Workforce management for teams too small for Erlang C

The Erlang C calculator is very sure of itself. Feed it your contact volume, your handle time, and your service-level target, and it hands back a precise, confident number: you need 3.7 agents. You have four. So you feel briefly, wonderfully staffed — right up until Thursday, when one person is out with a migraine and your queue quietly becomes fiction. The calculator wasn't wrong. It was answering a question a team your size never actually gets to ask.

The math is built for a bigger room

Erlang C and the workforce-management tools built on it assume the regularities of large numbers: contacts that arrive in a smooth statistical drizzle, a single pooled queue, agents who are interchangeable, and customers who wait patiently forever. At 150 contacts an hour those assumptions are roughly true. At six an hour they're a joke — your arrivals are lumpy, your agents each own something nobody else knows, and your customers abandon in minutes.

Worse, the economics run against you. Queues enjoy brutal economies of scale: the bigger the room, the more efficiently it absorbs randomness. A small team pays for that in occupancy it can never recover.

That 60% isn't slack; it's physics. The efficient band that large centres optimise into —

— is simply out of reach for you, because you can't run a four-person queue near 83% occupancy without it shattering the first day someone's out. And it will shatter, because small teams are fragile in a way big ones aren't.

One absence is a rounding error at fifty seats and a crisis at five.

Schedule coverage, not headcount

So stop trying to run a 200-seat staffing model in a four-person spreadsheet. Here's what actually works when you're below the threshold WFM software is built for.

Forecast the shape, not the total. You don't need a model; you need last quarter's history and the discipline to stop staffing to the average. You already know Monday is heavy and the post-release week is worse. Map the day-of-week and hour-of-day curve — forecasting without a data-science team walks through the spreadsheet — and staff the peaks, not the mean.

Define coverage windows, then put a floor under each. Decide the hours a customer can reach a human, and be honest about the hours they can't. A queue with a clear "back at 9am" beats a channel that looks open and answers on Tuesday. Deliberate gaps are fine; the accidental ones are what customers remember.

Staff for shrinkage, not for seats. The single most common small-team error is scheduling five people as if five people were available.

A five-person team is really about 3.3 people of live capacity once you subtract PTO, breaks, meetings, training, and the odd sick day. Schedule against that number or you will be permanently, avoidably surprised. (Shrinkage is the line item nobody wants to fund and everyone pays for.)

Stagger starts to cover the shoulders. If everyone begins at 9:00, the pre-9 and post-5 waves hit an empty queue. Offset start times by an hour or two so the peak's edges are covered without adding a single head.

Build a deferrable-work buffer. Keep a stack of real but droppable work — knowledge-base upkeep, QA, backlog cleanup — that flexes around live volume. It's the small-team substitute for an occupancy cushion: when the queue spikes, people surface; when it's quiet, the KB gets better instead of the team getting bored.

Cross-train against your single points of failure. If one person is the only one who understands billing, billing goes dark every time they take a Friday. Redundancy on the two or three highest-stakes areas buys more resilience than any scheduling trick.

The pressure only points one way:

Expectations don't scale down for small teams. You can't be everywhere, but you can choose your gaps on purpose instead of discovering them in the queue. That's the whole discipline: not replicating a big centre's math, but refusing to pretend you're a big centre. Small teams win on the things scale can't buy — flexibility, cross-training, judgment — and lose the moment they let a calculator built for someone else's room set their schedule. When you're ready to put numbers to it, the benchmarks give you targets you can actually hold.