Loading...

Customer Journey Analytics using Process Mining Based on the Markov Model

Torabi Ardekani, Saba | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57611 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Hassan Nayebi, Erfan
  7. Abstract:
  8. The analysis of customer journeys has gained significant attention due to the critical role of customer behavior data in enhancing business decision-making and formulating strategies for customer acquisition and retention. By segmenting customers based on their journey patterns, businesses can offer personalized recommendations, thereby improving customer engagement and loyalty. Additionally, predicting the next steps in a customer’s journey based on historical data allows for timely and appropriate interventions at various touchpoints. By understanding where customers are in their journey, businesses can provide targeted recommendations that increase the likelihood of converting potential customers into actual ones. To this end, the application of clustering algorithms for grouping customer journeys, coupled with predictive models to forecast subsequent steps, has been identified as a viable approach. Given their simplicity and interpretability, Markov models offer a robust method for such predictions. Customer journey data typically includes three key attributes: customer ID, touchpoint, and the time of occurrence of each touchpoint, which aligns with the structure of event logs in process mining. In this study, customer journeys are first mapped to event logs. Based on the nature of the event logs, an appropriate encoding method is selected, and customer journeys are encoded accordingly. Following this, clustering techniques are applied to segment the customer journeys into distinct groups, with the quality of clustering assessed using the silhouette score metric. After clustering the journeys, Markov and Hidden Markov Models are developed for each cluster to predict the next steps in the customer journey. The performance of these models is evaluated using accuracy, precision, recall, and F1 score metrics. The proposed model is applied to data from an online cosmetics and personal care store, where the touchpoints include “View,” “Add to Cart,” “Remove from Cart,” and “Purchase”. After data cleaning, journeys are encoded using one-hot encoding due to the limited number of customer touchpoints. The K-means clustering algorithm is then used to segment customer journeys. During this study, we identify five distinct clusters, which we label based on their journey characteristics: passive customers, active customers with a complete purchase cycle, hesitant customers, comparers, and confused customers. Markov and Hidden Markov Models are then constructed for the training data sets of each cluster, and their performance is evaluated on test data sets. In all clusters, the Markov model outperforms both the baseline model and the Hidden Markov model, achieving an average accuracy of 70%. The results demonstrate that the Markov model provides superior accuracy in predicting customer journey steps, offering valuable insights for improving customer engagement strategies
  9. Keywords:
  10. Customer Journey ; Clustering ; Markov Model ; Hidden Markov Model ; Process Mining ; Customer Segmentation

 Digital Object List

 Bookmark

No TOC