Sequence-Based Pairing Benchmarks for Contact Center Algorithms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for benchmarking contact center pairing strategies face challenges such as measurement errors due to agent performance variations, time-based biases, and random assignment biases, leading to inaccurate performance comparisons between algorithms.
Innovation Solution
A method involving sequence number-based alternation of pairing strategies, where contacts are assigned to agents using a first and second strategy based on sequence numbers, with a target number and ratio of contacts, and metrics are determined to compare performance, ensuring fair and accurate benchmarking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If agents are divided into multiple pools for testing different algorithms, then performance comparison between algorithms can be conducted, but measurement errors occur due to variations in agent performance ability across pools
Solution Approach 1:
The patent applies periodic action by alternating between test and control algorithms in sequential time periods (e.g., first period uses test algorithm, second period uses control algorithm, then repeats). This time-based alternation ensures that both algorithms are evaluated under identical agent pools and contact conditions, eliminating between-pool variability while maintaining measurement precision through systematic comparison.
Solution Approach 2:
The patent maintains continuity of useful action by keeping the same agent pool available throughout the entire benchmarking process, continuously alternating algorithm application between time periods rather than dividing agents into separate pools. This ensures uninterrupted evaluation with consistent resources, improving both measurement precision and reliability.
2Measurement precision
If contacts are divided into groups for algorithm testing, then performance metrics can be calculated, but time-based biases affect the results due to patterns in contact behavior correlated with arrival time
Solution Approach 1:
The patent uses periodic alternation of algorithms across time periods, where each algorithm is applied for a defined duration before switching. This systematic time-based approach allows for controlled comparison while the patent simultaneously addresses time-based biases through additional mechanisms like randomization within periods or statistical adjustments, preventing contact arrival time patterns from skewing results.
Solution Approach 2:
The patent introduces an intermediary benchmarking system that mediates between the test and control algorithms, using sequence numbers and structured assignment rules to distribute contacts fairly. This intermediary layer ensures that neither algorithm systematically receives contacts at advantageous times, eliminating time-based biases while maintaining measurement precision.
3Object-affected harmful factors
If random assignment of contacts to algorithms is used, then time-based artifacts are reduced, but other sources of bias are introduced affecting performance comparison
Solution Approach 1:
The patent combines periodic alternation with controlled randomization, where the overall structure follows regular time-based periods but within each period, specific assignment decisions may incorporate random elements. This hybrid approach reduces time-based artifacts while the periodic framework maintains measurement precision by ensuring balanced exposure of both algorithms to similar contact patterns.
Solution Approach 2:
The patent applies preliminary action by pre-establishing the periodic alternation schedule and assignment rules before benchmarking begins. Sequence numbers are assigned in advance, and the alternation pattern is predetermined, which prevents both time-based artifacts and random biases from affecting the comparison. This structured preliminary setup ensures fair and accurate performance measurement.
4Productivity
If PBR algorithms are used to select highest value contacts, then agent performance is improved, but waiting time for other contacts in queue increases reducing overall customer experience
Solution Approach 1:
The patent applies partial action by implementing PBR algorithms selectively during benchmarking periods rather than continuously, alternating with control algorithms that use different assignment criteria. This allows measurement of PBR's impact on agent performance while capturing the trade-off with waiting times, enabling optimized partial deployment strategies that balance productivity gains against customer experience losses.
Solution Approach 2:
The patent implements feedback mechanisms by measuring and comparing performance metrics including agent productivity and contact waiting times across different algorithm periods. This feedback loop enables identification of optimal PBR implementation strategies that achieve desired performance improvements while minimizing negative impacts on overall customer experience, allowing dynamic adjustment of algorithm parameters.
Data Source
AI summary
Provided are techniques at least two pairing strategies in a contact center system, involving a method, a system and an article of manufacture. The method of present disclosure includes: assigning sequence numbers to a series of events; initiating a first pairing strategy based on the sequence numbers; ending the first pairing strategy based on the sequence numbers; initiating a second pairing strategy based on the sequence numbers; ending the second pairing strategy based on the sequence numbers; assigning a first set of values to contacts assigned by the first pairing strategy; assigning a second set of values to contacts assigned by the second pairing strategy; and determining a metric that compares the first pairing strategy with the second pairing strategy based on the first and second set of values. The techniques of present disclosure can get rid of shortcomings and/or drawbacks of a time-based benchmark.