Plan against a range, not a guess.
Forecasts built from your own history, presented with the uncertainty visible so you can plan for the spread rather than a single number.
Useful when plans are built on one number that everyone privately knows is wrong.
Illustrative system example · fictional product and data · not client work
Forecast consoleHORIZON · illustrative system, fictional data
You may need this when
Some of this probably sounds familiar.
- Planning uses a single figure with no sense of how wrong it might be.
- Stock or staffing is set by last year plus a percentage.
- You are surprised by the same seasonal pattern every year.
- Nobody can say how accurate the last forecast turned out to be.
- A supplier needs commitments further ahead than you can confidently see.
What we could build
A forecast you can plan against.
Including an honest statement of how far ahead it stays useful.
- Demand forecastExpected volume with a range, at the granularity you plan at.
- Capacity and staffing modelWhat the forecast implies for people and shifts.
- Inventory forecastingReorder points that reflect variability, not just averages.
- Backtesting harnessHow the model would have performed on your real history.
- Scenario forecastsThe same model under different assumptions.
- MonitoringAlerts when reality drifts away from the forecast.
See an example
Look further ahead. Watch the confidence widen.
Change the horizon. The forecast stays honest by widening — a number that stays equally precise twelve months out is not telling you the truth.
Every option is shown below. The selector needs JavaScript, so it is not offered here.
OutcomeSix weeks out the range is tight enough to commit stock against.
OutcomeA quarter out the spread is wide enough that staffing should be flexible, not fixed.
OutcomeAt twelve months this is a direction of travel, not a number to sign a contract against.
Illustrative interfaces and fictional data. Not client work.
Planning against the range rather than the midpoint is what stops a forecast becoming a promise nobody agreed to.
What you receive
A forecast, and its track record.
- A working forecast produced on your schedule
- Backtest results showing how it performs on your own history
- Prediction intervals, not just a central estimate
- A stated horizon beyond which the forecast should not be relied on
- Agreed source code, data and infrastructure handed over according to the project scope and terms
- Handover documentation and any agreed support arrangements
Technical depth
Method
Statistical and machine-learning approaches compared on your data, with the simplest model that performs chosen rather than the most sophisticated.
Backtesting
Rolling-origin evaluation on held-out history, reported as error metrics you can interpret.
Uncertainty
Prediction intervals produced and carried into the interface, so the range is as visible as the point estimate.
Features
Seasonality, holidays, promotions and known events modelled explicitly where they drive the series.
Drift
Monitoring for when live accuracy degrades, with a defined trigger for retraining.
Limits
The horizon and conditions under which the forecast stops being useful stated plainly.
Often combined with
What usually sits next to this.
- Data & analytics →When the history needs assembling and cleaning first.
- Simulation & modelling →When you need to test a decision, not just predict a level.
- Automation & integrations →When the forecast should feed the systems that act on it.
Tell us what you are planning against.
Describe the decision the forecast would inform and how far ahead you need to see.