Module: Cannibalization

Executive summary

Cannibalization is defined as a decrease in sales of an item as an effect of another (similar) item’s increase in sales (typically due to a promotion).

The module is crucial in evaluating a real profitability of promotions. If we profit greatly by selling an item in promotion more, but – as a consequence – other products (with a higher margin) ceased to be sold (since they are not in promotion), the promotion as a whole might turn out not to be profitable at all. The not-in-promotion items that – hence – are not sold, are ‘cannibalized’ by the in-promotion items.

The module is also important for supply management, particularly for products with limited expiration date. Hence, not-in-promotion products must be ordered less, because of their (soon coming) expiration date.

Functional description

Data needed for computation

Not all products interact with one another. For instance, having a promotion on milk does not – in any way – affect sales of washing powder; on the other hand, a promotion on strawberry yogurt significantly affects sales of apricot yogurt. Product groups, whose regularity of sales interacts, are called cannibalization groups and they are defined during an implementation of a promotion; notice that they are depending on a type of a customer.

A group of products is defined as a cannibalization group, if all products (of the group) are found at a single store (sales of products at one store cannot affect sales of any other product at another store).

Given a rather large number of products in a leaflet (which, moreover, includes virtually all categories of products), it is rather impossible to forecast automatically, which products interact with which.

Therefore, for computing cannibalization, cannibalization groups (i.e., groups of products whose sale is influenced by another group of products) must be entered manually.

If you happen to have a well-designed product hierarchy (see the module Necessary inputs for STOCK) and if at the lowest level of the tree, the products are so similar that they directly cannibalize on one another, we can use the hierarchy – hence, you don’t need to supply any additional inputs.

Computing cannibalization for past promotions

In the following text, a cannibalizing item is an item in promotion such, that its increased sales affect negatively sales of a cannibalized item.

To determine cannibalization on two items, the following criteria must be met:

  • Both items belong to the same cannibalization group.
  • The length of promotion on the cannibalizing item must be between 5 to 31 days.
  • The cannibalized item can be in promotion for (maximally) 25% of the cannibalizing period – in other words, for at least 75% of the promotion days, the cannibalized item is NOT in promotion
  • Minimum degree of reliability of the forecast must be met; the parameter is adjustable and default is 70%.

The main aim is to calculate the amount by which the sales on cannibalized item were decreased. Lost sales is calculated as a difference between actual sales of cannibalized item and estimated sales amounts in the same time period, as if there is no promo activity on the item that cannibalizes others. Estimated sales amounts are calculated from sales of the cannibalized item using statistical methods from one month before and after a promo activity.

Forecasting cannibalization for future promotions

Cannibalization, i.e., a drop in sales of an item because of a promotion on a similar item (an item from the same cannibalization group) at the same warehouse, can be statistically estimated based on data from past cannibalizations.

As always, a cannibalized and cannibalizing item must belong to the same cannibalization group and they must be located at the same warehouse.

As the first step, historical cannibalizations are loaded. Data from various historical cannibalizations can be loaded, including data from other warehouses (than the actual warehouse for which cannibalization is to be computed). A groups of warehouses, from which the historical cannibalization data can be loaded, is determined during implementation and depends on a customer. It still must be the case that in a single group of warehouses, the behavior of the products must be comparable. For example, a group consists only of small warehouses, while big warehouses are members of another group, etc.

In the second step, historical cannibalizations are assigned weight based on their similarity to a historical promotion that caused cannibalization – and to the future one (that is likely to cause it).

In the final step, we compute (using weighted average) item’s cannibalization strength: both for an item in promotion and for an item outside of promotion.

Displaying cannibalization

Cannibalization can be seen in the report Evaluating promotions in the module Forecasting promotional sales.