Probabilistic ACD-CRM Record Matching Without Shared Identifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for joining Automated Contact Distribution (ACD) records with Customer Relationship Management (CRM) records in contact centers are imprecise due to the lack of shared unique identifiers, asynchronous system clocks, and inconsistencies in agent and customer identifiers, making it difficult to accurately correlate contact and customer data.
Innovation Solution
A probabilistic attribution method that involves creating subsets of ACD and CRM records based on unique identifiers where available, performing precise joins where possible, and using a predictive model to assign probabilistic correlations where unique identifiers are absent, thereby generating a consolidated outcome matrix for improved data correlation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional joining methods (timestamps, Agent ID, Customer ID) are used to match ACD records with CRM records, then the joining process can be performed without unique identifiers, but the matching accuracy deteriorates due to asynchronous system clocks and identifier inconsistencies
Solution Approach 1:
The patent introduces a probabilistic attribution model as an intermediary between ACD and CRM systems. This model uses machine learning to calculate probability scores that indicate the likelihood of a CRM record being the outcome of an ACD contact, serving as a mediator that resolves the mismatch between the two systems when traditional identifier-based joining fails
Solution Approach 2:
The patent transforms the joining problem from deterministic identifier matching to probabilistic scoring. By changing the parameter from binary match/no-match to a continuous probability score, the system can handle asynchronous clocks and identifier inconsistencies while maintaining high matching accuracy through learned patterns in the data
2Measurement precision
If probabilistic attribution models are used to improve matching accuracy, then data correlation precision improves, but system complexity increases due to model training and calculation requirements
Solution Approach 1:
The patent performs model training in advance using historical ACD and CRM data to learn the relationships between contacts and outcomes. This preliminary action creates a trained model that can then be applied to new records without requiring complex real-time computation, reducing the burden on the production system
Solution Approach 2:
The probabilistic attribution model is trained on the organization's own historical data, allowing the system to learn and adapt to its specific patterns and relationships. This self-service approach enables the model to improve over time using its own operational data without requiring external intervention or complex configuration
Data Source
AI summary
Techniques for assigning values from a customer record system to records in a contact record system are disclosed. The method of the present disclosure includes: receiving a first set of records in the contact record system; receiving a second set of records in the customer record system; creating a third set that is a subset of the first set; creating a fourth set that is a subset of the first set; creating a fifth set that is a subset of the second set; creating a sixth set that is a subset of the second set; creating a seventh set based on the third set and the fifth set; creating an eighth set based on the fourth set and the sixth set; training a model based on the seventh set to achieve a trained model; and creating a ninth set based on the eighth set and the trained model.