T-Mobile Polska is a well known leader in telecommunication services. As many companies that depend on retail, they need to optimize client acquisition through direct sale at physical stores (POS).
To optimize location selection process it's usually best to combine field expert knowledge with data analysis and prediction tools.
Artificial Intelligence & Machine Learning
Imagine you have to open dozens of stores in a short period of time, in all major cities in the region. Let’s say you have 2-3 possible locations for each store. That’s over 1000 potential locations, usually with little to no data history.
How to find locations with highest sales potential?
To answer this question TEONITE cooperated with T-Mobile business and big data teams.
TEONITE applied Data Science approach to collect critical information and lead the analysis process through:
- data engineering and feature engineering,
- building ML pipelines for evaluation of multiple predictive models,
- analyzing core GSM network data (animized behavior of users),
- analyzing correlation with sales data,
- visualization and information discovery.
The goal was to select the model with the highest success rate in predicting chosen sales KPIs. Our evaluation included various ML models and techniques: eXtreme gradient boosting, LightGBM, random forests and neural networks.
As a result, we've delivered predictive models and in-depth report to to help the business optimize the process of location selection.
- Python Data Science stack: Pandas, NumPy, scikit-learn
- JupLterLab and Jupyter Notebooks analytical environment
- Orange toolbox for data mining and data visualization
- Apache Spark data processing
- Custom Machine Learning pipeline
- PostgreSQL database
- Selected models: Extreme Gradient Boosting (XGB), Neural Networks, ensemble methods
Content-based platform for a global, NGOs support environmentRead case study
Planning a new project?
We know how to make it work!Get a free project estimate