Capacity Planning: Models and accuracy
Objective. Read which forecasting model is in use and its backtested accuracy and bias, treating a good backtest as past fit, not a future guarantee.
This screen answers: which model produces our forecast and how well has it predicted in the past? It is where the forecast earns - or loses - the right to be trusted.
See it in the demo
What this screen answers
The customer question is: what forecasting method is driving the plan and how accurate has it been. A CSM uses it to decide how much to trust the forecast before a plan is built on it; a rep uses it to show the forecast carries its own credibility evidence.
The task
View the model in use for the scope and read its backtested accuracy and bias. The job is to judge trust: is the model consistently over or under, and by how much.
How to read it
Accuracy here is backtested on history - it describes how well the model fit the past, not a guarantee for the next period. Bias (a consistent lean over or under) matters as much as the size of the error, because a biased forecast systematically over- or under-staffs. The classic misread is to treat a high historical accuracy as a promise the next forecast will be right. A clean backtest also cannot cover a regime change - a new client, a new product, an unmodelled event.
Honest limits
The specific methods and the accuracy metric used must be confirmed with product. A backtest is retrospective; it does not certify the future, and it is not an executed outcome. Where the customer is about to change their operation, say plainly that past accuracy may not carry.
Interpretation check
Model accuracy is backtested on history - it describes past fit, not a guarantee, and bias matters as much as error size.
Practical task
For a chosen scope, read the model's backtested accuracy and bias and write one line on how much trust the forecast earns - and one condition under which that trust would not carry.
Self-check
A customer says the model is 95 percent accurate so the forecast is reliable. What do you add?
That figure is backtested on history and describes past fit, not a guarantee for the next period - and bias (consistent over/under) can hurt staffing even at high accuracy. A regime change can break it.
Why does forecast bias matter as much as error size?
A biased model leans the same way every period, so it systematically over- or under-staffs even when average error looks small.
When should you warn that past accuracy may not hold?
When the customer is changing the operation - a new client, product or event the model has never seen. Backtests cannot cover a regime it has not observed.
Verified vs demo. Screen is live in the demo at /#/fcmodels; exact controls and figures are SME-validation-pending.