Module: Forecasting promotional sales

Executive summary

This module aims to estimate a quantity of a product that will be sold during future and ongoing promotions. The computation is based on the past promotions on the same (or similar) products.

If a promotion spans a several warehouses, the forecasts can be evaluated for individual warehouses – or all of them together, as a group. In computing the forecasted quantity (of the product to be sold), we use data on sales during past promotions.

Functional description

Computing quantity of product to be sold in a future promotion

The algorithm of promotion forecast consists of two parts:

  1. Loading historical promotions
  2. Forecasting quantity of product to be sold

Loading historical promotions

For the accuracy of the forecast, it is vital to choose the right historical promotions. So, all historical promotions are assigned weights that determine similarity with the forecasted promotion; the promotions with a higher weight are – of course – preferred. Thus, it is very important to set the parameters determining a similarity of two promotions correctly. The main criteria for determining similarities are:

  • Promotions are directed to the same customer
  • Promotions are at the same warehouse
  • Promotions are on the same products

Similarity of promotions is – in addition – determined by a duration of the promotion, discounts during the promotion, a leaflet and/or a hostess, placement (and secondary placement) of the given item in a store, etc. Each of these attributes is assigned a weight determining its importance.

A set of historical promotions is cleaned from extremes. A historical promotion can be labeled as extreme already in the input data.

Another setting allows you to consider stockouts and thus to force increases in sales in past promotions. That way, we get more accurate historical sales, thus more accurate forecast for future promotions. Expected sales for a promotion are computed by the module: Analysis of stockouts.

Forecasting quantity of product to be sold

Using a weighted linear regression, based on the weight of various historical promotions, we compute a forecasted quantity of a product to be sold in future promotions.

If there are overlapping promotions, the forecasted quantity might be adjusted.

Eliminating promotions from computation using SIDI

Each promotion can be eliminated from the automatic forecast of quantity (to be sold). This setting can be automatically imported from the primary ERP system.

STOCK features a setting that marks all promotions longer than X days; the default is 40 days. The whole module of promotion analysis is primarily intended for short-term promotions, with the idea that the price reduction (or having the product included in a leaflet) increases sales significantly.

Furthermore, each promotion can be set to be ignored in the selection of historical promotions. Therefore, it is possible to forecast it, but it will never be entered as a test promotion in the algorithm for forecasting future promotions.

This setting is especially useful for promotions with extremely high or low sales due to various unexpected reasons; eliminating such a promotion prevents an unnecessarily inaccuracy in future forecasts.

This setting can be automatically imported from the primary ERP system.

Report: Evaluation of promotions

This report lets the user to define a specific set of promotions and browse all the information related to it.

The set of selected promotions is defined via an input filter, selecting the date from-to (duration of the promotion or stock delivery), the promotion’s type and aggregation:

image0

The table displays (depending on whether the given promotion has been completed, whether it is an ongoing or future promotion, etc.) the following information on:

  • the type of promotion,
  • its duration from-to (both stock delivery and sales periods),
  • the size of discounts (%),
  • non-promotion and promotion prices,
  • estimated changes in sales,
    • estimated quantity to be sold during a single promotion (in quantity and value),
    • the actual quantity sold during a single promotion (in quantity and value),
    • estimated quantity sold in non-promotional sales (in quantity and value),
    • the promotion’s profitability (in value),
  • estimates obtained by computations (the system) and those put forth by a user,
  • total sales during the promotion and after the promotion,
  • stockouts during the promotion and total loss in quantity, revenues and profit.

Given the rather large amount of data available for a promotion, during the implementation period, the teams agree which kind of data are important for the given implementation – and only these data are then displayed in the report. It should make the report clearer.

Extensions

Setting a precise quantity and limits using SIDI

In some cases, the user knows ahead the quantity of the product to be sold in the promotion (or the maximum quantity that can be sold).

An example: there is just a limited quantity of a product in a warehouse intended for the promotion.

In these cases, the user can manually set the maximum (or the exact) quantity that can be sold in the promotion. The module Forecasting promotional sales respects the user’s input and thus the forecast is equal to the quantity (of the product) set by the user (or the quantity is not greater than the maximum quantity specified by the user).

The manually set quantity (of the product to be sold) is respected even when a promotion is eliminated (i.e., marked by a flag), the flag indicates that the promotion should not be computed automatically.

Copying effects of promotions from the past

Very often, promotions are plan just for a short period in future, for example, for 2-3 months. If the proportion of sales during promotions is high, the sales forecast for the whole year ahead is incomplete and – after the 3 months – it is significantly lower than in the past. It is because the promotions were not entered into the system.

The present extension can copy INCREMENTS of promotions from the past. That is, it does not copy the promotions themselves, but only an increase they caused, i.e., their effects. It is possible to set up the number of months not be copied – it is 3 in this case. These months will be computed based only on specified promotions. The rest of the months will be increased according to the historical sales. If there is to be entered a promotion to a month, which has been copied from the past, (e.g. to a 5th months from now), the effects of the promotion and copying from the past are not accumulated: only a maximum is taken. This should prevent doubling of the same data, that is, counting effects of a promotion and having copied its results from the past at the same time.