Case Study
Operations Analytics
Demand & Sales Forecasting
What demand should be expected, and how can an uncertain forecast support better decisions?
Download PDF
A complementary visual resource for a deeper view of the case.
The business challenge
What demand or sales should be expected by period and decision unit, and what is the cost of different errors?
Forecasts inform stock, capacity, budgets and staffing. An average metric can hide asymmetric costs and uncertainty.
Key business questions
- Which decision will use the forecast, and over what horizon?
- Which patterns, events and constraints explain demand?
- What is the cost of over- and under-forecasting?
Data & evidence
- History at a coherent granularity.
- Calendar, seasonality, promotions and availability.
- Product, location and channel hierarchies.
- Operational changes and external factors.
How I would approach it
- Define horizon, granularity and decision.
- Create baselines and rolling temporal validation.
- Compare models by segment.
- Generate intervals and translate errors into costs.
What you would get
- Forecast and baseline by relevant horizon.
- Uncertainty intervals and scenarios.
- Error, bias and difficult-segment diagnosis.
- Recommendations for stock, capacity or planning.
From analysis to action
Illustrative example: A forecast is a distribution of possible outcomes, not a certain number. Planned inputs do not prove their historical causal effect.
- Link stock or capacity to uncertainty.
- Prioritise review where risk is highest.
- Monitor bias and degradation.
How success should be measured
- MAE/WAPE and bias by horizon.
- Interval coverage.
- Operational cost and value versus baseline.
Other approaches to consider
Naive and ETS methods can outperform complex models; regression, trees or hierarchical models add value when strong drivers exist.