Systems and methods relating to implementation of predictive models in contact centers

1 . A method for facilitating an operational transition between a first model and a second model in a contact center, wherein the first model and the second model each comprises a predictive model for calculating an output related to a customer interacting with the contact center, the method comprising the steps of:

identifying the first model and the second model, wherein the first model and the second model are machine learning models that are trained based on one or more regression algorithms and empirical data;

defining a transition window, the transition window comprising a predetermined timeframe for operationally transitioning between the first model and the second model;

defining a pre-transition window, the pre-transition window comprising a predetermined timeframe occurring just previous to the transition window;

during the pre-transition window:

using the first model to determine first outputs in relation to respective pre-transition customers, which are ones of the customers interacting with the contact center during the pre-transition window, and, from the first outputs, determining first output probability distributions;

using, in real-time, the first outputs operationally in relation to the pre-transition customers;

during the transition window:

using the second model to determine second outputs in relation to respective transition customers, which are ones of the customers interacting with the contact center during the transition window, and, from the second outputs, determining second output probability distributions;

determining a target output probability distribution;

transforming a current one of the second output probability distributions into the target output probability distribution based on inverse transform sampling that includes determining a cumulative distribution function for a starting distribution and a target distribution and computationally determining an inversion of the target distribution and, based on the transformation, determining a transformed output for each of the second outputs contained in the current one of the second output probability distributions; and

compensating for model drift and reducing an impact on resource utilization in the contact center in a performance of automated actions in response to the transformed outputs by using, in real-time, the transformed outputs operationally in relation to the transition customers;

wherein the target output probability distribution comprises a mean output probability distribution derived from:

one or more of the first probability output distributions; and

one or more of the second output probability distributions.

2 . The method of claim 1 , wherein the output indicates a likelihood of a desired outcome occurring in relation to the customer based on behavior data associated with the customer; and

wherein the output is scored in relation to a predetermined action threshold for selectively determining whether to perform an automated action in relation to the customer for achieving the desired outcome, the automated action being taken when the output satisfies the action threshold and not taken when the output does not satisfy the action threshold.

3 . The method of claim 2 , further comprising the steps of:

determining that the predetermined timeframe of the transition period has ended; and

using, in real-time, the second outputs of the second model operationally in relation to respective post-transition customers, which are ones of the customers interacting with the contact center after the transition window has ended.

4 . The method of claim 2 , wherein the first and second model comprise versions of a same machine learning model, with the second model comprising a version of the first model that is retrained by a more current training dataset.

5 . The method of claim 2 , wherein the mean output probability distribution is weighted so to increasingly emphasize an importance of the one or more second output probability distributions over an importance of the one or more first probability output distributions as the transition window transpires; and

wherein the one or more second output probability distributions are each different from the current one of the second output probability distributions.

6 . The method of claim 2 , wherein the transforming the current one of the second output probability distributions into the target output probability distribution comprises mapping the current one of the second output probability distributions to a target output probability distribution so that a rank of output probabilities is preserved.

7 . The method of claim 2 , wherein the pre-transition window and the transition window are defined so that each is divided into sequential operating intervals; and

wherein:

the first probability output distributions of the first model correspond to respective ones of the operating intervals that make up the pre-transition window; and

the second output probability distributions of the second model correspond to respective ones of the operating intervals that make up the transition window.

8 . The method of claim 7 , wherein, as the transition window transpires, the mean output probability distribution is derived from:

an increasing number of the second probability output distributions; and

a decreasing number of the first probability output distributions.

9 . The method of claim 8 , further comprising the steps of:

determining a lookback period, wherein the lookback period comprises a sliding window of a predetermined number of the operating intervals measured from and occurring just prior to the current one of the operating intervals;

wherein the increasing number of the second probability output distributions and the decreasing number of the first probability output distributions are determined in relation to which of the operating intervals are captured within the lookback period as the lookback period advances in relation to the current one of the operating intervals during the transition window.

10 . The method of claim 9 , further comprising the steps of:

determining that the predetermined timeframe of the transition window has ended; and

using, in real-time, the second outputs of the second model operationally in relation to respective post-transition customers, which are ones of the customers interacting with the contact center after the transition window has ended.

11 . The method of claim 9 , wherein the first and second model comprise versions of a same machine learning model, with the second model comprising a version of the first model that is retrained by a more current training dataset.

12 . The method of claim 9 , wherein the transforming the current one of the second output probability distributions into the target output probability distribution comprises mapping the current one of the second output probability distributions to a target output probability distribution so that a rank of output probabilities is preserved.

13 . A system for facilitating an operational transition between a first model and a second model in a contact center, wherein the first model and the second model each comprises a predictive model for calculating an output related to a customer interacting with the contact center, the system comprising:

a processor; and

a memory, wherein the memory stores instructions that, when executed by the processor, cause the processor to perform the steps of:

identifying the first model and the second model, wherein the first model and the second model are machine learning models that are trained based on one or more regression algorithms and empirical data;

defining a transition window, the transition window comprising a predetermined timeframe for operationally transitioning between the first model and the second model;

defining a pre-transition window, the pre-transition window comprising a predetermined timeframe occurring just previous to the transition window;

during the pre-transition window:

using the first model to determine first outputs in relation to respective pre-transition customers, which are ones of the customers interacting with the contact center during the pre-transition window, and, from the first outputs, determining first output probability distributions;

using, in real-time, the first outputs operationally in relation to the pre-transition customers;

during the transition window:

using the second model to determine second outputs in relation to respective transition customers, which are ones of the customers interacting with the contact center during the transition window, and, from the second outputs, determining second output probability distributions;

determining a target output probability distribution;

transforming a current one of the second output probability distributions into the target output probability distribution based on inverse transform sampling that includes determining a cumulative distribution function for a starting distribution and a target distribution and computationally determining an inversion of the target distribution and, based on the transformation, determining a transformed output for each of the second outputs contained in the current one of the second output probability distributions; and

compensating for model drift and reducing an impact on resource utilization in the contact center by in a performance of automated actions in response to the transformed outputs by using, in real-time, the transformed outputs operationally in relation to the transition customers;

wherein the target output probability distribution comprises a mean output probability distribution derived from:

one or more of the first probability output distributions; and

one or more of the second output probability distributions.

14 . The system of claim 13 , wherein the output indicates a likelihood of a desired outcome occurring in relation to the customer based on behavior data associated with the customer; and

wherein the output is scored in relation to a predetermined action threshold for selectively determining whether to perform an automated action in relation to the customer for achieving the desired outcome, the automated action being taken when the output satisfies the action threshold and not taken when the output does not satisfy the action threshold.

15 . The system of claim 14 , wherein the instructions, when executed by the processor, further cause the processor to perform the steps of:

determining that the predetermined timeframe of the transition period has ended; and

using, in real-time, the second outputs of the second model operationally in relation to respective post-transition customers, which are ones of the customers interacting with the contact center after the transition window has ended.

16 . The system of claim 14 , wherein the mean output probability distribution is weighted so to increasingly emphasize an importance of the one or more second output probability distributions over an importance of the one or more first probability output distributions as the transition window transpires; and

wherein the one or more second output probability distributions are each different from the current one of the second output probability distributions.

17 . The system of claim 14 , wherein the transforming the current one of the second output probability distributions into the target output probability distribution comprises mapping the current one of the second output probability distributions to a target output probability distribution so that a rank of output probabilities is preserved.