Get 160-hour demand forecasts within hours

What was done

In an article published by the MagIT on 31 October 2025, Maxime Lelièvre, Senior Demand Planning Manager at the European Centre of Excellence in Danone, and Hugo Niel describe the transition of broken tools and Excel to SAP Integrated Business Planning, deployed in waves between 2019 and 2023, now in 38 countries, mainly in Europe. By the end of May 2023, 35 countries were already equipped. About 44% of turnover is forecast in IBP. More than 230 sales forecasters use the tool. The volume of data points observed per country has increased from 20,000 to 66 million, adding customer and channel dimensions to products and locations."From 160 to four hours to get the forecasts": it took 160 hours, it requires "20 minutes to four hours in first instance", including pretreatment; the four hours correspond to the execution time of the Python models for basic results. The forecast is for 75,000 product/segment combinations of customers, with millions of predictions per week, warnings to focus forecasters on important cases, and reliability that would "increase by 1 to 10%" by category. The article also states that this "would reduce waste by 35%": the latter two figures are conditional. This case is distinct from an already identified use, which concerns price control before invoicing.

Before you start

Tool cited by the source: SAP IBP and machine learning models (Danone); other tools cited: SAP IBP and machine learning models (Danone). Fill in the bracketed items: [pays, catégorie, horizon], [coller]. Strip confidential data that is not needed and check your tool’s access terms. Features and plans can change.

The method, in three steps.

1. 01

   Gather the context

   Spot the bracketed items in the prompt ([pays, catégorie, horizon], [coller]) and prepare what they ask for.

2. 02

   Adapt the prompt

   Copy the prompt below, replace the bracketed items, then paste it into the AI assistant you use. The cited tool (SAP IBP and machine learning models (Danone)) is not necessarily available to the public..

3. 03

   Reread before using

   Reread the result, correct it if needed and keep the final decision. Caveats are detailed below.

Your starting point.

Replace the bracketed items, then adapt the proposal to your situation. This prompt is written by the iadujour desk It is not a quotation from the source.

To be expected: [pays, catégorie, horizon]. Historical series, promotions, breaks: [coller].

1. Proposes a forecast, indicating the method and the variables actually used.
2. Lists product/client combinations that are too unstable to be reliable.
3. Prepare the warnings the forecaster should look at first.
Do not invent any past sales and does not turn a hypothesis into a definitive figure.

What deserves your attention

Useful skills

Official skills on the same task, paired by the desk: they are not part of the described case.

• Driver Azure Machine Learning (Microsoft · The SDK Azure Machine Learning v2 for Python controls workspaces, jobs, models, datasets, calculation and pipelines, mainly via MLClient.