Tuning Demantra's Statistical Engine can be a complex, time consuming and frustrating process but with suitable resource, data and direction it is possible to create an improved forecast or validate that existing settings are optimal. An extremely useful by-product of forecast tuning is that it will provide you with a greater understanding of your data. Tuning requires analysis of models, parameters and data usage through the forecast tree. Data segmentation, seasonality, sales and marketing bias, number accuracy of overrides, forecast horizons & levels, model usage, proportionality, product lifecycle... This knowledge can be viewed through Pareto to determine where resources (engine, planners & managers) should focus their efforts to maximise volume, value, margin and time.
The key Engine Tuning elements can be summarised thus:
- Demand Data Profiles (history)
- Node Processing (forecast decision and generation process)
- Forecast Tree (hierarchy levels used by engine)
- Engine Profiles (statistical engine model settings)
- Engine Parameters (methodology settings)
- Causals / Promotions (history and future effects)
- Proport Function (allocation & aggregation)
- Nodal Tuning (Settings per combination)
- Procedures (methodology and approach to tuning)
We can perform engine tuning on your behalf or train resources to become skilled in the science and art of creating improved forecasts. Engine Tuning can only be properly undertaken with access to all the administrative components of Demantra and preferably a dedicated Tuning environment.