Information processing device, information processing method, and program

Abstract

The purpose is to improve the efficiency of debt collection activities against debtors. [Solution] The information processing device 1 according to this embodiment includes a storage unit 12 that stores a classification model which is a machine learning model that has been trained using multiple sets of debtor attributes and effectiveness data indicating the effectiveness of reminders to encourage payment from debtors as training data, and which outputs effectiveness data when debtor attributes are input; an effectiveness identification unit 131 that identifies the effectiveness of reminders for each of the multiple target debtors based on the effectiveness data output by the classification model into which the attributes of the target debtors to be classified have been input; and a reminder target determination unit 132 that determines some of the multiple target debtors with relatively high reminder effectiveness as targets for reminders.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Techniques for improving the efficiency of debt collection operations are known (see Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Among debtors, there are those for whom collection efforts are effective, i.e., the probability of payment improves when a demand notice is issued. On the other hand, there are also those for whom collection efforts are not effective, such as those who will pay without a demand notice or those who will not pay even with a demand notice. The resources that a financial institution or the like that has provided financing to a debtor can use for issuing demand notices to the debtor are limited. Therefore, it is important to conduct efficient collection efforts, but there is a problem that efficiency is not good when demand notices are issued to those for whom collection efforts are not effective.

[0005] Therefore, the present invention has been made in view of these points, and an object thereof is to improve the efficiency of the collection operation for debtors.

Means for Solving the Problems

[0006] An information processing device according to a first aspect of the present invention includes: a storage unit that stores a classification model which is a machine learning model trained using multiple sets of debtor attributes and effectiveness data indicating the effectiveness of reminders to encourage payment to the debtor as training data, and which outputs the effectiveness data when the attributes of the debtor are input; an effectiveness identification unit that identifies the effectiveness of reminders for each of the multiple target debtors based on the effectiveness data output by the classification model when the attributes of the target debtors to be classified are input; and a reminder target determination unit that determines some of the target debtors among the multiple target debtors whose reminder effectiveness is relatively high as targets for reminders.

[0007] The classification model may output, in association with each of the multiple target debtors, a first effectiveness data indicating that the probability of payment being made without reminder is greater than or equal to a first threshold, a second effectiveness data indicating that the probability of payment being made after reminder is less than a second threshold (smaller than the first threshold), or a third effectiveness data indicating that the probability of payment being made without reminder is less than the first threshold, and the probability of payment being made after reminder is greater than or equal to the second threshold.

[0008] The debt collection target determination unit may determine the target debtor corresponding to the third effectiveness data to be the debt collection target.

[0009] The storage unit may store a debt collection capability value indicating the ability of the debt collection officer to perform debt collection duties, and the debt collection target determination unit may determine that the debt collection targets are those among the target debtors corresponding to the second effectiveness data whose outstanding debt balance is above a threshold, and that the debt collection officer whose debt collection capability value meets predetermined conditions will perform the debt collection duties.

[0010] The debt collection target determination unit may determine that some of the target debtors, from among the multiple target debtors, have a relatively high debt collection score calculated based on the effectiveness of the debt collection and the debtor's outstanding debt, as the target debtors for debt collection.

[0011] The storage unit may further store a payment prediction model that outputs the probability of payment for each collection method when the attributes of the debtor or the effectiveness data of the debtor are input, and the information processing device may further have a collection method determination unit that determines the collection method with a relatively high probability of payment output by the payment prediction model, which has the attributes of the person to be collected or the outputted effectiveness data of the debtor, as the collection method to be executed for each of the persons to be collected.

[0012] The debt collection means determination unit may allocate the total number of times debt collection can be performed to each of the debt collection means for each of the debt collection targets, such that the sum of the values ​​obtained by multiplying the probability of payment corresponding to the debt collection means for each debtor by the debtor's outstanding debt and the number of times the debt collection means has been performed, multiplied by the number of debtors, maximizes the total number of times debt collection can be performed for each debtor.

[0013] The debt collection means determination unit may determine the automatic phone call to be the debt collection means to be executed if the difference between the probability of payment made by manual phone calls by debt collection officers and the probability of payment made by automatic phone calls, as output by the payment prediction model, is less than a threshold.

[0014] The storage unit may further store a response probability prediction model that outputs the probability of the debtor responding to the demand for payment for each timing of the demand for payment, when the attributes of the debtor are input. The information processing device may further include a demand timing determination unit that determines the timing at which the response probability output by the response probability prediction model, into which the attributes of the persons to be demanded are input, is relatively high, as the demand timing for each of the persons to be demanded.

[0015] The storage unit may store the availability status of debt collection officers on a daily basis, and the information processing device may further include a schedule generation unit that generates a debt collection schedule by assigning debt collection tasks to debt collection officers who are available on the date and time corresponding to the debt collection timing.

[0016] The schedule generation unit may generate the debt collection schedule by assigning debt collection tasks to the debt collection officer in order of the period between the debt collection date issued by the debt collection officer and the payment execution date on which the debt collection target made the payment, starting with those targeting the debt collection officer.

[0017] The storage unit may further store a response probability prediction model that outputs the probability of the debtor responding to the collection for each collection method when the attributes of the debtor are input, and the information processing device may further have a collection method determination unit that determines the collection method with a relatively high response probability output by the response probability prediction model, into which the attributes of the person being collected have been input, as the collection method to be used for each of the persons being collected.

[0018] An information processing device according to a second aspect of the present invention includes: a storage unit that stores an effectiveness table in which the attributes of a debtor and effectiveness data indicating the effectiveness of reminders to encourage payment from the debtor are associated; an effectiveness identification unit that identifies the effectiveness of reminders for each of a plurality of target debtors based on the effectiveness data associated in the effectiveness table with the attributes of the debtors that have a similarity of a predetermined threshold or higher to the attributes of the target debtors to be classified; and a reminder target determination unit that determines some of the target debtors among the plurality of target debtors whose reminder effectiveness is relatively high as targets for reminders.

[0019] A third aspect of the present invention relates to an information processing method comprising: an effectiveness identification step, which is performed by a computer having a storage unit that stores a classification model that outputs the effectiveness data when the attributes of the debtors to be classified are input, and which identifies the effectiveness of the collection efforts for each of the multiple target debtors based on the effectiveness data output by the classification model into which the attributes of the target debtors to be classified have been input; and a collection target determination step, which determines that some of the multiple target debtors whose collection efforts are relatively high will be collected as collection targets.

[0020] The program according to the fourth aspect of the present invention is a machine learning model that is machine-learned using a plurality of sets of debtor attributes and effectiveness data indicating the effectiveness of a reminder for prompting payment to the debtor as teacher data. When the debtor attributes are input, a processor of an information processing device having a storage unit that stores a classification model for outputting the effectiveness data outputs the effectiveness data. Based on the effectiveness data output by the classification model when the attributes of the target debtor to be classified are input, an effectiveness specifying unit that specifies the reminder effectiveness for each of the plurality of target debtors, and among the plurality of target debtors, a reminder target determination unit that determines a part of the target debtors with relatively high reminder effectiveness as the reminder target persons are made to function.

Effects of the Invention

[0021] According to the present invention, there is an effect that the efficiency of the reminder service for debtors can be improved.

Brief Description of the Drawings

[0022] [Figure 1] It is a diagram showing an outline of the operation of the information processing system S. [Figure 2] It is a flowchart showing the flow of processing from debtor classification to reminder schedule generation. [Figure 3] It is a diagram showing an example of the configuration of the information processing device 1. [Figure 4] It is a diagram showing an example of a reminder ability value table. [Figure 5] It is a diagram showing an example of a reminder classification table. [Figure 6] It is a diagram showing an example of a reminder means table. [Figure 7] It is a diagram showing an example of a reminder execution count table. [Figure 8] It is a diagram showing an example of a reminder timing table. [Figure 9] It is a flowchart showing the flow of processing related to the generation of a reminder schedule. [Figure 10]This figure shows an example of a debt collection schedule table. [Figure 11] This figure shows an example of an effectiveness table. [Modes for carrying out the invention]

[0023] [Overview of Information Processing System S] The outline of the information processing system S according to this embodiment will be explained using Figures 1 and 2. Figure 1 is a diagram showing an overview of the operation of the information processing system S. The information processing system S comprises an information processing device 1 and an information terminal 2. The information processing system S may also include other equipment such as servers and terminals.

[0024] Among debtors, there are those who fail to make payments by the due date specified by the financial institution that provided the loan. While the number of such delinquent debtors is enormous, financial institutions face the problem of incurring significant costs and effort if they continue to pursue payment from delinquent debtors for whom collection efforts are ineffective. In particular, when collection agents directly call debtors to urge payment, they need to effectively utilize their limited number of agents. Therefore, the information processing device 1 identifies a select group of debtors for whom collection efforts are relatively effective as targets. This reduces the number of debtors targeted for collection, thereby improving the efficiency of collection operations.

[0025] Information processing device 1 is a computer, such as a server, that determines which individuals are subject to debt collection. Information terminal 2 is an information terminal used by the debt collection officer or their manager. Information terminal 2 may be a stationary terminal such as a desktop personal computer, or a portable terminal such as a smartphone or tablet personal computer. Information processing device 1 and information terminal 2 are connected to each other via a communication network such as the Internet.

[0026] Referring to Figure 1, the overview of the processing performed by the information processing device 1 will be explained. The information processing device 1 stores a classification model that has been machine-trained using multiple sets of data as training data, which consist of the attributes of debtors and effectiveness data indicating the effectiveness of reminders to encourage payment from debtors. When the attributes of the target debtor SA to be classified are input to the classification model, it outputs effectiveness data for the target debtor SA.

[0027] The information processing device 1 classifies multiple target debtors SA into three groups based on the effectiveness data output by the classification model: a group that tends to make payments without reminders, a group that tends to make payments when reminders are made, and a group that tends not to make payments even when reminders are made. The information processing device 1 then transmits data indicating the group that tends to make payments when reminders are made as target debtors T to the information terminal 2 used by the debt collection officer P.

[0028] Debt collection officer P can identify debtors T from information terminal 2 and then contact them by phone, email, etc., to request payment. Since the number of debtors T is small compared to the total number of delinquent debtors, and debtors T are individuals for whom debt collection is effective, debt collection officer P can carry out debt collection work efficiently.

[0029] Furthermore, while debt collection officer P will prioritize debt collection from the debtor T who belongs to the group for which debt collection is effective, for example, as soon as such debt collection is completed or in parallel with such debt collection, debtor SA who belongs to the group for which debt collection is not necessary and debtor SA who belongs to the group for which debt collection is invalid may also be contacted. In other words, debt collection officer P is not excluded from contacting debtor SA other than the debtor T. In addition, debt collection officer P may also contact the guarantor of debtor T.

[0030] Incidentally, for debt collection officer P to carry out debt collection, it is desirable to have a debt collection schedule that shows which debt collection officer P will contact which debtor T, how many times they will contact them, and when, using what means of collection (telephone, email, etc.). Therefore, the information processing device 1 may generate a debt collection schedule. The following describes the flow of processing from the classification of target debtors SA to the generation of the debt collection schedule.

[0031] Figure 2 is a flowchart illustrating the processing flow from debtor classification to the generation of a debt collection schedule. First, as described above, the information processing device 1 classifies the target debtors SA into three groups: those who do not require debt collection, those for whom debt collection is effective, and those for whom debt collection is ineffective, and determines that those in the group for whom debt collection is effective are the debtors to be collected (S1). Next, the information processing device 1 determines, from among several debt collection methods, the debt collection method for each debtor to be collected that has a relatively high probability of making a payment (S2). Subsequently, the information processing device 1 determines the number of times to execute the debt collection method determined for each debtor to be collected in S2 in order to maximize the total amount recovered through debt collection (S3). For example, the information processing device 1 determines the number of times to execute the debt collection method for each debtor to be collected, such as by executing the debt collection method more often for debtors to be collected with a larger outstanding debt balance.

[0032] Next, the information processing device 1 determines for each person T to whom a debt collection notice is requested, the timing of the notice in which the probability of the person responding to the notice by phone, email, etc. is relatively high (S4). For example, the information processing device 1 determines for each person T whether the notice should be sent to their landline or mobile phone, and at what time of day on weekdays or holidays, to increase the probability of a response. Note that the processing in S4 may be performed before S2 or S3, as long as it is performed after S1.

[0033] Finally, the information processing device 1 generates a debt collection schedule (S5) indicating that debt collection officer P will perform the debt collection method determined in S2 on the debt collection target T determined in S1, the number of times determined in S3, at the debt collection timing determined in S4. Debt collection officer P then performs the debt collection by referring to the generated debt collection schedule.

[0034] [Configuration of Information Processing Device 1] The configuration of the information processing device 1 will be described. Figure 3 shows an example of the configuration of the information processing device 1. The information processing device 1 comprises a communication unit 11, a storage unit 12, and a control unit 13. The control unit 13 comprises an effectiveness determination unit 131, a reminder target determination unit 132, a reminder means determination unit 133, a reminder timing determination unit 134, and a schedule generation unit 135.

[0035] The communication unit 11 is a communication interface for communicating with the information terminal 2 used by the debt collection officer P via a communication network such as the Internet. The communication unit 11 transmits the debt collection classification table input from the debt collection target determination unit 132 to the information terminal 2. The communication unit 11 also transmits the debt collection schedule table input from the schedule generation unit 135 to the information terminal 2.

[0036] The storage unit 12 is a storage medium including ROM (Read Only Memory) and RAM (Random Access Memory). The storage unit 12 stores the program that the control unit 13 executes. The storage unit 12 stores an information processing program that causes the control unit 13 to function as, for example, an effectiveness determination unit 131, a reminder target determination unit 132, a reminder means determination unit 133, a reminder timing determination unit 134, and a schedule generation unit 135.

[0037] The memory unit 12 stores deposit history data. The deposit history data is data showing the debtor's past deposit history and is the subject of learning by the machine learning model. In the deposit history data, information for identifying the debtor, the debtor's attributes, the deposit deadline, the date the debtor made the deposit, the deposit amount, and if a debt collection officer P made a demand for payment, further information for identifying the debt collection officer P, the means of demanding payment, and the date and time of the demand for payment are associated. Debtor attributes include, for example, the debtor's employment status (part-time worker, temporary worker, company employee, etc.), marital status (single, married), living situation (living alone, living with parents, renting, owning), family structure, age, occupation, workplace, work history, educational background, annual income, gender, personality, or history of using financial institutions.

[0038] The memory unit 12 stores debtor list data. The debtor list data is data that shows a list of debtors who currently have debts. The debtors included in the debtor list become target debtors SA, which are the subjects of classification by the classification model described later. In the debtor data, information for identifying the debtor, the debtor's attributes, the current debt balance, and the payment due date are associated.

[0039] The memory unit 12 stores a classification model for classifying multiple target debtors SA according to the effectiveness of debt collection efforts. The classification model is a machine learning model that uses multiple sets of debtor attributes and effectiveness data indicating the effectiveness of debt collection efforts to encourage payment to the debtors as training data. For example, the classification model uses payment history data to learn from the debtor's attributes, information indicating whether the debtor has made a payment, and, if a payment has been made, information indicating whether the payment was made before or after the debt collection effort by debt collection officer P. When the debtor's attributes are input to the classification model, it outputs effectiveness data. For example, when the debtor's attributes are input to the classification model, it outputs the pre-debt payment probability, which is the probability that payment will be made without debt collection efforts, and the post-debt payment probability, which is the probability that payment will be made after debt collection efforts, as effectiveness data, associated with each of the multiple debtors.

[0040] The memory unit 12 stores a payment prediction model for predicting the probability of payment by different collection methods. The payment prediction model is machine-learned using, for example, payment history data, the attributes of the debtor, and information indicating whether or not payment was made when collection was made by a predetermined collection method. The payment prediction model may also be machine-learned using effectiveness data output by a classification model and information indicating whether or not payment was made when collection was made by a predetermined collection method. Examples of collection methods include manual calls by collection officer P, automated calls such as predictive dialing (PD) or calls using an interactive voice response (IVR) system, email, short message service (SMS), or written letters or postcards. When the payment prediction model receives debtor attribute or effectiveness data as input, it outputs the probability of payment by different collection methods.

[0041] The memory unit 12 stores a response probability prediction model for predicting the probability of a debtor responding to a debt collection notice at different timings. The response probability prediction model is machine-learned using, for example, payment history data, and includes the debtor's attributes, the date and time of the notice, and information indicating whether the person T being asked for payment responded when the notice was made at that time. When the debtor's attributes are input, the response probability prediction model outputs the probability of the debtor responding to the notice at different timings.

[0042] The response probability prediction model may also be machine-learned using information indicating whether the debt collection is conducted via a landline or mobile phone to the debtor T. In this case, given the debtor's attributes as input, the response probability prediction model outputs the debtor's response probability to the debt collection for each collection method, either landline or mobile phone.

[0043] The memory unit 12 stores a debt collection ability value table that shows the debt collection ability value of debt collection officer P, which indicates P's ability to perform debt collection duties. Figure 4 is a diagram showing an example of the debt collection ability value table. In the debt collection ability value table, the debt collection officer ID and the debt collection ability value are associated. The debt collection officer ID is information used to identify debt collection officer P. The debt collection ability value is determined based on the debt collection officer P's years of experience in debt collection, the total number of debt collections performed, the success rate of debt collections, and the evaluation of debt collection duties from their superior. The debt collection ability value may be classified according to the skill level of debt collection officer P (expert level, intermediate level, beginner level), as shown in Figure 4, or it may be a numerical value such as a score or standard score.

[0044] The memory unit 12 stores an availability status table showing the availability of debt collection officer P on a daily basis. The availability status table is generated, for example, when debt collection officer P registers with the information processing device 1, at a time such as the end of the month, the dates and times on which they can perform debt collection work in the following month.

[0045] The control unit 13 is, for example, a CPU (Central Processing Unit). By executing the information processing program stored in the memory unit 12, the control unit 13 functions as an effectiveness determination unit 131, a reminder target determination unit 132, a reminder means determination unit 133, a reminder timing determination unit 134, and a schedule generation unit 135.

[0046] The effectiveness determination unit 131 determines the effectiveness of debt collection for each of the multiple target debtors SA based on the effectiveness data output by the classification model, which has been input with the attributes of the multiple target debtors SA to be classified. For example, when the classification model receives the attributes of target debtors SA included in the debtor list shown by the debtor list data stored in the memory unit 12, it outputs the pre-debt payment probability and post-debt payment probability as effectiveness data, associated with each of the multiple target debtors SA. In this case, the effectiveness determination unit 131 determines the effectiveness of debt collection for each of the multiple target debtors SA such that the debt collection effectiveness is higher for target debtors SA whose value obtained by subtracting the output pre-debt payment probability from the output post-debt payment probability is higher.

[0047] The classification model may, upon input of the attributes of target debtors SA included in the debtor list shown in the debtor list data, output first effectiveness data, second effectiveness data, or third effectiveness data associated with each of the multiple target debtors SA. First effectiveness data indicates that the probability of payment before reminder is above the first threshold (e.g., 80%). Second effectiveness data indicates that the probability of payment after reminder is below the second threshold (e.g., 60%), which is lower than the first threshold. Third effectiveness data indicates that the probability of payment before reminder is below the first threshold and the probability of payment after reminder is above the second threshold.

[0048] In this case, the effectiveness identification unit 131 classifies the target debtor SA corresponding to the first effectiveness data into the group that does not require collection, the target debtor SA corresponding to the second effectiveness data into the group that does not require collection, and the target debtor SA corresponding to the third effectiveness data into the group that requires collection. The effectiveness identification unit 131 identifies the collection effectiveness for each of the multiple target debtor SAs such that the collection effectiveness for the target debtor SA belonging to the group that requires collection is higher than the collection effectiveness for the target debtor SA belonging to the group that does not require collection and the collection effectiveness for the target debtor SA belonging to the group that does not require collection.

[0049] The debt collection target determination unit 132 determines that some of the target debtors SA, whose debt collection effectiveness is relatively high, will be designated as debt collection targets T. For example, the debt collection target determination unit 132 may determine that the target debtors SA corresponding to the third effectiveness data will be designated as debt collection targets T. The debt collection target determination unit 132 may also generate a debt collection classification table showing the debt collection classification group to which the target debtors SA belong and the debt collection targets T.

[0050] Figure 5 shows an example of a debt collection classification table. In the debt collection classification table, the debt collection classification group, the target debtor ID, the outstanding debt, the debt target flag, and the debt target ID are associated. The target debtor ID is information used to identify the target debtor SA that is subject to classification. The debt target flag is a flag used to identify the target debtor SA that is subject to debt collection. In Figure 5, all target debtor SAs belonging to the debt collection valid group are marked with a "○" flag, meaning they are subject to debt collection. The reason why target debtor SAs SA009 and SA011, which belong to the debt collection invalid group, are also marked with a "○" flag will be explained later. The debt target ID is information used to identify the debt target T that is subject to debt collection.

[0051] The debt collection target determination unit 132 may transmit the generated debt collection classification table to the information terminal 2 used by the debt collection officer P. This allows the debt collection officer P to prioritize debt collection for the debtor T, thereby improving the efficiency of debt collection operations for debtors. The debt collection target determination unit 132 may also store the generated debt collection classification table in the storage unit 12. This allows the schedule generation unit 135 to generate a debt collection schedule by referring to the debt collection classification table, as will be described in detail later.

[0052] Incidentally, among the target debtors SA belonging to the group for which debt collection is effective, corresponding to the third effectiveness data, there may be some whose probability of payment after debt collection is lower than or equal to the probability of payment before debt collection. If those who do not increase their probability of payment even after debt collection are excluded from the target debtors T, the debt collection officer P can carry out debt collection more efficiently. Therefore, the debt collection target determination unit 132 may determine that the target debtors T are those among the target debtors SA belonging to the group for which debt collection is effective, whose probability of payment after debt collection is higher than the probability of payment before debt collection. This allows the debt collection officer P to focus on debtors for whom debt collection is extremely effective.

[0053] While the target debtors SA corresponding to the first validity data have a high probability of making payments without reminders, there is a possibility that some may not make payments. Therefore, the reminder target determination unit 132 may determine that those target debtors SA corresponding to the first validity data who do not make payments within a predetermined period (for example, one week) from the payment deadline will be designated as reminder targets T. This maximizes the amount of debt recovered.

[0054] The debt collection target determination unit 132 may determine debtors T to be targeted for debt collection based on the outstanding debt of the target debtors SA, in addition to the effectiveness of the debt collection efforts, in order to maximize the amount of debt recovered through debt collection activities. For example, the debt collection target determination unit 132 may determine some of the target debtors SA to be targeted for debt collection T, from among a group of target debtors SA, whose debt collection score, calculated based on the effectiveness of the debt collection efforts and the debtor's outstanding debt, is relatively high. As an example, the debt collection target determination unit 132 may determine target debtors T to be targeted for debt collection T if the debt collection score for each target debtor SA, calculated by multiplying the effectiveness of the debt collection efforts identified by the effectiveness identification unit 131 by the outstanding debt, is above a threshold.

[0055] In the debt collection classification table shown in Figure 5, target debtors SA009 and SA011 belong to the group of debtors for whom debt collection is invalid, but because their outstanding debts are large, they are classified as target debtors T. In this way, the debt collection target determination unit 132 determines target debtors T based on the outstanding debts of target debtors SA, making it easier for target debtors SA with large outstanding debts to be classified as target debtors T. As a result, the amount of debt recovered through debt collection operations can be maximized.

[0056] However, recovering debts from target debtors SA belonging to the group of debtors whose debt collection efforts are deemed ineffective, as indicated in the second effectiveness data, is difficult. Therefore, the debt collection target determination unit 132 determines, for example, that large debtors among the target debtors SA corresponding to the second effectiveness data, whose outstanding debt balances are above a threshold, will be designated as debt collection targets T, for whom debt collection officers P whose debt collection ability values ​​meet predetermined conditions will be responsible for debt collection. As an example, the debt collection target determination unit 132 determines that large debtors among the target debtors SA belonging to the group of debt collection efforts are designated as debt collection targets T, for whom debt collection officers P whose debt collection ability values ​​are at the "expert level" in the debt collection ability value table shown in Figure 4 will be responsible for debt collection.

[0057] Thus, the debt collection target determination unit 132 decides that for large debtors among the debtors SA belonging to the group for which debt collection is invalid, a skilled debt collection officer P will carry out the debt collection. This increases the probability of recovering debts from debtors SA belonging to the group for which debt collection is invalid compared to when an intermediate or junior debt collection officer P carries out debt collection for those debtors SA belonging to the group for which debt collection is invalid. In addition, it reduces the loss of time and effort due to the failure of debt collection by intermediate or junior debt collection officers P, thus maintaining the motivation of intermediate or junior debt collection officers P.

[0058] After the debt collection target determination unit 132 determines the debtors T, the debt collection method determination unit 133 determines the debt collection method for each debtor T based on the payment probability for each debt collection method output by the payment prediction model for each debtor T. The debt collection method determination unit 133 determines the debt collection method to be executed for each debtor T to be the debt collection method with a relatively high payment probability output by the payment prediction model, which is input to the attributes of the debtor T or the effectiveness data output by the classification model. For example, the debt collection method determination unit 133 determines the one or more debt collection methods with a relatively high probability among the payment probabilities output by the payment prediction model for manual phone calls, automatic phone calls, emails, SMS messages, letters, and postcards by debt collection officer P to be the debt collection method to be executed. The debt collection method determination unit 133 may generate a debt collection method table showing the payment probability for each debtor T for each debtor T.

[0059] Figure 6 shows an example of a debt collection method table. In the debt collection method table, the debtor ID, the debt collection method, the probability of payment for each debt collection method, and the priority are associated. Debt collection methods with a higher probability of payment are assigned a higher priority. As will be described in detail later, the debt collection method determination unit 133 can determine the number of times each debt collection method will be executed by referring to the generated debt collection method table.

[0060] Incidentally, if the probability of payment being received through manual calls by debt collection officer P is about the same as the probability of payment being received through automated calls, it is preferable to use automated calls in order to reduce the workload of debt collection officer P. Therefore, if the difference between the probability of payment being received through manual calls by debt collection officer P and the probability of payment being received through automated calls, as output by the payment prediction model, is less than a threshold (for example, 10%), the debt collection means determination unit 133 determines that automated calls will be the debt collection means to be executed. This reduces the workload of debt collection officer P.

[0061] The debt collection method determination unit 133 may determine the debt collection method for each debt collection target T based on the response probability for each debt collection method output by the response probability prediction model for each debt collection target T. For example, the debt collection method determination unit 133 may determine the debt collection method to be executed for each debt collection target T to be the debt collection method with a relatively high response probability output by the response probability prediction model, which has the attributes of the debt collection target T as input.

[0062] For example, the debt collection method determination unit 133 may compare the probability of debtor T responding when debt collection is made by calling their landline phone, output by the response probability prediction model, with the probability of debtor T responding when debt collection is made by calling their mobile phone, and determine the debt collection method with the higher probability as the effective debt collection method. By doing so, the probability of debtor T responding to the debt collection increases, thereby improving the efficiency of debt collection operations against debtors.

[0063] After determining the appropriate collection method for each debtor T, the collection method determination unit 133 determines the number of times the collection method will be executed so as to maximize the total amount of debt recovered through the collection process. The expected value of the total amount of debt recovered through the collection process is obtained by multiplying the payment probability corresponding to the collection method by the outstanding debt balance of the debtor SA and the number of times the collection method will be executed, and then summing these values ​​for each debtor T. The collection method determination unit 133 allocates the total number of possible collection attempts to the number of executions of each collection method for each debtor T so as to maximize this total value.

[0064] The debt collection method determination unit 133 determines, for example, the number of times the debt collection method is executed for each debt collection target T shown in the debt collection method table in Figure 6, so that debt collection target T with a larger outstanding balance is executed more frequently. This increases the expected value of the total amount recovered. However, even for debt collection target T with a small outstanding balance, although the priority of debt collection is low, it is desirable that at least one debt collection is executed, as payment may be made with just one collection.

[0065] Therefore, the debt collection method determination unit 133 may determine the number of times each debt collection method is executed for each debt collection target T, such that the number of times each debt collection method is executed for all debt collection target T is at least once. This prevents debt collection from being completely ignored for debt collection target T with small outstanding debts, thereby suppressing debt collection losses. The debt collection method determination unit 133 may also generate a debt collection execution count table that shows the number of times each debt collection method is executed for each debt collection target T.

[0066] Figure 7 shows an example of a debt collection execution count table. In the debt collection execution count table, the debtor ID, the debt collection method used, and the number of times each debt collection method is executed are associated. As shown in Figure 7, debtor T T002, who has a relatively large outstanding debt, is assigned a higher number of debt collection executions. On the other hand, debtors T T001 and T003, who have relatively small outstanding debts, are assigned one debt collection execution. The debt collection method determination unit 133 stores the generated debt collection execution count table in the storage unit 12. As a result, the schedule generation unit 135 can generate a debt collection schedule by referring to the debt collection execution count table, although this will be described in detail later.

[0067] After the debt collection target determination unit 132 determines the debt collection target T, the debt collection timing determination unit 134 determines the debt collection timing for each debt collection target T based on the response probability for each debt collection target T output by the response probability prediction model for each debt collection target T. The debt collection timing determination unit 134 determines the timing at which the response probability output by the response probability prediction model, which has the attributes of the debt collection target T as input, is relatively high as the debt collection timing for each debt collection target T.

[0068] Incidentally, as mentioned above, the response probability prediction model can also output the response probability to reminder calls made to landlines and to mobile phones. Therefore, the response probability prediction model may output the response probability for each of the 16 combinations of reminder method (landline, mobile phone), date type (weekday, holiday), and time of day (four time slots per day). In this case, the reminder timing determination unit 134 may generate a reminder timing table that shows the response probability for each of these 16 combinations for each person T being reminded.

[0069] Figure 8 shows an example of a debt collection timing table. In the debt collection timing table, the debtor ID, the debt collection method, the date type, the time slot, and the response probability for 16 different combinations are associated.

[0070] The reminder timing determination unit 134 may determine one or more combinations with relatively high response probabilities from among the 16 possible combinations as reminder combinations, which are combinations of reminder means and reminder timing for the person to be reminded T. The number of one or more combinations is, for example, the number of reminder executions determined by the reminder target determination unit 132 for each person to be reminded T. In the example in Figure 8, for person T to be reminded T001, who has had one reminder execution, the reminder timing determination unit 134 determines the combination with the highest response probability (the combination corresponding to the response probability shown by halftone dots) as the reminder combination. Also, for person T to be reminded T002, who has had three reminder executions, the reminder timing determination unit 134 determines three combinations (the combination with the highest response probability, the second highest, and the third highest response probability, each corresponding to one of the three response probabilities shown by halftone dots) as reminder combinations.

[0071] The reminder timing determination unit 134 stores the generated reminder timing table in the storage unit 12. As a result, as will be described in detail later, the schedule generation unit 135 can generate a reminder schedule by referring to the reminder timing table.

[0072] The debt collection means determination unit 133 determines the debt collection means to be used and the number of times to be used for each person T to be subject to debt collection, and the debt collection timing determination unit 134 determines the debt collection timing for each person T to be subject to debt collection. After this, the schedule generation unit 135 generates a debt collection schedule. The schedule generation unit 135 generates a debt collection schedule by assigning debt collection tasks for the date and time corresponding to the debt collection timing determined by the debt collection timing determination unit 134 to debt collection personnel P who are available to work on the date and time corresponding to the debt collection timing determined by the debt collection timing determination unit 134. For example, the schedule generation unit 135 generates a debt collection schedule by assigning debt collection tasks for the debt collection timing determined by the debt collection timing determination unit 134 to the date and time on which debt collection personnel P, as indicated by the availability status table stored in the storage unit 12, are available to perform debt collection tasks.

[0073] The schedule generation unit 135 may generate a debt collection schedule by assigning debt collection tasks to debt collection officers P in order, starting with those for debtors T who tend to make payments quickly. For example, the schedule generation unit 135 generates a debt collection schedule by assigning debt collection tasks to debt collection officers P in order, starting with those for debtors T whose standard payment period is the period from the date of the debt collection notice issued by the debt collection officer P to the date of payment execution when the debtor T makes the payment. In this case, it is preferable that the standard debt collection date used is a debt collection date within the most recent three months. If there are multiple periods between the debt collection notice date and the payment execution date for a single debtor T, the schedule generation unit 135 may use the shortest of the multiple periods as the standard payment period for that debtor T, or it may use the average of the multiple periods as the standard payment period for that debtor T.

[0074] By doing this, compared to assigning debt collection tasks to debt collection officer P without prioritizing the debtors T, the timing of payments will be faster, making it possible to recover the debt earlier.

[0075] The schedule generation unit 135 may generate a debt collection schedule such that the number of debt collection attempts per person T per day is no more than three, in order to ensure compliance with laws and regulations regarding debt collection. This allows the debt collection officer P to conduct debt collection in compliance with laws and regulations, thereby preventing damage to the image of the financial institution collecting the debt. The process related to the generation of the debt collection schedule will be described in detail below.

[0076] Figure 9 is a flowchart showing the processing flow related to the generation of a debt collection schedule. First, the schedule generation unit 135 determines whether the target debtor SA, indicated by the debtor list data stored in the storage unit 12, belongs to the group that does not require debt collection, by referring to the debt collection classification table shown in Figure 5 stored in the storage unit 12 (S51). If the target debtor SA belongs to the group that does not require debt collection (S51: YES), debt collection is not required for the target debtor SA, and the process ends.

[0077] On the other hand, if the target debtor SA does not fall into the group that does not require debt collection (S51: NO), the schedule generation unit 135 determines whether the target debtor SA indicated in the debtor list data falls into the group that requires debt collection by referring to the debt collection classification table shown in Figure 5 stored in the storage unit 12 (S52). If the target debtor SA falls into the group that requires debt collection (S52: YES), the schedule generation unit 135 assigns an intermediate or beginner debt collection officer P to the target debtor SA (S53).

[0078] On the other hand, if the target debtor SA does not belong to the group for which debt collection is effective (S52: NO), then the target debtor SA belongs to the group for which debt collection is invalid. In this case, the schedule generation unit 135 refers to the debt collection execution count table shown in Figure 7, which is stored in the storage unit 12, and determines whether the means of debt collection for the target debtor SA belonging to the group for which debt collection is invalid is other than a manual phone call (S54). If the means of debt collection for the target debtor SA belonging to the group for which debt collection is invalid is other than a manual phone call (S54: YES), the schedule generation unit 135 assigns a debt collection officer P of intermediate or beginner level to the target debtor SA (S53).

[0079] On the other hand, if the means of debt collection for debtor SA belonging to the group of debt collection invalid is a manual phone call (S54:NO), a high level of verbal negotiation skills with debtor SA is required, so the schedule generation unit 135 assigns a skilled debt collection officer P to the debtor SA (S55). In this way, the valuable human resource of a skilled debt collection officer P can be effectively utilized.

[0080] Through the processes described so far, the type of debt collection officer P to be assigned to each of the target debtors SA (a junior level or intermediate level debt collection officer P, or a skilled level debt collection officer P) has been determined. Next, the schedule generation unit 135 assigns a debt collection timing with a relatively high response probability, as shown in the debt collection timing table shown in Figure 8, which is stored in the storage unit 12, to the debt collection means shown in the debt collection execution count table shown in Figure 7, which is stored in the storage unit 12 (S56).

[0081] For example, the schedule generation unit 135 assigns "Manual Call: Execution Count 1" for target debtor SA T001 in the debt collection execution count table shown in Figure 7 to "Debt Collection Method: Mobile Phone, Date Type: Holiday, Time Slot: 11:00-13:00" in the debt collection timing table shown in Figure 8, which has the highest response probability. Similarly, the schedule generation unit 135 assigns "Manual Call: Execution Count 3" for target debtor SA T002 in the debt collection execution count table shown in Figure 7 to "Debt Collection Method: Landline, Date Type: Weekday, Time Slot: 11:00-13:00", "Debt Collection Method: Landline, Date Type: Weekday, Time Slot: 17:00-20:00", and "Debt Collection Method: Mobile Phone, Date Type: Holiday, Time Slot: 13:00-17:00" in the debt collection timing table shown in Figure 8, which have relatively high response probabilities.

[0082] Finally, the schedule generation unit 135 assigns a collection date and time that the collection officer P, as indicated in the availability status table stored in the storage unit 12, can perform collection duties to the collection timing assigned in S56 (S57). For example, for collection timings for collection target T to which a junior or intermediate level collection officer P has been assigned in S53, the schedule generation unit 135 assigns a collection date and time that the junior or intermediate level collection officer P can perform collection duties. Similarly, for collection timings for collection target T to which an expert level collection officer P has been assigned in S55, the schedule generation unit 135 assigns a collection date and time that the expert level collection officer P can perform collection duties. In this way, the schedule generation unit 135 can generate a collection schedule table.

[0083] Figure 10 shows an example of a debt collection schedule table. In the debt collection schedule table, the debt collection officer ID, the debt collection target ID, the debt collection method, and the debt collection date and time are associated. The schedule generation unit 135 transmits the generated debt collection schedule table to the information terminal 2 used by the debt collection officer P. This allows the debt collection officer P to refer to the generated debt collection schedule and carry out debt collection.

[0084] [Effects of Information Processing Device 1] As explained above, the information processing device 1 identifies the effectiveness level of debt collection for each of the multiple target debtors SA based on the effectiveness data output by the classification model into which the attributes of the target debtors SA to be classified are input, and determines that some of the target debtors SA with relatively high debt collection effectiveness levels will be designated as debtors T. As a result, the number of debtors subject to debt collection is narrowed down, thereby improving the efficiency of debt collection operations.

[0085] Furthermore, the information processing device 1 generates a debt collection schedule that shows the schedule for debt collection by the debt collection officer P. By referring to the generated debt collection schedule, debt collection officer P can easily understand which debt collection method to use and at what timing to send a debt collection notice to which debt collection target T. As a result, the efficiency of debt collection work can be further improved.

[0086] <Variation> In the above embodiment, the effectiveness identification unit 131 identified the effectiveness of debt collection for each of the multiple target debtors SA based on the effectiveness data output by the classification model, which is a machine learning model. In the modified example, the effectiveness identification unit 131 identifies the effectiveness of debt collection for each of the multiple target debtors SA based on an effectiveness table instead of a machine learning model. In the modified example, the storage unit 12 stores an effectiveness table in which the attributes of the debtors are associated with effectiveness data indicating the effectiveness of debt collection efforts to encourage payment from the debtors.

[0087] Figure 11 shows an example of an effectiveness table. In the effectiveness table, the debtor's attributes are associated with effectiveness data indicating the effectiveness of the debt collection efforts. In Figure 11, "Not Required" represents the first effectiveness data, "Ineffective" represents the second effectiveness data, and "Effective" represents the third effectiveness data.

[0088] The effectiveness identification unit 131 identifies the effectiveness of debt collection for each of the multiple target debtors SA based on effectiveness data associated in the effectiveness table with the attributes of debtors that have a similarity of a predetermined threshold or higher to the attributes of the target debtor SA to be classified. For example, the effectiveness of debt collection for each of the multiple target debtors SA is determined such that the effectiveness of debt collection for target debtors SA whose effectiveness data is "effective" is higher than the effectiveness of debt collection for target debtors SA whose effectiveness data is "unnecessary" and for target debtors SA whose effectiveness data is "ineffective".

[0089] From this point onward, the processes executed by the debt collection target determination unit 132, the debt collection means determination unit 133, the debt collection timing determination unit 134, and the schedule generation unit 135 are the same as those described in the above embodiment, so their explanation will be omitted.

[0090] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of its gist. For example, all or part of the apparatus can be configured by functionally or physically distributing and integrating in any unit. Furthermore, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effects of the new embodiments resulting from the combinations are combined with the effects of the original embodiments. [Explanation of Symbols]

[0091] 1. Information Processing Device 11 Communications Department 12 Storage section 13 Control Unit 131 Effectiveness Identification Unit 132 Department for Determining Who Will Be Subject to Collection 133 Dunning Means Determination Department 134 Determination of Debt Collection Timing 135 Schedule Generation Unit 2 Information terminals S Information Processing System

Claims

1. A machine learning model trained using multiple sets of debtor attributes and effectiveness data indicating the effectiveness of reminders to encourage payment from the debtor as training data, comprising a storage unit that stores a classification model that outputs the effectiveness data when the debtor's attributes are input, An effectiveness determination unit that determines the effectiveness of debt collection for each of the multiple target debtors based on the effectiveness data output by the classification model into which the attributes of the target debtors to be classified are input, A debt collection target determination unit that determines, among multiple target debtors, some of the target debtors whose debt collection effectiveness is relatively high as targets for debt collection, An information processing device having

2. The classification model outputs, for each of the multiple target debtors, a first effectiveness data indicating that the probability of payment being made without reminder is above a first threshold, a second effectiveness data indicating that the probability of payment being made after reminder is below a second threshold (smaller than the first threshold), or a third effectiveness data indicating that the probability of payment being made without reminder is below the first threshold, and the probability of payment being made after reminder is above the second threshold. The information processing apparatus according to claim 1.

3. The debt collection target determination unit determines the target debtor corresponding to the third effectiveness data to be the debt collection target. The information processing apparatus according to claim 2.

4. The memory unit stores a debt collection ability value that indicates the debt collection officer's ability to perform debt collection duties. The debt collection target determination unit determines that the debtors whose outstanding debt balance is above a threshold among the target debtors corresponding to the second effectiveness data are the debt collection targets for whom the debt collection officer whose debt collection ability value meets predetermined conditions will perform the debt collection work. The information processing apparatus according to claim 2.

5. The debt collection target determination unit determines, among a plurality of target debtors, some of whom have a relatively high debt collection score calculated based on the effectiveness of the debt collection and the debtor's outstanding debt balance, to be the target debtors for debt collection. The information processing apparatus according to claim 1.

6. The memory unit further stores a payment prediction model that outputs the probability of payment for each collection method when the debtor's attributes or the effectiveness data are input. The information processing device further includes a debt collection means determination unit that determines the debt collection means to be used for each of the debt collection targets based on the attributes of the debt collection target or the output effectiveness data, which are output by the payment prediction model, and which have a relatively high probability of payment. The information processing apparatus according to claim 1.

7. The debt collection method determination unit allocates the total number of times debt collection can be performed to each of the debt collection methods for each of the debt collection targets, such that the sum of the values ​​obtained by multiplying the probability of payment corresponding to the debt collection method for each debtor by the debtor's outstanding debt and the number of times the debt collection method has been performed, multiplied by the number of debtors, maximizes the total number of times debt collection can be performed. The information processing apparatus according to claim 6.

8. The debt collection means determination unit determines that the automatic call is the debt collection means to be executed if the difference between the probability of payment made by manual calls by debt collection officers and the probability of payment made by automatic calls, as output by the payment prediction model, is less than a threshold. The information processing apparatus according to claim 6.

9. The memory unit, upon inputting the debtor's attributes, further stores a response probability prediction model that outputs the debtor's response probability to the debt collection notice for each timing of the notice. The information processing device further includes a debt collection timing determination unit that determines the timing at which the response probability output by the response probability prediction model, which has the attributes of the person to be debted as input, is relatively high, as the debt collection timing for each of the persons to be debted. The information processing apparatus according to claim 1.

10. The aforementioned storage unit stores the availability status of the debt collection officer for each date and time. The information processing device further includes a schedule generation unit that generates a debt collection schedule by assigning debt collection tasks to the debt collection personnel who are available to work at the date and time corresponding to the debt collection timing. The information processing apparatus according to claim 9.

11. The schedule generation unit generates the debt collection schedule by assigning debt collection tasks to the debt collection officer in order of priority, starting with those for whom the period from the date of the debt collection officer's debt collection request to the date of payment execution of the payment made by the debt collection officer is shortest. The information processing apparatus according to claim 10.

12. The memory unit, upon inputting the debtor's attributes, further stores a response probability prediction model that outputs the debtor's response probability to the collection method for each collection method. The information processing device further includes a debt collection means determination unit that determines the debt collection means for each of the debt collection targets based on the relatively high response probability output by the response probability prediction model, which has the attributes of the debt collection targets input. The information processing apparatus according to claim 1.

13. A storage unit that stores an effectiveness table in which the attributes of the debtor and effectiveness data indicating the effectiveness of the debt collection efforts to encourage payment from the debtor are associated, The effectiveness identification unit identifies the effectiveness of debt collection for each of the multiple target debtors based on the effectiveness data associated with the attributes of the debtors having a similarity of a predetermined threshold or higher to the attributes of the target debtors to be classified in the effectiveness table, A debt collection target determination unit that determines, among multiple target debtors, some of the target debtors whose debt collection effectiveness is relatively high as targets for debt collection, An information processing device having

14. A machine learning model trained using multiple sets of debtor attributes and effectiveness data indicating the effectiveness of reminders to encourage payment from the debtor as training data, wherein a computer having a memory unit that stores a classification model that outputs the effectiveness data when the debtor's attributes are input is executed. An effectiveness identification step, in which the effectiveness of debt collection is identified for each of the multiple target debtors based on the effectiveness data output by the classification model into which the attributes of the target debtors to be classified are input, A step of determining who to target for debt collection, in which a portion of the aforementioned target debtors, whose debt collection effectiveness is relatively high, are designated as targets for debt collection, An information processing method having

15. A machine learning model trained using multiple sets of debtor attributes and effectiveness data indicating the effectiveness of reminders to encourage payment from the debtor as training data, wherein the information processing device has a processor that stores a classification model that outputs the effectiveness data when the debtor's attributes are input, An effectiveness determination unit that determines the effectiveness of debt collection for each of the multiple target debtors based on the effectiveness data output by the classification model into which the attributes of the target debtors to be classified are input, A debt collection target determination unit that determines, among multiple target debtors, some of the target debtors whose debt collection effectiveness is relatively high as targets for debt collection, A program designed to function as such.

Images