Team members:
Ting Xiao, Automation Developer
Rafael Maia C, Automation Developer
Michel Benites Nascimento, Analytics Solution Designer
Yao Lu, Supply Chain Manager
Country:
United States
Organization:
Ericsson
Description:
Ericsson provides high-performing solutions to enable its customers to capture the full value of connectivity. The Company supplies communication infrastructure, services and software to the telecom industry and other sectors.
Awards Categories:
- AI Democratization & Inclusivity
- Value at Scale
Challenge:
As the world leader in the rapidly changing environment of communications technology, Ericsson operates many warehouses on a global scale. Optimizing the use of this space is a key part of Ericsson’s lean supply chain.
Our project goal is to provide accurate estimates of the space for current and future inventory need. For most of the products stored, the occupied space can be calculated using simple formulas. However, for the remaining products this is not possible, thus historically the estimates of the available space were inaccurate. Simply applying the formulas will not satisfy the stakeholder requirements for high accuracy.
Our challenges can be summarized as such:
- Impossible to measure every single packaging on a daily basis
- Logic to calculate the size of unknown packaging is unclear, and so complex that it cannot be defined by the business
- No centralized platform to perform the calculation
- Stakeholders require high reporting accuracy
Solution:
After being introduced to the Dataiku platform at Ericsson, we realized that Machine Learning could be used to estimate the space occupied by those remaining products in order to increase the accuracy - and all this, just in one platform. Since we already had the data stored in our data lake, Dataiku made it easy to extract, clean and transform the data.
The seamless way it integrates the data flow simplified the process of dealing with the data sources. The data selection using Dataiku is made easy through the dataset explorer window, picking the right variable type, identifying incompatible data types is very easy and fast.
We could also test multiple different Machine Learning algorithms for benchmarking, without having to code them. We were able to compare KNN, Random Forest, XGBoost and a few others. The hyperparameter setup, metric selection, and comparison with the charts output made it easy to spot the best algorithm. Not only that, but the auto retrain and selection of the best technique also allow our models to pick a better approach should it change with future data.
Model lifecycle management is achieved by scheduled model retrain, which will pick up the changes in warehouse behavior and to provide more accurate estimation. The entire flow of pulling data, scoring, and outputting is automated via the Scenario feature.
The output of our project is a dataset containing accurate estimates for the space available in all warehouses. This dataset is read by a dashboard in Tableau, displaying the insights visually to all supply hub managers.
Impact:
The ultimate financial impact of our project is that Ericsson saves on real-estate costs, by continuously optimizing the use of the existing warehouses. This is thanks to the more accurate estimates from our project.
On top of the financial savings, our project provides operational visibility on demand, a data-driven warehouse space management process, and aligns with our strategy for digital transformation.
The work that we initiated in our Americas region is now being showcased within our Group Supply organization. Being reusable around the world, our work will have an even greater impact on Ericsson's transformation journey.
Personally, as Citizen Data Scientists, we are able to use AI to augment and optimize an existing process, all without writing a single line of code. This inspires us to do the same for many other use cases at Ericsson.
