- Commodity Management & Trading
- Customer & Marketing Strategy
- Digital Engineering
- Station Studio
- Supply Chain & Industry 5.0
- Sustainability
AI-Based Customer Claims Classification
AI-based classification helps energy retailers turn high-volume customer claims and requests into more consistent, actionable routing decisions. By learning from historical interactions, it supports faster service handling, reduces manual effort and creates a stronger foundation for customer-care prioritization.
The transformation required more than a model. It needed a way to organize heterogeneous information, interpret varied customer language and make analytical outputs practical for business and service teams.
Customer claims and requests arrived through telephone, web, offices and other channels, creating varied text inputs that could not be treated as a single standardized flow.
Manual categorization created inconsistent assignments and additional rework when requests or claims were sent to the wrong category or team.
The initiative required customer, interaction and consumption signals to be brought together in a form that analytical models and operational teams could use consistently
Model outputs needed to be understandable, traceable and reusable inside existing commercial and service processes, rather than remaining a stand-alone analytics exercise.
NTT DATA combined data preparation, language analysis, deep-learning classification and operational reporting to create a workflow that could move from historical information to day-to-day case routing.
Processed and analysed five years of customer claims and requests, including free text, target categories, language detection, channel and subtype analysis, annual language evolution and dictionary construction.
Built an ensemble of deep-learning models designed to assign each request or claim to the most appropriate category.
Organized the process from data preparation and feature definition through modelling, scoring and operational consumption of results.
Produced technical evaluation and recurring reporting so classification quality could be monitored and outputs could be incorporated into customer, case or action prioritization.
The initiative achieved 95% top-3 classification accuracy and 82% top-1 classification accuracy, with 25% of FTE capacity available for reallocation and an approximate annual classification budget saving of EUR 215k. The source also indicates improved customer-service quality and response time through more reliable routing, reduced manual classification workload and less rework from incorrect category assignments. Together, these results establish a repeatable AI pattern for high-volume customer interaction processes.
success stories
