Forecasting that shows its work.
Workforce forecasts with honest uncertainty, built into the scheduling tool you already run. Know whatโs coming. Staff for what could.
Ten models compete. Your data picks the winner.
Soon doesnโt bet your schedule on one method. Ten forecasting models are backtested against your own history, and the one that would have predicted your past best gets to predict your future.
- Rolling-origin backtests run on your own history, not a benchmark dataset
- Blends are used only when they beat the best single model out-of-sample
- Per-line model attribution: every forecast tells you who drew it
Backtest leaderboard
rolling origin ยท your history
WAPE, out-of-sample. Lower is better โ the blend only ships when it wins.
Honest about uncertainty
A single line is a guess wearing a suit. Soon draws the band around it: confidence intervals measured from the modelโs real errors on your data, separately for each time of day โ because Monday 9am doesnโt behave like Sunday 3am.
Plan on the p50 when youโre optimizing cost. Staff to the p90 when missing service level is the expensive mistake. The band makes that a decision, not an accident.
Measured error, per time-of-day
From calls to headcount
A demand curve isnโt a plan. Soon converts forecasted volume into required agents with exact Erlang C and Erlang A queueing math, then stress-tests the plan by simulating 2,000 weeks of traffic in under a second.
The output is the question every planner actually has: how likely am I to hit service level each day โ and what does one more agent buy me? The budget-vs-service-level curve shows exactly where the next euro stops working.
Exact Erlang C/A
validated against reference implementations to 14 decimal places
2,000 weeks
simulated in under a second
Daily attainment
the probability you hit your service level, day by day
Budget vs. service level
simulated ยท 2,000 weeks
Every point is a staffing plan. The flat part is money that buys nothing.
Grab the levers. Feel the consequences.
What if volume jumps 10% next month? What if handle time drops after the new macros ship? Drag the levers โ volume, handle time, targets โ and watch risk update while the slider is still moving.
Enterprise suites batch this into simulation jobs and make you wait minutes per answer. Soon renders the consequence in the same motion as the question.
Open the Scenario StudioTry it before you import anything
The Sandbox ships with four sample operations, forecast by the real engine โ not a canned demo. Walk the whole loop from data to forecast to staffing to risk, flip models yourself, and see how the leaderboard reacts.
Contact centre
sample operation
Support desk
sample operation
Back office
sample operation
Field ops
sample operation
- Data
- Forecast
- Staffing
- Risk
Plays well with your stack
Actuals flow in from the platforms you already run, so the forecast keeps learning from what really happened.
Live today
On the roadmap
In live comparisons on real customer traffic, up to 46% lower forecast error than the industry-standard N-week average.
And one product owns the loop: forecast โ staffing โ schedule.
Your next schedule could take 2 minutes.
Import your team, set your rules, hit auto-fill. Most teams are live the same day.
30 days free ยท No credit card required
Already have an account?Sign in
