Forecasting support volume with Erlang C
Support teams miss service levels because they are staffed to a monthly average. Contact volume is not an average, and the maths for this is a century old.
A support floor that misses its service level every month is usually not lazy or badly managed. It is staffed to the wrong shape. Somebody took last quarter's contacts, divided by the working days, and rostered against the result.
Contact volume does not behave like an average. It follows your release calendar, your campaign calendar and your delivery promises. Monday is not Thursday. The day after an email send is not the day before it. Staffing to the mean guarantees you are overstaffed half the time and failing the other half.
The tool for this is older than the industry it serves. Erlang C, developed for telephone exchanges in 1917, takes arrival rate, handling time and target answer time and tells you how many agents you need. It is not exotic; it is a spreadsheet function.
The inputs are where the work is. You need historical contact data at half-hour granularity, a realistic handling time per contact type, and an honest shrinkage figure covering breaks, training, sickness and everything else that stops a rostered agent from being available. Teams routinely assume shrinkage of ten percent when the real number is thirty.
Then you re-forecast during the day. A forecast made on Monday is wrong by Wednesday; what makes the difference is noticing at eleven that the queue is running ahead and moving people before the afternoon, not explaining it in the monthly report.
On one account this took service level from missed every month to 82% of calls answered within twenty seconds, without adding headcount. The team was the same size. It was on shift at different times.