| CPC G16H 20/10 (2018.01) [G06Q 30/0251 (2013.01); G06F 2218/12 (2023.01)] | 24 Claims |
|
1. A non-transitory computer-readable storage medium comprising processor-executable instructions, the instructions including:
receiving user data for a user derived from at least one or more user interactions with an initial service channel;
determining a set of features from the user data, wherein the set of features characterizes a likelihood of the user to change from the initial service channel to an alternative service channel;
generating an input encoding for the set of features from the user data;
training a predictive model with a plurality of training samples, wherein:
the plurality of training samples corresponds to recipients of one or more drug therapies via the initial service channel, and
the plurality of training samples includes training labels that indicate whether a respective recipient defected from the initial service channel;
determining, using a predictive model, a probability that indicates how likely the user is to change from the initial service channel to the alternative service channel prior to a service defection opportunity, wherein the predictive model is trained to:
receive, as input, the input encoding, and
generate, as output, a respective probability that indicates how likely a respective user is to change from the initial service channel to the alternative service channel;
determining whether the probability from the predictive model satisfies a first defection threshold or a second defection threshold;
generating a remedial action based on the probability from the predictive model, wherein generating the remedial action includes:
in response to a determination that the probability from the predictive model has met the first defection threshold, automatically transmitting a first communication using a first communication protocol, and
in response to a determination that the probability from the predictive model has met the second defection threshold, automatically transmitting a second communication using a second communication protocol that is different from the first communication protocol; and
validating an accuracy of the predictive model, wherein validating the accuracy includes:
determining, based on a comparison of a set of predicted probabilities from the predictive model to a set of actual results associated with the set of predicted probabilities, whether an accuracy of the predictive model meets an accuracy threshold; and
in response to a determination that the accuracy of the predictive model has not met the accuracy threshold:
obtaining a set of data associated with a second recipient,
determining training labels for the set of data associated with the second recipient,
generating a second plurality of training samples based on the determined training labels and the set of data associated with the second recipient; and
retraining the predictive model with the second plurality of training samples to increase the accuracy of the predictive model.
|