Staffing and Forecasting for Support: An Erlang-Free Primer

Most support staffing advice opens with the Erlang C formula and loses you by the second variable. You don't need it to get this right. Support staffing forecasting is mostly arithmetic plus honesty about how people actually work, and you can do the core of it on one screen without a queuing-theory degree.

Here's the practical version: predict the volume, account for the time people don't spend on the queue, and don't run your team at a pace that guarantees burnout.

Start with the volume drivers, not last year's average

Forecasting begins with predicting how many contacts will arrive — and "we got 4,000 last month, so plan for 4,000" is where most forecasts go wrong. Volume isn't flat. It's driven by things you can see coming.

The drivers worth modeling:

• Customer base size. More customers, more contacts. Tie volume to your active-customer count, not a raw historical number.
• Contact rate per customer. Contacts divided by customers. If this is climbing, something upstream is broken — a product issue, a confusing policy, a billing change.
• Seasonality and events. A promotion, a launch, a holiday, a price change. These are predictable spikes; put them on the calendar before they hit.
• Known one-offs. A shipping delay, an outage, a recall. You often know about these days ahead. Forecast them in.

Forecast at the level you'll staff at — by channel and ideally by hour, because staffing breaks at the peak hour, not the daily total. A day that averages comfortably can still drown you at 2pm.

Two numbers everyone forgets: shrinkage and occupancy

This is where back-of-the-envelope forecasts fall apart. A rep on payroll is not a rep on the queue.

Shrinkage is the share of paid time not spent available for contacts — breaks, lunch, meetings, training, coaching, admin, time off. It runs 25–35% on most teams. If you ignore it, you'll plan for 10 reps' worth of capacity and field about 7.

Occupancy is the share of available time reps spend actively handling contacts. It feels like you'd want this near 100%, but you don't. A team at 95% occupancy has no slack to absorb a spike, every queue backs up, and people burn out fast. Healthy synchronous teams sit around 80–85%, leaving real room to breathe.

If your forecast assumes everyone is available all day and busy every minute, it isn't a forecast. It's a recipe for missed SLAs and turnover.

The plain-English staffing math

You can size a team with four numbers and no special formula:

So roughly 20 reps scheduled to safely cover a peak hour that looks, on paper, like it only needs 12. The gap between those two numbers — 12 versus 20 — is exactly what shrinkage and occupancy account for, and it's exactly what naive forecasts miss. Synchronous channels (chat, voice) need this buffer most; asynchronous email can run leaner because contacts can wait in a queue without anyone feeling abandoned.

Do this per channel, per peak window, and you have a real schedule instead of a hopeful one.

How AI deflection changes the math

AI agents don't just "reduce volume." They reshape the curve, and the forecast has to account for that:

1. They cut the resolvable, low-variance volume first. Order status, password resets, return labels. Subtract the deflected contacts from your forecast before you size the human team — but only the ones that truly resolved and didn't reopen.
2. They flatten peaks more than averages. AI agents answer instantly at 2pm and 2am alike, so they shave the spikes hardest — which is precisely where staffing strained most. That can let you trim the peak-hour buffer.
3. They raise your team's average handle time, on purpose. What's left for people is the complex, variable work. Your human AHT will rise, and your forecast should expect it, not flag it as a regression.
4. Reopens belong back in the forecast. A deflection that bounces back is volume that returns to the human queue. Forecast from true deflection — resolved and stayed resolved — not from "AI touched it."

Get this right and the staffing math gets easier, not harder: a smaller, more predictable human workload sitting behind an AI layer that absorbs the noisy peaks.

In BearScope, AI agents carry the resolvable volume and leave a receipt for every action, true deflection is measured net of reopens, and contact volume is broken out by channel and intent — so you can forecast from the work that's actually left for people. See how AI agents and your team share the queue, or book a walkthrough to model your own staffing curve.

Predict the drivers, respect shrinkage and occupancy, and net out true deflection. That's the whole forecast — no Erlang required.