Michał Adamczyk

A reinforcement learning approach to balance the exploration-exploitation trade-off when buying traffic in the real-time bidding environment. The module guarantees ramp-up of unknown traffic by prioritizing exploration first and then allows the traffic to be bought organically by gradually reducing the exploratory aggression. The solution was A/B tested and improved all relevant KPIs for the client. It was ultimately deployed to all production traffic.

As a member of an external data science team, I contributed to the digital transformation of an online shop from data-unaware to data-driven. I performed extensive data analyses and AI-based data modeling on the shop's customers and concluded with detailed reports and strategic recommendations. As a result of this work, online reporting in Tableau is used on a daily basis to uncover underlying trends, and the company uses AI-driven, automated identification of high-chance and high churn-risk customers.

A machine learning model that identifies customers with a high risk of churn. Customers of the CRM use the product to process their sales funnel. The model collects activity information about user sessions and predicts users who should be approached because they have a high risk of churn. The solution reduced the churn rate and improved the company's performance on KPIs.