Stock optimization for a new product launch
Launching a new product, how to predict the need for stock?
01
Problematic
A distributor with warehouses in several countries offers many new products every year. Each launch is prepared by the marketing teams, who plan the launch request, and by the logisticians who configure the stock requirements in each warehouse.
Despite this, nearly one product launch out of two ends up either out of stock or in over-stock, before gradually restoring the situation in the stabilized phase.
02
“Product launches are better anticipated and Supply Chain tensions are greatly reduced.”
Process
The approach consisted in regrouping products into clusters with similar launch behaviors, based on 10 – 20 main parameters. These clusters have also demonstrated non-trivial similarities between products from distant ranges.
We have defined a small number of specific KPIs that we are trying to optimize (including the number of products in stock, and the cost of stock).
The machine learning model then used a 3-year history of product launch to build predictions on launch behavior. These predictions were tested on a part of the history that had not been used in the learning.
The best model was thus selected and used to support decision making for suppliers.
3
years of history
3000
products analyzed
10
warehouses
Result
03
Result
The model is able to predict with a good probability the launch trend after only about ten days. Previously, this analysis could only really be done after more than a month.
The time saved makes it possible to manage replenishments or reallocations of certain warehouses in the event of a risk of stock shortage, and to launch commercial operations in the event of a launch that falls short of marketing expectations.
In addition, the model makes it possible to make recommendations for optimal stock allocation and to alert in the event of a high probability of stock shortage in one or more warehouses.