Unbiased Call Routing via Bias-Corrected Quality Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for evaluating and routing calls in internet telephony do not adequately account for biases caused by variations in customer traffic and sample size, leading to unfair comparisons and decisions about provider quality, which can result in suboptimal route selection and service quality.
Innovation Solution
A method and apparatus that determine a standard quality metric by accounting for customer traffic bias and sample size bias, allowing for unbiased evaluation of provider quality by decomposing quality metrics into customer contributions and adjusting standards based on the number of completed calls, thereby making fair decisions about routing and scrubbing providers from the routing lineup.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If standard quality metrics are used to evaluate providers, then routing decisions can be made automatically, but biases from customer traffic variations and sample size differences cause unfair comparisons
Solution Approach 1:
The patent adjusts quality metric thresholds dynamically based on provider-specific factors such as customer traffic volume and sample size. Instead of using fixed standards, the system modifies evaluation parameters to account for variations in provider characteristics, ensuring fair comparison while maintaining automated decision-making capability.
Solution Approach 2:
The patent introduces adjustment factors as intermediary elements between raw quality metrics and final provider evaluation. These intermediaries (bias correction factors, sample size weights) mediate the comparison process by normalizing differences in traffic volume and measurement reliability, enabling fair automated decisions without manual intervention.
2Reliability
If providers are scrubbed from routing lineup based on observed quality metrics, then poor quality routes are eliminated, but providers may be unfairly removed due to biases in the metrics
Solution Approach 1:
The patent modifies the scrubbing threshold by applying provider-specific adjustment factors. Providers with higher traffic volumes or more consistent performance receive different evaluation thresholds compared to smaller providers, preventing unfair removal while maintaining quality standards. The threshold becomes a dynamic parameter rather than a fixed value.
Solution Approach 2:
The patent performs preliminary bias correction and adjustment calculations before making the scrubbing decision. By pre-computing adjustment factors based on historical traffic patterns and sample size analysis, the system ensures that the final scrubbing decision is based on fair, adjusted metrics rather than raw observed values.
3Productivity
If quality metrics are compared across providers with different traffic volumes, then routing optimization is achieved, but providers with different sample sizes are compared unfairly
Solution Approach 1:
The patent applies sample size-based weighting factors to quality metrics before comparison. Providers with smaller traffic volumes receive adjusted thresholds that account for higher statistical variability, while providers with large volumes are evaluated against tighter standards. This parameter adjustment enables fair comparison across different scales.
Solution Approach 2:
The patent creates equipotential evaluation conditions by normalizing quality metrics through provider-specific adjustment factors. All providers are transformed to a common evaluation baseline that accounts for their traffic characteristics, ensuring that comparisons occur under equivalent conditions despite differences in volume and sample size.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A decision about provider quality based on a quality metric observed says little about the quality of the provider. Further, the decision may be biascd by a variation in customer contributions to the quality metric observed or by a variation in a number of completed calls received by a provider. Accordingly, a method and corresponding apparatus are provided to evaluate quality and to correct bias by determining a standard that accounts for at least one source of bias, comparing an observed measure of a provider against the standard to produce an evaluation of the observed measure of the provider, and making a decision about the quality of the provider based on the evaluation. As a result, an unbiased decision, for example, to scrub a provider can be made and in some instances, a provider may be rescued from being scrubbed.