Workforce Management for Insurance Agencies: Erlang C Without the Spreadsheet

The Four Numbers WFM Math Needs

Why Insurance Agencies Need Erlang at All

Erlang C is the formula traffic engineers have used since A. K. Erlang's original 1917 work to size telephone-network capacity, and it answers exactly the question every agency operator asks: given expected call volume, average handle time, and a service-level target, how many agents do I need on the phones to hit it? The peer-reviewed traffic-engineering literature has refined the model over a century, but the agency-relevant version is straightforward enough to run on the back of a napkin once you've seen it twice.

The reason agencies need Erlang and not "agent count = call volume × AHT / hour" is that calls don't arrive evenly. Two hundred calls in an hour with 6-minute AHT does not require 20 agents (200 × 6 / 60 = 20). It requires 24–28 agents to hit an 80/30 service level, because calls cluster and a queue forms. Linear math under-staffs by 20–40% in the windows where it matters most. ICMI's contact-center benchmarks and the broader Erlang traffic literature both make this point: queue dynamics, not throughput math, drive headcount in any meaningful service-level model.

The Four Inputs You Actually Need

Erlang C wants four numbers. The math gets done by a calculator (free, browser-based — search "Erlang C calculator"). Your job as the operator is to feed it the right inputs.

A Worked Example

Imagine an inbound Medicare retention queue during the AEP off-season. Forecast: 120 calls in a 30-minute interval. AHT: 360 seconds. Target: 80% answered within 30 seconds. Shrinkage: 32%.

Step one: convert volume + AHT to Erlangs (workload). 120 calls × 360 seconds = 43,200 seconds of work. Divided by the 1,800 seconds in the interval = 24 Erlangs.

Step two: feed 24 Erlangs and 80/30 into the Erlang C calculator. It returns ~28 agents required on calls (occupancy ~86%).

Step three: gross up for shrinkage. 28 / (1 − 0.32) = 41 agents on schedule. That's the headcount the schedule must put on shift to deliver 28 agents actually available to answer calls in that window.

Where Insurance Calls Break the Generic Model

Erlang C assumes Poisson arrivals and exponentially distributed handle times. Insurance calls violate both — sort of. The variance in AHT between a 90-second wrong-number and a 28-minute Medicare appointment is enormous, and that variance pushes required headcount higher than vanilla Erlang predicts. Two practical adjustments handle this for agency-level planning.

Forecasting Volume When You Have Three Years of History

The single biggest source of WFM error isn't the formula — it's the volume forecast. The default heuristic that works for most agencies: take prior-year same-day volume per 30-minute interval, apply this year's growth rate (typically +5–25% for a growing book), and adjust for known event-driven anomalies (carrier launches, election-cycle Medicare news, holidays). This gets you within 10% on a normal week and within 20% during AEP, which is good enough to plan against.

Agencies with strong inbound queues end up forecasting at the 30-minute interval, the day-of-week, and the week-of-year level. Agencies with mostly outbound campaigns forecast contact attempts and conversion rates instead, but the discipline of looking at the data at-interval rather than at-day is the same. Tying this to your seasonal staffing model is what turns the forecast into actionable hiring.

Shrinkage Is Where Most Agencies Lie to Themselves

Most agencies under-report shrinkage by 10–15 percentage points. They count PTO and breaks but skip training, coaching, team meetings, system issues, payroll questions, AHT spikes from carrier portal slowness, and the time agents spend reading internal Slack messages. Honest measurement means logging every minute paid agents are not actually on calls during scheduled production hours, then dividing by paid hours.

Occupancy: The Counter-Pressure on Service Level

Erlang C will tell you to staff for 80/30 with 86% occupancy. But occupancy above 88–90% destroys agents — burnout, errors, attrition. If your service-level target requires 92% occupancy in peak hours, you're not understaffed; you're over-targeting service level. Either lower the SL target (80/45 instead of 80/30) or hire more agents. The honest math is what tells you which.

Mature WFM functions track three metrics in tension: service level achieved, occupancy, and shrinkage. Moving any one of them moves the others. Operators who treat each independently always end up with a queue that's either bleeding agents or bleeding callers.

When You Don't Need a WFM Platform Yet

Most insurance agencies run their first 100 agents on Erlang plus a spreadsheet. The decision to buy a dedicated WFM tool is usually triggered by one of three things: more than three skill groups to schedule independently, multiple time zones with overlapping shifts, or a fast-arriving compliance audit that requires precise adherence reporting. Below those triggers, the spreadsheet plus an Erlang calculator is enough — provided your dialer and queue platform actually exposes the historical interval data you need to forecast.

Key Takeaways for Agency Operators

Erlang isn't hard. It's just unfamiliar. Spend an afternoon with a calculator and historical data, run last week's actuals through the math, and you'll see exactly where the queue broke and what the schedule should have been. After that, it's just discipline.