21.02.2019, Lesezeit: ~3min
In this 4-part blog series, the BI team at Cards & Systems present an example project, and explain how all of the parts of our team work together to provide our solution. Fern Watson, writing for the Data Science team, gives the technical solution. Sebastian Pisarski, representing our Data Visualisation experts, writes about the way that dashboards support the project. Finally, Claudia Grünberger gives the marketing perspective, explaining how to turn consumer behaviour insights into actions.
A grocery store wishes to sort customers entering their website into one of two categories. They are very concerned about assigning customers to these groups correctly. Success could lead to a boost in sales, but incorrect assignment might cause people to leave the site without making a purchase. The BI team at Cards & Systems explain this use case in greater detail.
A grocery store plans to run simultaneous promotions on meat and vegan products in their online store. They have one advertising banner for each of these promotions, and a ‘neutral’ one, just showing a generic selection of items.
They prefer not to show the generic banner because it won’t drive many additional sales, but they really don’t want to show the wrong offer to the wrong customer! In this case it could cause offense, and might result in some customers choosing not to make a purchase at all.
The company develops a strategy to show the neutral banner at first, switching it to either the meat or the vegan promotion when they are confident that the customer falls into one of these categories. If there is too much uncertainty, the neutral banner can remain on the screen, minimising the risk.
In this online shop, the customer does not have to have an account. This means that it is impossible to know what they have bought in the past, and that information isn’t available to inform any strategy. In this case, the algorithm needs to categorise customers ‘on the fly’, as they interact with the website.
However, historical (anonymous) data about user sessions and their associated shopping baskets is available. While it’s clear that buying items such as steak or tofu will indicate different preferences, machine learning algorithms can find more subtle patterns in this data.
To solve this problem, the data scientists at Cards & Systems can begin by separating the historical purchase data into session information, including users’ behaviour during the first few actions on the site (actions are things like searches, clicks, and additions of items to their basket), and the final basket. The final basket is then categorised as ‘meat’ or ‘non-meat’ so that it can be used as the label in the machine learning phase of the project.
In the next blog, we will reveal the technical solution used to solve this problem, why it was selected, as well as some alternatives. Also stay tuned for part 3, which will cover the visualisation of this data, and part 4 which looks at the project from a marketing point of view.
Dr. Fern Watson