In today’s digital world, marketers have access to millions of data points which are helpful to understand and predict consumer and customer behavior. With the help of social media data, clickstream and web journey data, as well as consumption data, marketers can better segment and target consumers, enhance cross-selling through recommendations, predict customer churn, optimize consumer communication and marketing budget allocation. Nevertheless, the many opportunities made possible by the data availability do not come for free as they require marketers to understand complex analytical methods that all rely on artificial intelligence and machine learning.

Although companies have been using AI for several years already, available methods are not yet fully exploited by marketers. A recent survey by McKinsey showed that only 14% of companies are using machine learning for customer segmentation and 17% for customer-service analytics. These numbers are in great opposition to the potential revenue increase to be gained from AI in marketing. To tackle this gap, the JPM therefore offered a research seminar in the summer term 2021 to investigate the opportunities and challenges of machine learning in marketing. Here, students got the opportunity to work together in groups on marketing problems which had to be solved by applying unsupervised and supervised machine learning algorithms.

To ensure that everyone is on the same page, students were provided with access to DataCamp courses in advance, covering basic coding in R instructions as well as several machine learning topics. Subsequently, a joint introduction to the subject matter assured that students were equipped with all necessary tools to complete their machine learning projects. Students then familiarized themselves with datasets containing information on customer behavior and product features. The discussed topics covered three unsupervised, three supervised and one semi-supervised machine learning problem from which the groups could choose.

Working in groups of three, students who were interested in unsupervised topics investigated how to segment customers as well as clustering products to make appropriate recommendations and targeting. For customer segmentation, survey data from the airline industry was used to identify different clusters of airline customers with the help of the k-means algorithm to give airlines recommendations which clusters deserve more or less attention and how to optimally address the different clusters. Another group was looking at the products itself to make product recommendations and used market basket analysis tools which are often applied on retailing websites like Amazon. Relying on a whiskey data set, students had the opportunity to develop their very own recommender system for a set of whiskeys with different bodies and characteristics. Diving deeper into the topic of recommender systems, the third group examined a large shopping data set from one of Europe’s largest online retailers with information on very different product groups. Using cluster analysis, the students were able to detect several patterns in buying behavior and group products which are commonly bought together. Watch the videos below to learn more about the projects applying unsupervised machine learning algorithms.

Since losing and re-acquiring customers is quite expensive, it is not only important for marketers to cluster their customers, but also to predict which consumers are likely to quit the company before they actually do so. With the help of support vector machines, the first group focusing on supervised machine learning algorithms therefore used previous churn data to identify customers who might leave the company. The students had access to real data from a large German telecommunication company to train their own model and prediction skills. Working on a related topic, another group investigated the chances and obstacles of predictive analytics. Using a large number of advanced machine learning algorithms such as random forest, adaptive boosting and extreme gradient boosting, they looked at credit card data to predict which customers deserve more ore less attention. Further, addressing another important topic in marketing analytics, the sixth group applied marketing mix modeling to understand how specific channels and ads are convincing customers to behave in the intended way and to eliminate the non-working ones. With access to click-stream data from an online retailer that uses social media ads, search engine advertising and banner advertising, they relied on different regression models to filter out the non-working channels. To see how the group made inferences about the optimal marketing budget allocation and to learn more about the other projects described above, see the videos below:

The last group focused on semi-supervised machine learning and scraped more than 36,000 Amazon reviews across ten product categories. These reviews provide valuable insights for marketers as they can help them to gain a better understanding of potential product or service improvements. With the help of topic models, the group members then made sense of the obtained review data and derived specific recommendations for companies. A summary of the collection process and analysis is described in the video below:

At the end of the very successful seminar, the groups got the opportunity to present their results and potential obstacles to the other students and engaged in lively discussions about the insightful managerial implications. All in all, the seminar provided students interested in machine learning and marketing analytics a suitable and detailed introduction to the large and quickly evolving field of AI and provided them with useful tools for dealing with huge data sets while working on real-life marketing problems.