Predicting Next Best Action(s) To Improve Sales Metrics For Pipedrive Customers
Name
Amey Chandrakant Darekar
Abstract
Predictive process monitoring (PPM) techniques exploit the full potential of historical event log data by applying data mining and machine learning methods to predict future process behavior, such as predicting or recommending the next best activity (or action). Modern techniques for recommending the next best action, particularly those using Deep Neural Networks (DNNs), have achieved near-perfect accuracy in predicting future process behavior in business environments. Despite this, since these techniques do not take into account Key Performance Indicators (KPIs), the metrics used by businesses to measure process performance making these techniques are limited in their ability to improve business processes in real-world applications. Process simulation has been used in the past to incorporate KPIs to optimize the process flow of business transaction activities, but this technique is limiting when there is a lack of definitive outcomes for action. In such cases, attempts to use process simulation alongside decision support for controlling action flows often yield unfavorable outcomes. We propose an approach inspired by business process optimization that relies on the probabilistic distribution of action sequences to predict the next best action(s). We attempt to implement this technique by taking into account KPIs that optimize the success rate of the sales transactions, using real-world event logs extracted from Pipedrive CRM. We also conducted experiments with heuristic search strategies to measure their usefulness when paired with our proposed strategy. We compare the performance of our proposed framework with the traditional control-flow simulation-based technique.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Mahmoud Kamel Akila Soliman Shoush
Defence year
2023