Automated Customer Service Interaction Classification Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated customer service systems, such as IVR, face challenges in balancing cost and customer satisfaction due to difficulties in determining when an interaction is 'going bad,' leading to either unnecessary transfers to live agents or lower customer satisfaction.
Innovation Solution
A computerized method that classifies historical interactions, trains classification models, and uses a log likelihood ratio to determine when to transfer a user from an automated service to a live agent, with dynamic threshold adjustment based on external factors and user lifetime value, employing N-gram and Boostexter classification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated customer service is used to reduce costs, then operational efficiency is improved, but customer satisfaction deteriorates
Solution Approach 1:
The patent introduces a classification model as an intermediary between the automated system and the live agent. This model analyzes interaction data in real-time and predicts the likelihood of negative outcomes, serving as a mediator that determines when automated service should be interrupted and a live agent should be introduced, thereby maintaining both cost efficiency and customer satisfaction
Solution Approach 2:
The system dynamically adjusts the interaction by transitioning from fully automated mode to live agent mode based on real-time predictions. The classification model continuously evaluates interaction data and adapts the service delivery mode, making the system flexible rather than static, thus resolving the contradiction between automation efficiency and service quality
2Reliability
If an oversensitive algorithm is used to detect bad interactions, then customer satisfaction is improved through timely transfers, but cost increases due to unnecessary live agent interventions
Solution Approach 1:
The classification model uses feedback from historical interaction data and real-time interaction metrics to continuously refine its predictions. By incorporating feedback loops that learn from past outcomes, the system optimizes its sensitivity thresholds, reducing false positives while maintaining accurate detection of truly problematic interactions
Solution Approach 2:
The system changes parameters such as the log likelihood ratio threshold dynamically based on learned patterns from training data. By adjusting these parameters optimally through the classification process, the system achieves the right balance between sensitivity and cost-effectiveness, avoiding both oversensitivity and undersensitivity
3Productivity
If an under-sensitive algorithm is used to detect bad interactions, then cost is reduced by minimizing live agent interventions, but customer satisfaction deteriorates due to missed transfers
Solution Approach 1:
The system performs preliminary classification and prediction during the interaction itself, analyzing multiple features and data points before making the transfer decision. This preliminary action allows the system to accumulate sufficient information to make accurate predictions, reducing the risk of missing problematic interactions while avoiding premature transfers
4Measurement precision
If complex classification models are used to accurately predict interaction outcomes, then detection precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the classification task into distinct components: feature extraction from interaction data, training data preparation, model training phase, and real-time prediction phase. This segmentation allows each component to be optimized independently, managing complexity while maintaining high detection precision through specialized processing at each stage
Data Source
AI summary
A balance between customer satisfaction and cost to providing customer care can be achieved based on the use of online interaction classification techniques. Such techniques can use measurements such as a log likelihood ratio to determine if an interaction should be removed from automation to live support if it appears necessary.