Project - Predictive Complaints Analysis
Using Natural Language Processing and machine learning to understand and predict the causes of customer complaints to prevent them from occurring.
How do we shift from reacting to complaints to predicting and preventing them?
Summary
Complaints result from the building frustration as customers encounter friction navigating processes, systems and bureaucracy. Machine learning can make it possible to see everything that has happened leading up to a complaint, recognising stress points and predicting when and how complaints are likely to arise.
Our value goal was to predict the circumstances and when a complaint is likely, enable preventative fixes and achieve a 10% reduction in complaints.
Impact
We analysed over 43 billion data points, identifying areas requiring priority attention to avoid potential complaints. The work replaced a manual process that took 4 days to classify just 20% of complaints, with a new automated process that takes just 40 minutes to classify 100% of complaints.
Our analysis directed remediation work to areas of the web application which were common triggers for complaints, building support prompts at typical friction points in order to improve the customer experience and prevent potential future complaints.
Our approach
Automate complaint categorisation
We changed the way that complaints were analysed, using Natural Language Processing to identify root causes, replacing the existing manual process.
Link complaints to customers and events
We built an automated solution that identified the individual customer that each complaint related to and the timeline of interactions leading up to the complaint.
Machine learning path analysis
We used machine learning to identify the features of journeys which were highly predictive of a complaint, from online API calls to specific call handler interactions.