Module: Computing daily sales forecast in promotions

Executive summary

The module splits up the total forecasted quantity (of the products to be sold during the promotion) into individual days.

This split is particularly important for supply chain management; as a matter of facts, supply management would not be possible without it.

As in the module Forecasting monthly sales forecasts (and its complement, the module Computing daily sales forecast), we first forecast the whole promotion and only then, using various formulas, we split the forecast into the daily sales.

Functional description

Splitting sales into days for promotions

When forecasting future promotions, it is important to split up the forecasted quantity (of the product to be sold) into individual days.

There are four different ways for the split (and all of them can be further set up, as needed):

  • FIRST DAY
    • A certain quantity is proposed for the first day of promotion.
    • Advantage: the quantity is already at the given warehouse at the beginning of the promotion; hence, a daily variability during promotion is not an issue.
    • Disadvantage: requirements on a particular warehouse & unsuitable for long-term promotions.
  • UNIFORM
    • The desired quantity is split up evenly throughout the period of the promotion.
    • Advantage: simplicity.
    • Disadvantage: inaccuracy of computation, and thus a greater risk of failure.
  • USER
    • The split is governed by data imported from the primary ERP system. The desired quantity is entered in absolute values.
    • This setup ignores quantity adjustments from the module Fast adaptation of sales forecast in promotions.
  • AUTO
    • Based on data from historical promotions and use of statistical methods, the split is computed automatically.
    • The recommended settings.

Algorithm for AUTO setting

When computing the split into days for a future promotion, a past sale of an item in past promotions is taken into consideration. The computation has the following steps:

  1. Preparation of historical statistical data
    1. For each future promotion, to find a sufficient number of similar historical promotions is a must.
    2. Each historical promotion is assigned a weight adjusted in accordance with the similarity of the historical and actual (i.e., future) promotion.
    3. Promotions with a greater weight are preferred.
    4. Each historical promotion is divided into 2-6 sections.
    5. For each item in the planned promotion, a percentage of sales of the given item in a given section of the past promotion is evaluated.
    6. The data serve as a “model” for future.

Extremes are cleaned from the set of historical promotions.

Splitting sales into days for an ongoing promotion

If an ongoing promotion develops in an unexpected way (i.e., in a way different than forecasted), there are two remedy functionalities:

  • Algorithm for correcting forecasts of promotional sales described in the module: Fast adaptation of sales forecast in promotions. This forecast changes the expected sales of the promotion – and it splits it into days subsequently.
  • A setting ‘Keep’. There are two values:
    • No: the quantity sold in the promotion so far is ignored and a split up remains unchanged (regardless of the quantity already sold).
    • Yes: from the sales history, the quantity sold so far in the promotion is loaded into the system. The quantity is deducted from the total forecasted quantity (to be sold), hence, only the remaining forecasted quantity is split up into the remaining days. If the sale is greater or equal to the forecasted sales, there is nothing left to be split into the remaining days (thus nothing is split).

The split functions behave differently with respect to these two remedy functionalities (Fast adaptation and Keep for short):

  • FIRST DAY
    • Fast adaptation – ignored
    • Keep – accepted
  • UNIFORM
    • Fast adaptation – accepted
    • Keep – accepted
  • USER
    • Keep – ignored
  • AUTO
    • Fast adaptation – accepted
    • Keep – ignored

Displaying daily sales in promotions

Given that the split into days is used mainly for inventory management, the split is – indeed – visible in the expected modules: Orders and its part: Explaining orders.

Extensions

Splitting sales into days for retail suppliers

Retailers and retail chains purchase products differently from end customers. Retail chains purchase a large portion of products they want to promote in their own promotions ahead of time, during stock delivery period. Also: retail chains typically do not purchase products every day; rather, they purchase larger quantity of products at once.

To determine the effect of various purchase behavior of chains and retailers, different strategies are designed to explore purchase strategy of retail chains between two strategic days: the first day of stock delivery and the first day of the (chain’s own) promotion. During this period, the retailers and chains ‘stock up’ before promotions.

From a supplier’s point of view, it is crucial to identify (and pinpoint) the delivery period of a retailer, forecast the quantity the retailers plan to purchase and the exact day of the delivery period.

Fast adaptation of sales forecast in promotions

The extension adjusts the forecasted quantity of products intended for a promotion after an unexpected change in forecasted sales. After a certain (adjustable) time from the onset of a promotion, the algorithm starts to evaluate the promotion’s progress and detects whether the real sales match the forecasted sales (as computed in the module Forecasting promotional sales). If a relative accuracy of sales forecast fails to be in a user-specified interval (i.e., should the forecasted quantity be higher or lower), the present extension adjusts the remaining sales for the promotion.