How many agents do you need?

Erlang C is the classic contact-center staffing formula. Given how many calls arrive in an interval and how long each one takes, it estimates the chance a caller waits and how long they wait for a given number of agents. It assumes calls arrive randomly, every caller waits as long as it takes (no one hangs up), and any agent can take any call.

Erlang A adds abandonment: callers have a limited patience and some hang up if the wait is too long. That is closer to real life. Because abandoned calls leave the queue, Erlang A usually needs a few fewer agents than Erlang C for the same target, and it also tells you how many calls you should expect to lose.

Use Erlang C when you want a conservative, widely understood number, when abandonment is very low, or when you need to match a WFM tool that uses it. Use Erlang A when abandonment is meaningful (typically above 3–5%), for busy queues where Erlang C tends to overstaff, or when you want to plan around an abandon-rate target.

Service level is the share of calls answered within your threshold, such as 80% within 20 seconds. In Erlang A, it is measured against all offered calls, so calls that abandon count against you. Some ACD reports exclude short abandons or measure against answered calls only, so your platform's number may read a little higher.

Shrinkage is the share of paid time agents are not available to take contacts: breaks, lunches, meetings, coaching, training, PTO and unplanned absence. If you need 14 agents on the phones and shrinkage is 30%, you need to schedule 14 ÷ (1 − 0.30) = 20 agents. Typical centers run 25–35%.

Erlang C and Erlang A assume each agent handles one contact at a time. Chat agents juggle several, but each chat goes slower while they do. Patreon supporters can pick concurrent chats per agent (N) and an efficiency factor; we then treat each agent as N × efficiency agents' worth of capacity by using an effective handle time of AHT ÷ (N × efficiency). With 2 chats at 85%, a 10-minute chat counts as about 5m 53s of agent time. Workload and occupancy come out right on average, but wait times and service level are rougher than for voice, because a chat really waits for a free slot, not a free agent. Use it as a planning estimate and check it against your chat history.

Which handle time should I enter? Most chat platforms report handle time measured while agents juggle chats, open to close. That number already includes the slowdown, so choose "Platform-reported": we then use AHT ÷ N and skip the efficiency factor. Counting efficiency on top would double-count the slowdown and understaff. Choose "Solo (one chat at a time)" only if your AHT comes from one-at-a-time handling (for example, time studies or a period when concurrency was 1).

At high volume, the math can hit service level with agents busy 92% of the time or more. That is hard to sustain and tends to raise AHT, errors and attrition. A max-occupancy cap (often 85–90%) adds agents until the load is humane.

Join the Fratz WFM tier (or All Access) on Patreon, then come back and click Log in with Patreon. Supporter features unlock in that browser; on another device, log in with Patreon there too. Patreon handles billing, receipts, cancellation and refunds. We only receive your Patreon first name, tier and membership status. More on the support page.

In your browser, and nowhere else. Every calculation runs on your device, and your inputs, imports, scenarios and settings are saved only in this browser's storage. Nothing is uploaded to us. To back up your work or move it to another device, use Export all my scenarios in the Day Planner, then import the file on the other device. Clearing this site's data in your browser removes everything saved here. See the privacy page for details.

These are estimates, not guarantees. Erlang models assume steady, random arrivals within the interval, a single skill, and an exponential distribution of handle time and patience. Real queues have spikes, multi-skill routing and outages. Use the results as a planning baseline, check them against your actual history, and leave headroom for what the math can't see.