Skip to content
Chatbotscape
Call Centre Helper's Erlang C worked example, its articles on Erlang C for chat, chat capacity planning and Erlang A, and Assembled's article on planning staffing with Erlang C were read on 16 November 2026. The worked example below was recalculated independently on the same date.
Erlang C· Support operations
Erlang C is a queueing formula that estimates how likely a customer is to wait, and how many contacts are answered within a target time, given how many contacts arrive, how long each takes and how many agents are available. Support teams use it to work out the smallest team that meets a service level such as 80 percent answered within 60 seconds.
By Chatbotscape Editorial· Methodology· Published 17 November 2026· Updated 17 November 2026

Erlang C — How Many Agents a Response Time Target Needs

Quick answer: Erlang C takes four inputs (contacts per period, average handle time, the number of agents, and a target answer time) and returns the probability that a contact has to wait, the share answered inside the target, and the average wait. You raise the number of agents until the share meets your goal, then add extra people for breaks, meetings and absence, known as shrinkage. It was built for phone calls, so for live chat, where one agent handles several conversations at once, practitioners adjust the inputs.

Where the name comes from

The formula is named after Agner Krarup Erlang, a Danish engineer who studied telephone exchanges. Assembled dates the formula to 1917, when it was used to work out how many switchboard operators a phone exchange needed. An "Erlang" is also a unit: one Erlang is one hour of work arriving per hour, so 10 Erlangs means enough work to keep 10 people busy non-stop. Call Centre Helper describes three related formulas: Erlang B, for counting phone lines, which it says is rarely used in contact centers now; Erlang C, for staffing queues where customers wait; and Erlang A, which adds customers who give up.

The inputs and what they mean

InputWhat it isWhere small teams get it
VolumeContacts arriving in a period, usually an hour or 30 minutesHelp desk reports, by hour of day
Average handle timeAverage minutes an agent spends per contact, including wrap-upHelp desk reports; see average handle time
Target answer timeThe wait you aim to stay under, such as 60 secondsYour SLA
Service level goalThe share of contacts that should meet the target, such as 80%Your own decision
ShrinkagePaid time agents are not available: breaks, meetings, training, absenceRota and HR records
Maximum occupancyThe highest share of time you want agents busyYour own decision

Only the first four go into the formula itself. Shrinkage and maximum occupancy are applied afterwards.

What it calculates

  • Traffic intensity (A). Contacts per hour times handle time in hours. 200 calls of 3 minutes each is 600 minutes of work per hour, or 10 Erlangs.
  • Probability of waiting. The chance a new contact finds every agent busy. This is the core Erlang C result.
  • Service level. The share answered within the target time, calculated from the probability of waiting, the gap between agents and traffic, and the target compared with handle time.
  • Average speed of answer. The average wait across all contacts, including those answered at once.
  • Occupancy. Traffic divided by agents: the share of time agents are busy with contacts.

The number of agents must be larger than the traffic intensity. With 10 Erlangs of work and 10 agents, the queue never settles and waits grow without limit.

A worked example, checked

Call Centre Helper's worked example uses 100 calls per 30 minutes, a 3-minute handle time, and a goal of 80 percent answered within 20 seconds. We recalculated each step:

StepResult
Calls per hour100 × 2 = 200
Traffic intensity200 × 3 ÷ 60 = 10 Erlangs
Service level with 13 agents79.6%, just short of 80%
Service level with 14 agents88.8%, meets the goal
Probability of waiting, 14 agents17.4%
Average speed of answer, 14 agents7.8 seconds
Occupancy10 ÷ 14 = 71.4%
Agents to schedule with 30% shrinkage14 ÷ 0.7 = 20

The jump from 79.6 percent to 88.8 percent with one extra agent shows why Erlang C answers are always rounded up to whole people, and why small teams feel each absence sharply. Call Centre Helper suggests checking occupancy against a ceiling (85 percent in its example) because very busy agents burn out and handle times rise.

Erlang C for live chat

Erlang C assumes one agent handles one contact at a time. A chat agent often works two or three conversations at once (see chat concurrency), so the raw formula overstates the team needed. There is no agreed fix, and two Call Centre Helper articles describe different shortcuts:

  • Shorten the handle time. Divide handle time by a concurrency factor to get an effective handle time, then run Erlang C as normal. Its suggested factors are lower than the chat count because agents are not always at their limit: 1.7 for two chats, 2.5 for three, 2.9 for four. A 600-second chat at three-at-once becomes 240 seconds.
  • Divide the result. Run Erlang C as if chats were calls, then divide the agents by the average number of chats an agent actually handles. In its example, 51 agents divided by 1.6 gives about 32.

Our response time SLA calculator uses a third common approximation: it multiplies agents by the chats each handles. All three are estimates. Call Centre Helper calls the factor approach "not ideal" but the most widely used, and notes that handle time tends to rise as concurrency rises, so test the factor against your own data.

Limits and assumptions

Erlang C treats arrivals as random and evenly spread within the period, handle times as varying in a regular pattern, and every agent as able to take every contact. It also assumes nobody leaves the queue. Erlang A adds abandonment using the customers' average patience; Call Centre Helper notes that Erlang A tends to underestimate staff and suggests using both. Assembled warns that Erlang C is not suited to email, where replies take hours or days, and that a forecast off by 25 percent can leave a team several people short or over. For a team with separate skills or languages, run each group separately, as chat routing would split them.

Where chatbots fit

A bot changes two inputs. It lowers volume when it answers questions fully, which is your deflection rate. It can also change handle time: when a bot collects an order number before the human handoff, agents may finish faster, but the chats that reach people are often the harder ones, so handle time can go up. Run Erlang C on the chats that actually reach people, not on all conversations, and measure the human wait separately from the bot's instant reply, as explained in first response time.

FAQ

What is the Erlang C formula used for? To estimate how many agents are needed to answer a given share of contacts within a target time, based on volume and handle time.

What inputs does an Erlang C calculator need? Contacts per period, average handle time, target answer time and a service level goal. Most calculators also ask for shrinkage and maximum occupancy, which are applied after the formula.

Does Erlang C include shrinkage? No. The formula gives agents needed at the moment. You divide by one minus shrinkage to get agents to schedule: 14 agents with 30% shrinkage means 14 ÷ 0.7 = 20.

Can Erlang C be used for live chat? Yes, with an adjustment for agents handling several chats at once. The usual shortcuts shorten handle time or divide the result by a concurrency factor; both are estimates to check against real data.

What is the difference between Erlang B, C and A? Erlang B counts lines needed when blocked calls are lost. Erlang C staffs a queue where customers wait. Erlang A extends Erlang C with customers who abandon.

Is Erlang C accurate for small teams? It is a useful starting point, but each whole agent moves the result a lot at small sizes, and absences, breaks and uneven arrivals matter more. Check the forecast against real waits every week.

Sources