Dynamic Contact Center Routing via Machine Learning Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional skill-based routing in contact centers is static and does not adapt to real-time changes, leading to inefficient agent interaction allocation and requiring significant manual effort, resulting in suboptimal customer experience and increased costs.
Innovation Solution
A system that dynamically routes interactions to contact center agents based on real-time context data, including agent profiles, customer profiles, and interaction intent, using a network model to estimate the expected value of each agent and select the best fit, balancing exploration and exploitation needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional skill-based routing is used, then agent skill matching is achieved, but the system is static and does not adapt to real-time changes
Solution Approach 1:
The patent transforms the static skill-based routing system into a dynamic one by continuously gathering context data (agent state, customer history, interaction type) and using machine learning models to adapt routing decisions in real-time. The system dynamically adjusts routing based on changing conditions rather than relying on fixed skill categories.
Solution Approach 2:
The system uses automated machine learning models to autonomously make routing decisions based on gathered context data, eliminating the need for manual routing configuration and reducing dependency on pre-defined skill models. The system self-optimizes by learning from past interactions and automatically adapting to new patterns.
2Reliability
If traditional skill-based routing is used, then routing decisions are made quickly, but the customer experience is suboptimal
Solution Approach 1:
The system pre-gathers context data about agents (skills, availability, performance metrics) and customers (history, preferences, issue type) before routing decisions are needed. Machine learning models are pre-trained on historical data to enable rapid real-time predictions, so when a routing decision is required, the system can quickly retrieve and evaluate relevant context without extensive processing.
3Measurement precision
If refined skill models are constructed manually, then routing precision is improved, but the cost and effort increase significantly
Solution Approach 1:
The system uses automated machine learning models that self-train on historical interaction data to continuously refine routing precision without manual intervention. The models automatically learn from past routing outcomes and improve their accuracy over time, eliminating the need for manual model construction and maintenance while maintaining high routing precision.
Solution Approach 2:
The system implements feedback loops where routing outcomes are continuously monitored and fed back into the machine learning models. This feedback mechanism allows the models to automatically learn from successes and failures, continuously improving routing precision without requiring manual model reconstruction. The system adapts its routing logic based on actual performance data.
4Adaptability or versatility
If a large pool of agents with equivalent skills is created, then routing flexibility is improved, but the matching efficiency decreases
Solution Approach 1:
Instead of treating all agents with equivalent skills uniformly, the system applies local quality by considering individual agent characteristics (current state, performance history, specific strengths) even among agents with similar skill sets. The machine learning model evaluates each agent's unique attributes to make precise matching decisions, thereby maintaining flexibility while improving matching efficiency through personalized evaluation.
Data Source
AI summary
A system that is adapted to route interactions to contact center agents. More specifically, the system is adapted to identify an interaction to be routed, and identify a group of agents based on one or more constraints for generating one or more candidate agents. The system is also adapted to gather context data surrounding the candidate agents. For each agent of the candidate agents, the system is adapted to estimate an expected value to be obtained by routing the interaction to the agent. The system is further adapted to select a particular agent of the candidate agents based on the estimates, and signal a routing device for routing the interaction to the particular agent.