Demand and sales forecast
A key module of the platform used for general budgeting, procurement, production, integrated planning, etc.
A flexible system of settings:
- Time scale selection: day, week, month
- Grouping options: by store, by goods, by sku, by combination of store - goods, the entire chain of stores, etc.
- Different units of measurement: items or money
- Setting the planning horizon: X days ahead, Y weeks ahead, Z months ahead
The forecasting system takes into account the following factors:
- Retrospective of data on sales, stock: goods are divided into new and old, and different forecast algorithms are applied to each group
- Availability, duration and type of promotions: the user has the opportunity to download the historical and planned promotional calendar and include it in the forecast module
- Seasonality and temperature dependence of goods: we collect external data, such as an extended production calendar and a calendar of unofficial holidays, historical and forecast weather (temperature, precipitation) - and include them in factors that affect the forecast
- Taking into account current processes: accounting for substitute goods, accounting for reconstructions and closures of outlets and stores, accounting for out-of-stock situations
A wide range of forecasting methods are used, from the simplest to the most advanced:
- Naive
- Average and average weekly
- Linear Regressions with Regularization
- Decision Trees
- boosting
- Forests with automatically tuned hyperparameters
The forecasting system works with high accuracy:
- A system for clustering goods and outlets by demand has been developed
- It is possible to group similar forecast units, increase the training sample for the algorithm and make the forecast more reliable.
- One optimal model is built for each cluster, which is selected from a variety of models by maximizing the accuracy on the test sample
Providing the results of the forecasting system:
- As a result of model training, a downloadable final file is generated from forecasts in the required sections
- There is a visualization of the forecast with highlighting the accuracy on the test period
- All forecasts are saved in the history of forecasts, and the user can open and see old results at any time
