Demand forecasting, Erlang staffing, shift rosters and a metrics report, for contact centre planners and operations leads.
Free. Your figures never leave your browser — no signup, nothing you enter is uploaded. The chat assistant is the one part that talks to a server, and only what you type into it is sent.
YYYY-MM, +12%, reason.
Holidays here mean behaviour that survives the working-day count, a Christmas lull or a New Year surge,
not the lost days themselves.The volumes here come from the Volume Forecast tab. This history is used only to measure the day of week shape, which decides where inside each month the volume falls.
Every WFM tool rests on a model. These are the ones behind these numbers, so you can judge where they hold and where they do not.
| What is assumed | Why it matters |
|---|---|
| Service level assumes nobody abandons | Abandonment itself is modelled properly (Erlang A, an M/M/n+M queue), and it is one of the three constraints that can drive headcount. But the service level figure is still Erlang C, which assumes callers wait forever. That makes service level mildly conservative, which is the safe direction. The Erlang A service-level formula was written and deliberately not shipped, because it counts an abandoned caller as a success. |
| Monthly forecast: the pattern holds | The 12 month forecast assumes next year behaves like the last two or three: same seasonal shape, same direction of travel. It cannot know about a contract won, a product recalled or a competitor collapsing. That is what the holiday and external adjustment boxes are for, and using them is not cheating, it is the job. |
| Monthly forecast: two methods, not one | With 24 months or more the tool does full classical decomposition: measured trend, seasonal indices, business cycle and an out-of-sample backtest. With 12 to 23 months it switches to seasonal naive, repeating the same calendar month a year earlier with a working-day correction and a growth rate you supply; in that mode no seasonality and no trend are estimated, because under two years a seasonal peak and a year of growth are the same wiggle and nothing can separate them. Below 12 months no monthly method applies. The panel always states which method produced the numbers you are looking at. |
| At exactly 24 months, every index rests on one observation | Full decomposition becomes possible at 24 months, but each calendar month has a single ratio behind it, so no outlier can be detected and one freak month sets that month's index outright. Three years is materially better, and the tool shows how many observations each index rests on. |
| Trend damping | A straight line extrapolated twelve months out is the classic way to forecast a business into a number it never reaches. Damping below 1.00 flattens the trend as the horizon extends, which is what the forecasting literature consistently finds beats an undamped line. Set it to 1.00 if you want the pure straight line. |
| Poisson arrivals | Contacts arrive randomly and independently. Marketing bursts, outage spikes and scheduled campaigns break this, and those intervals will be understaffed by any Erlang model. |
| Exponential handle times | The maths assumes handle time varies exponentially around the AHT. Very consistent handle times make it slightly pessimistic; very variable ones make it optimistic. |
| Steady state inside the interval | Each interval is solved on its own, with no queue carried in or out. Short intervals with heavy overflow between them will read better here than in reality. |
| One skill, one queue | No multi-skill routing or overflow groups. Pooled multi-skill operations need fewer agents than the sum of separate queues. |
| Occupancy uses handled contacts | Occupancy is handle time over agent time, and abandoned callers are excluded because they never reach an agent. The occupancy ceiling that drives headcount uses the same basis. Measuring it on contacts offered instead reads high by roughly the abandonment rate. |
| Occupancy uses logged-in time | Handle time divided by time actually logged in, not by scheduled hours. Judging occupancy against scheduled hours understates it by roughly the shrinkage rate. |
| Service level is measured on contacts offered | Answered within target divided by offered. Some clients contract it on contacts answered, which produces a flattering number. Check which basis your contract uses. |
Erlang C is the industry standard for staffing a single queue and is what most WFM platforms use. It is a model, not a measurement. Sense-check the output against what your operation actually did.
Chat is not just cheaper voice. An agent runs several chats at once, so each one consumes a fraction of an agent rather than a whole one — and if the two channels share a pool rather than sitting in separate teams, the whole operation needs fewer people. This works out both, and the gap between them is what cross-training is worth.
The Staffing tab tells you how many people this day needs. This asks the next question: what could change that number, and what is each lever measured to be worth? Every figure is produced by re-running the staffing engine over the same day with one change applied — nothing here is a rule of thumb. Where the number cannot practically be improved, it says so and names the reason.