Erlang C in Plain Language: How Many Agents Do You Really Need?
A formula more than a century old is still the foundation of contact center staffing. We explain it without heavy math, with a worked example step by step.
PeekAI · · Workforce Management
"We need more people." Every finance director has heard it from the contact center manager, and always replies with a question: "How many, exactly? And why?" The most precise answer still relies on a formula a Danish engineer, Agner Erlang, developed in the early twentieth century to plan telephone exchanges: Erlang C.
You won't need to solve it by hand. Every workforce management system and free calculator does that for you. But you do need to understand its logic, so you know when to trust the result and when to question it.
The idea in one sentence
Calls don't arrive evenly. Ten may arrive in one minute and two in the next. So dividing workload by headcount isn't enough. Erlang C calculates the probability that a customer will have to wait when calls arrive at random, and every other number is derived from that probability, as this plain-language guide explains.
The three inputs
According to the Society of Workforce Planning Professionals (SWPP), you need:
- Forecast call volume for the interval (half hour or hour).
- Average handle time (AHT): talk time + hold time + after-call work.
- Target service level: the most common in the industry is 80/20, meaning 80% of calls answered within 20 seconds.
The example, step by step
Say a contact center expects 300 calls between 10:00 and 10:30 a.m., with an AHT of 4 minutes (240 seconds) and a target of 80/20.
Step 1: calculate the workload
Workload = calls × AHT ÷ interval length
= 300 × 240 ÷ 1,800 seconds = 40 Erlangs.
Put simply: if calls arrived perfectly evenly, you'd need 40 agents busy all the time. That's the theoretical minimum, not the number you need.
Step 2: test different staffing levels
This is where Erlang C comes in. We calculated the results for different numbers of agents actually available on the phones:
| Agents on the phones | Service level (within 20s) | Average speed of answer | Occupancy |
|---|---|---|---|
| 43 | 57.9% | 43 seconds | 93% |
| 44 | 69.1% | 26 seconds | 91% |
| 45 | 77.5% | 16 seconds | 89% |
| 46 | 83.9% | 11 seconds | 87% |
| 47 | 88.6% | 7 seconds | 85% |
| 48 | 92.0% | 5 seconds | 83% |
The smallest number that meets 80/20 is 46 agents on the phones.
Step 3: add shrinkage
Those 46 must be on the phones, actually taking calls. But part of paid time never goes to answering: breaks, training, meetings, coaching, absence. SWPP puts this typically between 20% and 35%.
With 30% shrinkage:
Scheduled agents = 46 ÷ (1 − 0.30) = 46 ÷ 0.70 = 65.7, so 66 scheduled agents for this interval.
Notice the gap: from 40 "in theory" to 66 on the actual schedule. That gap is what you need to explain to finance.
Three lessons from the table
- One agent makes a big difference. Going from 45 to 46 agents raised service level by about 6 points. Going from 44 to 43 dropped it by more than 11. As SWPP puts it, removing staff near the minimum causes service to collapse fast.
- Returns diminish. Beyond 48 agents, each extra person improves service only slightly while adding the same cost.
- Very high occupancy is risky. At 43 agents occupancy hits 93%, so agents barely catch their breath between calls. Over time that level leads to burnout and attrition.
When not to trust Erlang C
- When customers hang up while waiting. Erlang C assumes every caller waits until answered. In reality some abandon, so the formula slightly overstates staffing.
- In concurrent channels. In chat, agents handle several conversations at once, which the formula wasn't designed for.
- When the forecast itself is wrong. The formula is only as accurate as its inputs. A bad forecast for Ramadan or billing cycles produces a bad answer, however precise the math.
What this means for your team
Before your next budget meeting, calculate the workload for the busiest half hour of your day, run Erlang C on it, then apply your actual shrinkage rather than an assumed one. That number, with a table showing the effect of each agent, is far stronger than a general request for "more people."