The “Analysis” page is intended for building predictive models and for conducting a detailed examination of the factors that influence the formation of demand forecasts.

Open

The page interface consists of two main sections:

• Parameters — used to select and configure the key forecasting settings.

• Results — displays the outcomes of the generated forecast and provides the ability to analyze the factors that influenced its formation.

Open

MySales uses a high-performance mathematical forecasting engine, validated on hundreds of millions of “product–store” combinations.
The system automatically generates consistently high-quality forecasts for each item in every retail location, taking into account historical data and influencing factors.

The forecast is generated using a hierarchical approach — from more aggregated levels to more granular ones:

• Product category — entire network

• Product category — region

• Product category — store

• SKU — entire network

• SKU — region

• SKU — store

This approach makes it possible to leverage demand patterns at different scales and ensures forecast stability even when historical sales data is limited.

With each recalculation, the system analyzes up to 3–4 years of historical data, including:

• actual sales

• prices and promotional activities

• inventory levels

• product hierarchy and reference data

• geographical structure of the network

• external factors (weather conditions, macroeconomic indicators, competitive environment)

Depending on the scale of the network, the volume of processed data can range from gigabytes to terabytes.

The forecast generation process is automatic, multi-step, and iterative.
The main stages include:

The system loads historical data from databases and data warehouses, performs validation, and normalizes it.
For new products or new stores, data from similar items is used for periods where no own historical data is available.

Historical sales are adjusted to account for factors that distort baseline demand:

• promotional activities

• price changes

• seasonal peaks

• lost sales due to stockouts

This allows the system to determine the true level of demand.

The system determines:

1. the nature of seasonal fluctuations (additive or multiplicative)

2. long-term demand trends

3. average and median sales values

Seasonality of the average price at the product group level is also analyzed.

The system evaluates the dependency of sales on key factors:

1. price levels

2. promotions and discounts

3. weather conditions

4. macroeconomic indicators

5. cross-product interactions

If dependencies cannot be determined at a detailed level, data from more aggregated forecasting levels is used.

The system automatically:

1. calculates correlation coefficients

2. tests various forecasting models

3. selects the model that provides the highest accuracy based on historical data

A baseline sales forecast is then generated.

The forecast is adjusted to account for:

• planned promotional activities

• expected price changes

• forecasted values of external factors