Demand forecasting and planning at Danone based on Jume platform

Artificial Intelligence capabilities have transformed the demand planning process at Danone

Industry:

Dairy products and health food production

Task:

Demand forecasting and planning

Client:

Danone

Background

Danone is the largest dairy producer across the world, developing brands in dozens of product categories, including milk, sour cream, yogurt, and baby food. Danone’s business is complex.

Given the complexity of the business, the company has three types of demand forecasts: short-term, medium-term, and long-term. The quality of each of these forecasts determines the successful functioning of certain related processes. For example, the accuracy of the medium-term forecast used for promotional planning directly affects the ROI of promotions, the level of service, the number of write-offs, client relations, storage and transportation costs, and much more: the price of increasing accuracy by every percentage point can lead to a significant increase in efficiency. And given that the vast majority of the company’s business is perishable, it always goes the extra mile to build and improve forecast quality.

Following the parent company’s global strategy of harnessing artificial intelligence (AI) technologies to improve the efficiency of their business processes, the project team, which included Danone employees from supply chain management, marketing-review, IT, and the Data team, decided to continue the digitalization of planning with the Jume team, which had already implemented a platform for sales forecasting during promotions at key national retail clients.

Solution

The data science team designed and implemented a demand forecasting platform:

Three planning horizons: an operational sub-day forecast two weeks ahead, a mid-term forecast for the next quarter for promotional planning purposes, and a long-term 12-month monthly forecast for annual financial planning.

A wide range of input data sets were used to train the ML algorithms and to build forecasts: historical shipments, shelf sales, promotional plans, price levels, client orders, inventory levels, calendar events, assortment changes.

Especially for Danone, the basic configuration of the platform based on XGBoost methods was enhanced with additional Machine Learning algorithms such as Ridge, Lasso, and KNN.

A separate model for demand forecasting for key Retail client (store formats: minimarkets and hypermarkets) was developed and integrated into a single loop.

Automated integration between incoming data and the platform was set up.

Business value

+ 73,5 %

ML forecast accuracy increased to 74.3% compared to the forecast created by the promotional planning team using the existing IT solution (70.6%).

- $750k

Reduced product write-offs by USD $750k per year due to improved forecast accuracy.

Danone

Artificial Intelligence capabilities have transformed the demand planning process at Danone

Background

Solution

Business value

Other customer stories

Unilever

Greenfields

Get in touch to learn more

Thank you!

Error!