Optimising customer interactions

Demonstrated the value of various Machine Learning driven approaches to dynamic pricing for customer acquisition and retention.

Interim project lead for a real-time multi-channel Next Best Action system build on AWS to optimize customer interactions. The goal of the project was to expand the successful implementation in place for the consumer market to the small business market.

Worked on various small projects to help handle the post-corona peak in projects for this boutique consultancy company.- Created a ML model to understand customer happiness measured in NPS, dis-aggregating the effect of operational excellence, customer service channels and demographic effects.- Helped to measure and improve marketing effectiveness for an upsell focused marketing team.

As part of the Data Science team I worked on:- Improving marketing for small business b2b acquisition and cross-sell campaigns by optimizing customer selection using machine learning.- Increasing open-rates by 10% for the b2c cross-sell campaign using reinforcement learning (multi-armed bandit) to a large-scale test with 43 subject lines and pre-headers.

As an external advisor support the forecasting project group with direction, model evaluation methods and implementation advice both technical and organizational.

Worked in a team of 14 Data Scientists en Actuaries to set the risk- and commercial pricing fordamage insurance policies. The main task was to create a reproducible and re-usable process forpricing using Python+H2O on Azure cloud. We did this while updating the NN and ING car insurance tariffs.

Worked in a team of 8 Analysts focussed on KPN consumer market. The most important skill used was to start projects that both help with company objectives and are innovative.– Responsible for the evaluation of marketing campaigns around the renewal process for KPN mobile. Enabled a better estimation of campaign impact in euros by using Survival Modelling techniques.– Improved customer selection for the telephonic renewal campaign, reducing wake-up churn as the goal, by using Uplift Modelling techniques in a daily scheduled automated process written in R.– Increased the open-rate and conversion of a cross-sell email campaign by optimizing the used subject line with a Multi-Armed Bandit algorithm (Thompson Sampling) written in Python.– Created a prediction of churn, renewals and base numbers 24 months into the future for a financial forecast by using Markov Chains.

Worked in a team of 6 Data Scientists. The most important skill used was to create daily-updated production worthy models, with a shared codebase for the team.– Used Machine Learning techniques (Generalized Linear Models, XGBoost, Heterogenous Ensemble models) to predict daily sales, aiding in more efficient personnel planning for logistics and customer service.– Wrote recommender system for substitute products (Alternative producten). Implemented using Google BigQuery and a Apache Spark cluster on Google Cloud. Winning idea at Coolblue Hackathon 2015.– Wrote re-usable tooling in Python for model selection, parameter tuning, cross-validation, visualization of results and deployment of the best model.

Shareyourmeal.net (Dutch: Thuisafgehaald.nl) is a location-based social community aimed at improving social interaction in your neighbourhood through the sharing of food with your neighbours.I did small scale data analysis to find the factors that make the difference between successful and less successful homecooks.

A small team (10 students), main goals:* Frontend (HTML/CSS/Javascript) and backend (PHP/MySQL) developer for A.S.V. Dizkartes' website www.dizkartes.nl (2000 visits/day, 20.000 hits/day).* Network infrastructure manager (Windows, 20 Client computers, 2 Servers).Chairman from September 2010 to September 2012

- Wrote tools in Perl to ease processing of the PDTB Corpus for the course Computational Discourse- Coordinated lab sessions and graded assignments for the course Signal Processing

Promoting Apple products and solutions.