Artificial intelligence payment timing models
Disclosed in some examples, are methods, systems, and machine-readable mediums which build and utilize an artificial intelligence model to predict debtor payment timing. The past debtor payment history and other debtor information may be for a plurality of accounts over a past time period. Once the model is created, it may be used when a debtor misses a payment to determine a prediction of when the debtor will pay. The model uses characteristics of past debtors and their payment dates to predict, based upon the characteristics of the late debtor, when the late debtor will make a payment. The predicted timing may include a predicted probability for whether the payment will be made within the predicted timing. The predicted timing may be a specific date, or a window (e.g., a three-day window).
1. A method for modifying a contact based upon a predicted payment date, the method comprising:
using a hardware processor configured to perform operations comprising:
training a prediction model using a gradient boosted tree machine-learning algorithm based on a training data set comprising a plurality of features including past payment history of a plurality of debtors on debt accounts and information about non-debt accounts of the plurality of debtors, the training data including at least one non-numerical feature, the training comprising converting the non-numerical features into a binary vector with each position of the vector representing a possible value of the non-numerical features, the model comprising one or more data structures that describe resultant nodes, leaves, and edges of the tree;
determining that a debtor has exceeded a threshold amount of time past a payment date of debt obligation without making a payment;
determining past payment history information of the debtor;
determining account history of a non-debt account of the debtor;
determining a predicted payment date using the past payment history information and the account history of the non-debt account of the debtor as inputs to the model; and
automatically causing a modification to a contact to the debtor related to the debt, the modification based upon the predicted payment date, the modification changing a timing of the contact, a type of the contact, or a call script of the contact.
2. The method of claim 1 , wherein causing a modification to the contact to the debtor comprises instructing a contact center system to delay a contact until after expiry of the predicted payment date.
3. The method of claim 1 , wherein causing a modification to the contact comprises modifying a call script to include a prompt to ask the debtor if they are paying by the predicted payment date.
4. The method of claim 1 , wherein the method further comprises:
determining that the predicted payment date has passed without the debtor making the payment; and
responsive to determining that the predicted payment date has passed without the debtor making the payment, determining a plurality of other payments the debtor has made within a predetermined threshold period of time;
determining that one of the plurality of other payments was misapplied;
responsive to determining that the one of the plurality of other payments was misapplied, reversing the one of the plurality of other payments and applying the one of the plurality of other payments to the debt.
5. The method of claim 1 , wherein causing a modification to the contact comprises changing a format of a contact from a first format to a second format, the first and second formats comprising one of a call, an email, a text message, or a notification.
6. A computing device for modifying a contact based upon a predicted payment date, the computing device comprising:
a hardware processor;
a memory, the memory storing instructions, the instructions, when executed by the hardware processor, causing the computing device to perform operations comprising:
training a prediction model using a gradient boosted tree machine-learning algorithm based on a training data set comprising a plurality of features including past payment history of a plurality of debtors on debt accounts and information about non-debt accounts of the plurality of debtors, the training data including at least one non-numerical feature, the training comprising converting the non-numerical features into a binary vector with each position of the vector representing a possible value of the non-numerical features, the model comprising one or more data structures that describe resultant nodes, leaves, and edges of the tree;
determining that a debtor has exceeded a threshold amount of time past a payment date of a debt obligation without making a payment;
determining past payment history information of the debtor;
determining account history of a non-debt account of the debtor;
determining a predicted payment date using the past payment history information and the account history of the non-debt account of the debtor as inputs to the model; and
automatically causing a modification to a contact to the debtor related to the debt, the modification based upon the predicted payment date, the modification changing a timing of the contact, a type of the contact, or a call script of the contact.
7. The computing device of claim 6 , wherein the operations of causing a modification to the contact to the debtor comprises instructing a contact center system to delay a contact until after expiry of the predicted payment date.
8. The computing device of claim 6 , wherein the operations of causing a modification to the contact comprises modifying a call script to include a prompt to ask the debtor if they are paying by the predicted payment date.
9. The computing device of claim 6 , wherein the operations further comprise:
determining that the predicted payment date has passed without the debtor making the payment; and
responsive to determining that the predicted payment date has passed without the debtor making the payment, determining a plurality of other payments the debtor has made within a predetermined threshold period of time;
determining that one of the plurality of other payments was misapplied;
responsive to determining that the one of the plurality of other payments was misapplied, reversing the one of the plurality of other payments and applying the one of the plurality of other payments to the debt.
10. The computing device of claim 6 , wherein the operations of causing a modification to the contact comprises changing a format of a contact from a first format to a second format, the first and second formats comprising one of a call, an email, a text message, or a notification.
11. A non-transitory machine-readable medium storing instructions, the instructions, when executed by a machine, cause the machine to perform operations comprising:
training a prediction model using a gradient boosted tree machine-learning algorithm based on a training data set comprising a plurality of features including past payment history of a plurality of debtors on debt accounts and information about non-debt accounts of the plurality of debtors, the training data including at least one non-numerical feature, the training comprising converting the non-numerical features into a binary vector with each position of the vector representing a possible value of the non-numerical features, the model comprising one or more data structures that describe resultant nodes, leaves, and edges of the tree;
determining that a debtor has exceeded a threshold amount of time past a payment date of a debt obligation without making a payment;
determining past payment history information of the debtor;
determining account history of a non-debt account of the debtor;
determining a predicted payment date using the past payment history information and the account history of the non-debt account of the debtor as inputs to the model; and
automatically causing a modification to a contact to the debtor related to the debt, the modification based upon the predicted payment date, the modification changing a timing of the contact, a type of the contact, or a call script of the contact.
12. The non-transitory machine-readable medium of claim 11 , wherein the operations of causing a modification to the contact to the debtor comprises instructing a contact center system to delay a contact until after expiry of the predicted payment date.
13. The non-transitory machine-readable medium of claim 11 , wherein the operations of causing a modification to the contact comprises modifying a call script to include a prompt to ask the debtor if they are paying by the predicted payment date.
14. The non-transitory machine-readable medium of claim 11 , wherein the operations further comprise:
determining that the predicted payment date has passed without the debtor making the payment; and
responsive to determining that the predicted payment date has passed without the debtor making the payment, determining a plurality of other payments the debtor has made within a predetermined threshold period of time;
determining that one of the plurality of other payments was misapplied;
responsive to determining that the one of the plurality of other payments was misapplied, reversing the one of the plurality of other payments and applying the one of the plurality of other payments to the debt.