Data integration
Data management for machine learning
Complexity theory
Description
At Picnic, we currently have an assortment of some 8k articles. However, in practice, many play a similar role from the customer’s perspective - a customer wants to buy e.g. cola or bananas, but there will be multiple colas or multiple bananas that the vast majority of customers will only buy one variety of. Likewise, if a customer wants to cook a certain meal, say pasta bolognese, there will be many different collections of articles that could represent this; and for the role of ‘minced meat’, there could be a number of varieties of minced meat in different sizes and of different origin, including vegetarian and vegan options. These are similar concepts, they are a sort of ‘product archetype’ that play a certain role in a customer’s shopping.
However, for different types of product archetypes, and also for different products within certain archetypes, the behaviour might be slightly different. Some articles might be directly interchangeable (e.g. two brands of bananas, or two sizes of exactly the same product), but others not or only in one direction (minced meat vs the vegan option). In some categories, a customer will only buy one product from the archetype in a specific shopping session, while in others (e.g. zoutjes) customers regularly buy multiple. And finally, what should be shown might also depend on the context (where in the app, e.g. in search or in recipe) and on the customer.
This project aims to formalise these inter-product relations. Can we systematically identify these product archetypes? Can we quantify how interchangeable two products are, in different contexts? What different contexts are relevant? And is it generally possible to generate context-specific sets of products by starting from a broad archetype and applying filters or re-ranking, or is this too simplistic?
Work Environment
Picnic Technologies is known for its innovative approach to grocery shopping and delivery. The work environment reflects this innovative and dynamic culture. A lot of creativity, open communication, and a willingness to come up with new ideas!
Expectations
We’re aiming for quality thesis work where the intern is able to deepdive into the topic at hand within the context of potentially applying it to Picnic’s business on the long term, with guidance from Picnic of course. We require at least 5 months, full-time internships; there’s an online code test and two interviews in the process.
Modelling complex systems & networks
Data integration
Algorithms and datastructures
Description
Customer preference changes over time. Customers might switch their shopping behaviour for various reasons, e.g.:
-they find a similar but cheaper/better-tasting/preferable alternative
-they start or stop buying certain types of products online (e.g. some customers are wary of fresh products from online)
-their tastes change
-their diet changes
-their family/living situation changes
At Picnic, we have an extensive and very well-maintained Data Warehouse, where relational databases store e.g. personal purchasing behaviour, historical status changes of articles and customer addresses, and so on. There is also considerable data on in-app events that can give insights in how customers are using the app, how they navigate and what actions are performed. Such factual information is quite reliable; other data points such as family size or number of pets are self-reported and require interpretation. Knowledge of SQL is required to unlock this primary source of relevant data; in addition, Python is the language of choice in the Data Science team.
There are a number of interesting research questions in this direction, which can be further refined during the thesis. the A first question is how shopping behaviour changes over time, and if specific stages can be identified during this change in behaviour. This requires systematic detection of patterns in the available historical purchasing data, potentially also including relevant in-app events. Depending on the type of shopping behaviour we are interested in, different kinds of dynamics may occur; i.e., it might be a simple case of transitioning to a different product, but also wider changes in behaviour such as ‘ordering more sustainably’, ‘becoming a more mature customer’ (whatever that means exactly is an open question), or ‘switching to a vegetarian diet’ are of interest.
A similar, but more forward-looking question is whether we can predict behaviour change trajectories for current customers. From the academical point of view, this allows for a validation of the more historically focused analysis in step 1. Likewise, If we can observe customers that are likely to exhibit certain future purchasing behaviour, this can be used to show e.g. more relevant products, choose promotions that might induce them to explore new categories, or proactively reach out to avoid (partial) customer churn. That brings us to a next question: is it possible to nudge customer behaviour via e.g. recommendations or presentation to induce healthier or more sustainable choices? And in what ways can this be achieved, i.e. what are the articles or methods that are best suited to do so? Can different customer groups be identified here that may need to be treated in distinct ways? This is a research question that not only brings recommendation systems directly into play, but is also a next step from the business perspective. Eventually, alignment will be necessary with business-facing and tech teams for implementation of such nudging experiments, but it is certainly possible to validate the research and see if the nudging predictions hold water in real life.
Work Environment
Picnic Technologies is known for its innovative approach to grocery shopping and delivery. The work environment reflects this innovative and dynamic culture. A lot of creativity, open communication, and a willingness to come up with new ideas!
人挺多的估计，等等吧