WFM Forecaster
Forecasts calls, chats, handle times, abandon rate and shrinkage from your daily history, then works out staffing per interval. Runs fully offline; your data never leaves this browser.
Daily CSV columns: date (YYYY-MM-DD or M/D/YYYY), calls, chats, aht, chat_aht, abandon_rate, shrinkage. Any metric column can be left out; missing days are filled in. Calls and chats should be offered (answered + abandoned): a "calls_offered" or "offered" column works, or give "answered" and "abandoned" and the app adds them (and works out abandon_rate if missing).
Holidays & events
One line per day: date, % change to calls, % change to chats (blank = same as calls), note. Use -100 for closed.
Let the app learn the effect: list the holiday's past dates too (e.g. last two Black Fridays) with the same note, and leave the % blank on the future date. The app compares those past days with a normal day and applies the average. Without a note, it matches the same calendar date (good for Christmas).
Past event days are left out when learning the normal weekly pattern, so spikes don't distort the rest of the forecast.
Let the app learn the effect: list the holiday's past dates too (e.g. last two Black Fridays) with the same note, and leave the % blank on the future date. The app compares those past days with a normal day and applies the average. Without a note, it matches the same calendar date (good for Christmas).
Past event days are left out when learning the normal weekly pattern, so spikes don't distort the rest of the forecast.
Weather & economic drivers
A CSV with a date column plus any numeric drivers, e.g.
Caution: slow-moving drivers (unemployment, CPI) look a lot like the general trend, so with 1–2 years of history their measured effect can be misleading. Fast-moving ones (rain, storms, promotions) are measured well.
date,rain_mm,temp_c,storm_alert,unemployment_pct,cpi,marketing_spend. Include past dates (so the app can measure each driver's effect) and future dates (your weather forecast or economic assumptions). Each metric uses drivers only if they make its backtest more accurate. Future dates without values use the last 28 days' average.Caution: slow-moving drivers (unemployment, CPI) look a lot like the general trend, so with 1–2 years of history their measured effect can be misleading. Fast-moving ones (rain, storms, promotions) are measured well.
Staffing (Erlang A)
Patience = how long a typical caller waits before hanging up; set 0 to ignore abandons (Erlang C). Chat handle time is per chat while multitasking and is only used when your CSV has no chat_aht column. Without an intraday CSV, volume is spread over opening hours with a typical late-morning peak. An intraday CSV with date, time (HH:MM) and calls columns gives each weekday its own pattern and sets the hours.
Agents by interval (selected day)
scheduled (incl. shrinkage)required (logged in)
FTE per day (whole forecast)
FTEselected day — click a bar to select