Writing your thesis at Veneficus

Almost done with your study and looking for a place to write your thesis? Consider Veneficus! When working on our projects, you get to see the practical applications of data science right away. At Veneficus, we have several subjects we want to research further. For most of them we have data available which means you start right away.

Writing your thesis at Veneficus does not only give you the opportunity to take on a challenging project, you can also get a close look of how we work as a company. While working on your thesis, you will support our "regular" projects as well, if you want. This gives you a sense of what it would be like to work for Veneficus.

Veneficus specializes in data analytics and in factual decision making on the basis thereof. We support companies by using discerning analyses to provide properly substantiated, innovative, and understandable insights. To do so, we make use of the web tools we develop ourselves, among other things. Our focus is on the following industries:

  • Retail
  • Real estate
  • E-commerce
  • Insurance
  • Public

Veneficus is young, innovative, and enterprising. Our thinking is solution-oriented, and we have a fresh take on our field. Our clients love our sense of passion, but most of all, the fact that we manage to actually solve their problems. In doing so, we never shy away from a challenge. In our view, the phrase ‘I can’t’ does not exist!

Thesis Topics

The thesis topics we are interested in range across the ones mentioned below. If you have an interesting idea for a project you want to take on, let us know. If not, below are some topics we would like to see researched.

Media and Marketing Effectiveness

Companies have a lot of ways to spend their marketing/media budget. They can for example send email campaigns, air tv commercials or buy Google Adwords. Most of our clients don't know the optimal way to divide their budget over these media channels.

The biggest challenge is related to indirect effects. Someone might see an online add, check the website for information and decide to buy it a week later in a real store (ROPO, Research Online, Purchase Offline). Google Store Visits can help to get this data, but more advanced techniques are necessary to properly model the complete media effectiveness.

You could think of using a Bayesian, multivariate adaptive regression splines or generalized additive model. Long term effects could for example be modelled with Adstock, but when you write your thesis at Veneficus you'll get the freedom to look for the best techniques for this problem.

Mixed model forecasting

For a client in the delivery food business we forecast the number of orders for the next week. In order to do this we apply model selection/averaging, using several different modeling techniques and selecting the best ones based on out of sample predictions.

The aim of this project is to improve this process, by improving either the different models or the selection/averaging process. Possibilities include:

  • Analyse which models do well on certain days, possibly with a model (model x does better on holidays, y when it is raining, etc)
  • Use model to combine individual model results (instead of averaging) (stacking)
  • Models in series, each model using the error terms of the previous model as an input (boosting)

Customer clustering in E-commerce

To obtain insights in the customer base one can perform a clustering of customers. Customers are clustered based on behaviour (purchases, purchase time/frequency, store/website visits, ...) and if available, other data like age, gender, income, and others. This helps companies to form ideas of their customers, which can be useful for targeted marketing.

We have a lot of experience with customer clustering. However, so far this is mostly done in the offline retail case. The online environment offers new possibilities when it comes to data tracking customer behaviour. Can we obtain a more precise clustering?

Explain efficiency scores in location potential analysis

At Veneficus we perform location analysis for our retail clients. For this analysis we use data about demographics and competition to assign a potential revenue for each store location. Using this, the revenues of different stores can be compared more objectively.

On top of this potential per store, the model output contains information about which variables drive the potential (positively or negatively). However there is discussion about the method used for this. Does it yield unbiased and efficient estimators?

The aim of this project is to research and gain understanding of these estimators, in order to know whether we can trust them. If this turns out not to be the case, there should be alternatives to obtain a good description of what drives potential.

Risk assessment of insurances

Insurance companies create risk profiles for their customers. These profiles are usually generic, while in theory, it is possible to make a more precise risk profile for a specific group of customers. The goal of this project will be to obtain more specific risk profiles. First, customers can be clustered, using several variables like age, education level, income, and many more. After the clusters are obtained, a risk profile can be assigned to each cluster, and hence to each individual. These risk profiles will be more specific and more realistic, allowing the insurance company to make a better assessment of future costs.

Interested?

Are you interested in doing research at Veneficus? Do you like a challenge? If so, please send your resume and cover letter to recruitment@veneficus.nl. Would you like to know more about Veneficus and this position, please contact Robbert Bos via phone +31 10 40 47 333.


Back