Automatic Forecast System Checklist

Looking for automating sales forecasts for retail chain? Use these questions as part of your vendor research

• Does the system allow for building promo forecasts and simulating various scenarios? Is it taking into account the interplay between product, price, seasonality and promo factors?

• Does the system take into account the effects of seasonality and weather for each combination of store and SKU individually?

• Does the system take into account the impact of price changes on the forecast? Context: Particularly important for fresh goods so that write-offs do not increase after price increases.

• Can the system account for cannibalization in the forecast? Context: Let’s draw an example from eggs where the appearance of a new top position in the group leads to a significant drop in sales of other positions.

• Is the system providing daily forecast recalculation, taking into account the latest daily sales, price change, seasonality and weather data? Is it made separately for promo sales and regular sales?

• Is the forecast calculated according to a top-down or bottom-up approach? Context: Top-down approach means global forecast decomposed to store level. Bottom-up is a separate model for each combination of store and SKU. Bottom-up approach performs better in capturing store specifics. Weather, location in the city, local purchasing power, among other factors.

• Is the promo forecast module allowing for taking into account factors such as marketing activities? How customizable is it?

• Does the system provide insights on why given predictors were chosen in a particular model?

• Is the system building a separate ML model for each combination of store and SKU? Does it support daily or even more frequent recalculations?

• How many competing models is the system building before choosing the best one?

• Is the system allowing for building custom reports?

• Is promo forecast automatically calculated into overall forecast?

• Is it possible to exclude periods of demand anomalies for the sake of future forecasts? Can this be done in bulk or automatically? Context: Covid quarantine, war in Ukraine led to increased purchases of certain categories.

• Is the system allowing for using competitor’s prices as one of the predictors?

• Is the system allowing for customizing for how long into the future and to the past should the forecast be done?

• In the case of a sale (price cut, e.g. due to the upcoming expiry date), is the system able to categorize the record as a sale, so that the increased sales volume does not result in a disproportionate increase in the forecasted sales for subsequent periods?

• Is the system allowing for using competitor prices as a factor in a sales forecast?