Background
Customer journey thinking is getting more and more established in companies. Customers are getting more volatile with higher expectations and competition is fierce. With customer journey analytics it is possible to use a personal approach on a large dataset of customer behaviour and customer experience. In the field of customer journey analytics multiple well-known fields are combined, like process mining, anomaly detection, text analytics (NLP), stream mining and predictive analytics.
Assignment goal
For companies, it is not enough anymore to look at their customer data in hindsight. The business is moving to real time analytics to allow intervention into the customer journey when necessary. In particular, the detection of undesired behaviors close to the time of their occurrences is a key in all customer journey related businesses. Traditionally, such anomaly detection task required either a “golden standard” business model for performing conformance checking or a large amount of labeled journeys to perform a supervised anomaly detection. However, with the large possible variabilities of customer behaviors, it is not realistic to assume the availability of enough labeled data or reliable normative process models. This limits the applicability of anomalous customer journey detection in real life. Therefore, companies need to invest in techniques focusing on online, unsupervised anomaly detection methods. Recently, several approaches with limited access to a large scale of labeled data started benefiting from pretrained Larg Language Models (LLMs). After fine tuning, such approaches profit from the (i) massive amounts of (pre)training data, (ii) large-scale compute power and (iii) highly expressive, billion-scale parameterized transformer models.
In this assignment, the task is to investigate the applicability of pretrained LLMs for:
- performing an online unsupervised anomaly detection over partial or full customer journeys, and,
- interpreting detected anomalies by applying a semantic anomalies analysis.
In the above-mentioned analysis, all available data will be leveraged. This includes logged data of the journey (workflow process data, call center data, online click trails, social media data), online and offline feedback and non-transactional data (product and background information of the customers).
De Volksbank
Being one of the large banks of the Netherlands, it delivers products and services to more the three million customers under four brands: SNS, BLG Wonen, RegioBank and ASN Bank with a shared back-office. Each of the brands has its own focus, but as a whole it wants to stand out by making a social impact on lives across the Netherlands by making banking about more than just money. Customer journey analytics fits this mission by really trying to understand the customer. More information about de Volksbank can be found at: www.devolksbank.nl/
Contact
For more information, contact dr. ing. Marwan Hassani.