At every level of the hierarchy — total customer, product family, or individual item — you choose which forecasting model to use as your base. The model can be different at every level and is saved as your configured strategy for that node in the hierarchy.
4.1 The Two Base Model Types
Model Type | How It Works |
Sales Forecast | Uses the Sales Forecast from the Sales Forecasting module as the base. Reflects the commercial team’s assumptions about promotions, new distribution, and lift. Best for new items, new customers, or situations where you want to fully align with the sales team’s view. |
Statistical Forecast | Runs a model on your historical shipment data (baselined or raw) to project future shipments. Confido evaluates multiple models and suggests the best fit based on historical accuracy. Best for mature, steady-state customers with consistent shipment patterns. |
Select the best fit strategy in your demand planning workspace.
4.2 Statistical Model Options
When you select a statistical forecast, Confido evaluates several model types and recommends the best fit based on which would have been most accurate historically (lowest MAPE). You can accept the recommendation or manually select a specific model:
Best Fit — Confido selects the most accurate model for that customer/item automatically, applied at the item level within the customer
Moving Average — smooths out recent shipment trends; useful for steady, low-volatility customers. The lookback period can be configured by the demand planner.
Single Exponential Smoothing — weights recent data more heavily; responsive to recent trends. The alpha determines the weight this model places on the most recent periods. It is a value between 0 and 1, with 1 placing the highest weight on the latest periods.
Seasonal Naive — repeats last year’s pattern; useful for highly seasonal businesses with predictable year-over-year patterns
Holt-Winters — handles both trend and seasonality; good for growing brands with seasonal patterns
Croston’s Method — designed for intermittent or sporadic demand; good for slow-moving items
Prophet — probabilistic model; good for business time series with strong seasonal patterns
For each model, based on backtesting and cross-validation, Confido can display accuracy metrics including MAPE (Mean Absolute Percentage Error) and bias, so you can evaluate which model has performed best historically before committing to it. Additionally, when you click “Compare methods”, the different volume forecasts will appear in a separate table in the demand planning workspace.
|
4.3 Setting and Saving Model Strategies
When you select a model at any level of the hierarchy, it is saved as your configured strategy for that customer/product combination. The workspace always shows you which model is currently in use at the level you are viewing. If different items within a customer are using different models, the customer-level display will show “Multiple Models”.
In the forecast settings menu (click the settings icon in the DP workspace → Forecast Strategies → Add Strategy), you can set rules to automate model selection over time. For example: use the Sales Forecast for any customer or item with less than 4 months of shipment history, then automatically switch to the best-fit statistical model once that threshold is met (this example falls under the multi-horizon strategy). Furthermore, you can scope the forecast strategy to apply only to a specific product family or product, as well as a specific demand planning group or customer. This allows the demand plan to evolve naturally as your business matures, without requiring manual intervention for every new item or customer.