory ceiling.

Results

After deploying Erlang-C pacing across our beta customers:

Abandon rate: dropped from 4.8% to 1.9% (60% reduction)

Agent utilization: increased from 62% to 78%

TCPA complaints: zero in the first 90 days

The math works. More importantly, it keeps our customers compliant.