Ask a finance director how many agents it takes to lift service level from 80% to 90% and you will usually get a linear assumption back: if 10 agents got us to 80, another 5 should get us close to 90. Erlang C does not work that way. The relationship between headcount and service level is a curve, steep in the middle and almost flat at both ends, and knowing where you are sitting on that curve is one of the most useful things a WFM planner can bring to a staffing conversation.
A worked example
Take a realistic half-hour interval: 500 calls, an AHT of 300 seconds, and a target of answering within 20 seconds. That works out to a traffic intensity of 83.3 Erlangs, which is the theoretical minimum number of agents needed just to handle the work with zero queue tolerance. Here is what actually happens to service level as you staff above that floor:
Three things stand out immediately.
The cliff at the bottom
At 84 agents, barely above the traffic intensity floor, service level is 12.6% and average speed of answer is over six minutes. One extra agent takes you to 28.8%. Another takes you to 42.3%. In this zone every single agent is worth an enormous amount, because the queue is compounding on itself. This is exactly why a shortfall of two or three agents in a peak interval does far more damage than a planner expecting linear behaviour would predict, and why real-time recovery matters so much when an interval starts slipping.
The steep middle, where targets live
Between roughly 88 and 94 agents, each additional agent is still buying meaningful service level, somewhere between two and eight percentage points each. Most operations set their targets right here, and it is where the trade-off conversation is genuinely balanced: an extra body costs real money and delivers a real, visible improvement.
The flat top, where money disappears
Past about 97 agents the curve goes almost horizontal. Going from 100 to 104 agents, four extra full-time equivalents, moves service level from 98.4% to 99.5%. Just over one percentage point for four people. If someone asks why you cannot simply add headcount to guarantee 99.9%, this section of the curve is the answer. You can, technically, but the cost per point of service level becomes absurd, and you have not eliminated risk, you have just bought a very expensive buffer.
Occupancy moves the opposite way
The same calculation has a second story in it. At 84 agents, occupancy sits at 99.2%, which is not a sign of an efficient operation, it is a sign of one about to break. Agents at that occupancy have no recovery time between contacts, and sustained periods there drive both AHT inflation and attrition. By 95 agents, occupancy has dropped to a much healthier 87.7% while service level is at 93.3%. The staffing level that protects your service level is usually also the one that protects your people, which is a useful thing to be able to demonstrate with numbers rather than assert.
What to do with this
When you are asked to model a service level change, do not quote a rate of agents-per-percentage-point, because there is not one, it depends entirely on where the current staffing sits on this curve. Run the actual calculation for your own volume, AHT, and target, and present the shape rather than a single number. A finance conversation goes very differently when it starts with "here is where each extra agent stops paying for itself" instead of "we need more people".
You can generate this curve for your own operation with the Erlang C Staffing Calculator, which runs the same maths used to build the chart above, and factor in shrinkage to convert available-agent requirements into actual scheduled headcount.