Estimation system, estimation method, and program

Abstract

To streamline actions toward a target person.SOLUTION: An effect information acquisition unit (102) of an estimation system (1) acquires effect information related to effects obtained when a predetermined action is performed at predetermined timing using predetermined means for each of a plurality of target persons. A cost information acquisition unit (103) acquires cost information related to costs required when an action is performed at the timing using the means for each of the plurality of target persons. A constraint information acquisition unit (104) acquires constraint information related to cost constraints permitted for the action. An estimation unit (105) estimates a combination of the target person for which an action is to be performed and at least one of the timing and the means so that the efficient results can be obtained as a whole within the constraints on the basis of effect information, cost information, and constraint information for each of the plurality of target persons.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to an estimation system, an estimation method, and a program. [Background technology]

[0002] Conventionally, actions such as reminders have been taken against those who have defaulted on their credit card payments. For example, Patent Document 1 describes a system in which, based on a learning model prepared using a machine learning technique, a card company operator estimates, from among multiple individuals who have defaulted on their credit card payments, individuals who would be most likely to be reminded by telephone or other means, and the operator then makes the reminders to those individuals. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2023-047398 Summary of the Invention [Problem to be solved by the invention]

[0004] However, while the technology in Patent Document 1 can estimate the effectiveness of reminders for each of multiple subjects who are behind on their credit card payments, it cannot make an estimate that focuses on the overall efficiency of reminders, and this point is also true in other situations other than credit card payment reminders as in Patent Document 1. Conventional technology was not able to determine whether or not to take action against subjects while taking into account the overall efficiency.

[0005] One of the purposes of this disclosure is to streamline actions taken against targets. [Means for solving the problem]

[0006] The estimation system according to the present disclosure includes an effect information acquisition unit that acquires effect information regarding the effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects; a cost information acquisition unit that acquires cost information regarding the cost required when the action is performed by the means at the timing for each of the plurality of subjects; a constraint information acquisition unit that acquires constraint information regarding the cost constraints allowed for the action; and an estimation unit that estimates a combination of the subjects for whom the action is to be performed and at least one of the timing and the means, based on the effect information and cost information for each of the plurality of subjects and the constraint information, so that the effect is obtained efficiently overall within the scope of the constraints. [Effects of the Invention]

[0007] The present disclosure can streamline actions taken against targets. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of the estimation system. [Figure 2] FIG. 2 is a diagram illustrating an example of functions realized by the estimation system. [Figure 3] FIG. 10 is a diagram illustrating an example of a subject database. [Figure 4] FIG. 10 is a diagram illustrating an example of the relationship between input and output of a probability estimation model. [Figure 5] 10A to 10C are diagrams illustrating examples of processing by an effect information acquisition unit, a cost information acquisition unit, a constraint information acquisition unit, and an estimation unit. [Figure 6] FIG. 2 is a diagram illustrating an example of processing executed by the estimation system. [Figure 7] FIG. 10 is a diagram illustrating an example of a screen displayed on a staff member terminal. [Figure 8] FIG. 10 is a diagram illustrating an example of a function realized in a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0009] [1. Hardware configuration of the estimation system] An example of an embodiment of an estimation system, estimation method, and program according to the present disclosure will be described. FIG. 1 is a diagram illustrating an example of the hardware configuration of the estimation system. For example, the estimation system 1 includes a server 10, a person in charge terminal 20, and a subject terminal 30. Each of the server 10, the person in charge terminal 20, and the subject terminal 30 is connected to a network N such as the Internet, a LAN, or a public telephone line. While FIG. 1 illustrates one server 10, one person in charge terminal 20, and one subject terminal 30, there may be multiple servers 10, one person in charge terminal 20, and one subject terminal 30.

[0010] The server 10 is a server computer. For example, the server 10 includes a control unit 11, a storage unit 12, and a communication unit 13. The control unit 11 includes at least one processor. The storage unit 12 includes at least one of a volatile memory such as RAM and a non-volatile memory such as a flash memory. The communication unit 13 includes at least one of a communication interface for wired communication and a communication interface for wireless communication.

[0011] The staff terminal 20 is a terminal of the staff member, which will be described later. For example, the staff terminal 20 is a smartphone, a tablet, a mobile phone not classified as a smartphone, a landline phone, a personal computer, or a wearable terminal. The staff terminal 20 includes a control unit 21, a memory unit 22, a communication unit 23, an operation unit 24, and a display unit 25. The hardware configurations of the control unit 21, the memory unit 22, and the communication unit 23 may be similar to those of the control unit 11, the memory unit 12, and the communication unit 13, respectively. The operation unit 24 is an input device such as a touch panel or a mouse. The display unit 25 is a display such as a liquid crystal or organic electroluminescence (EL) display.

[0012] The subject terminal 30 is a terminal of the subject, which will be described later. For example, the subject terminal 30 is a smartphone, a tablet, a mobile phone not classified as a smartphone, a landline phone, a personal computer, or a wearable terminal. For example, the subject terminal 30 includes a control unit 31, a memory unit 32, a communication unit 33, an operation unit 34, and a display unit 35. The hardware configurations of the control unit 31, the memory unit 32, the communication unit 33, the operation unit 34, and the display unit 35 may be similar to those of the control unit 11, the memory unit 12, the communication unit 13, the operation unit 24, and the display unit 25, respectively.

[0013] The programs stored in the storage units 12, 22, 32 may be supplied to the server 10, the person in charge terminal 20, or the subject terminal 30 via the network N. Also, at least one of a reading unit (e.g., a memory card slot) that reads a computer-readable information storage medium and an input / output unit (e.g., a USB port) for inputting and outputting data to and from an external device may be included in the server 10, the person in charge terminal 20, or the subject terminal 30. For example, a program stored in an information storage medium may be supplied to the server 10, the person in charge terminal 20, or the subject terminal 30 via at least one of the reading unit and the input / output unit.

[0014] Furthermore, the estimation system 1 may include at least one computer. The computers included in the estimation system 1 are not limited to the example in FIG. 1. For example, the estimation system 1 may include only the server 10 and the person in charge terminal 20. In this case, the subject terminal 30 exists outside the estimation system 1. The estimation system 1 may include only the server 10. In this case, the person in charge terminal 20 and the subject terminal 30 exist outside the estimation system 1. For example, the estimation system 1 may include the server 10 and another computer not shown in FIG. 1.

[0015] [2. Overview of the estimation system] In this embodiment, a predetermined action is performed on each of a plurality of targets at a predetermined timing by a predetermined means. A target is a person who is the target of the predetermined action. In this embodiment, the predetermined action may not be performed on a target, so the target can also be said to be a candidate for the predetermined action. For example, a target is a person who uses a predetermined service. A predetermined service is a service for which some action may occur on the target.

[0016] In this embodiment, a case where a credit card service corresponds to the predetermined service is taken as an example. Therefore, the description of the credit card service can be read as the predetermined service. The predetermined service may be any service. The predetermined service is not limited to a credit card service. For example, the predetermined service may be an e-commerce service, a payment service, a travel reservation service, a communication service, a financial service, an online flea market service, or other services. The target person may not be using the predetermined service. The target person may be a potential customer of the predetermined service.

[0017] The predetermined timing is the timing at which a predetermined action is performed. The predetermined timing may be a pinpoint point in time, or may have a certain length of time (for example, one hour, one day, or one week). In this embodiment, an example is given in which the predetermined timing is indicated by a date, but the predetermined timing may be indicated by a date and time including not only the date but also the time. The predetermined timing may be indicated by only the time without including date information. The predetermined timing may be indicated by information other than the date and time. The predetermined timing may be indicated by temporal information. For example, the predetermined timing may be indicated by a day of the week, a weekday, a weekend, the beginning of the month, the end of the month, the beginning, the middle, the end of the month, or other information. There may be only one predetermined timing, or there may be multiple predetermined timings.

[0018] The predetermined means is a means used for a predetermined action. In other words, the predetermined means is a specific method for the predetermined action. There may be only one predetermined means, or there may be multiple predetermined means. For example, if contacting a target person corresponds to the predetermined action, the predetermined means is a contact means. The contact means may be various known means. For example, the contact means may be a human call, which is a call made by a human, a robot call, which is a call made by a robot, SMS (Short Message Service), email, SNS (Social Networking Service), a messaging app, chat, a notification function of an application such as a smartphone app, or other means.

[0019] It should be noted that the predetermined means is not limited to a contact means. The predetermined means may be any means corresponding to the predetermined action. For example, if the presentation of an advertisement corresponds to the predetermined action, the predetermined means is an advertising medium used for the advertisement. For example, the advertising medium may be a banner, push, pop-up, email, paper direct mail, public broadcasting, digital signage, or other medium. If the sales of a predetermined service corresponds to the predetermined action, the predetermined means is a sales method used in the sales. For example, the sales method may be the various contact means mentioned above, a direct visit, setting up a booth at an event venue, or other means. If the predetermined action is an act other than contact, presentation of an advertisement, or sales, the predetermined means may be any means used in the other act.

[0020] A predetermined action is an action taken toward a target person. The predetermined action may be an action taken for any purpose. In this embodiment, a case where contacting a target person corresponds to a predetermined action is taken as an example. The predetermined action may be any action. The predetermined action is not limited to contacting a target person. For example, the predetermined action may be presenting an advertisement or making a sale. The predetermined action may be an action other than contacting, presenting an advertisement, or making a sale. The person who takes the predetermined action may be any person. The person who takes the predetermined action may be someone other than the target person. For example, the person who takes the predetermined action may be an employee of a business that operates a predetermined service, an employee of another business that cooperates with the business, a part-time worker, or someone else.

[0021] In this embodiment, an example is given in which the estimation system 1 is used in a debt collection situation in a credit card service. For example, each of multiple subjects is a delinquent debtor who has fallen behind on his / her credit card payments. A delinquent debtor is someone who was unable to make a credit card payment by the due date due to insufficient funds in their bank account, etc. The specified means is a means of contacting a delinquent debtor to demand payment. In this embodiment, the description of the contact means can be read as "specified means." A demand is a demand for a payment that a subject has fallen behind on. Since the specified means is a means used for demand payment, it can also be called a demand means. Since the specified means is a means used for debt collection, it can also be called a collection means. The specified action is a demand for a delinquent debtor. In this embodiment, the description of demand payment can be read as "specified action."

[0022] In this embodiment, an example is given of a case where a person working for a card company that issued a credit card sends a reminder to a person. Furthermore, an example is given of a case where the person in charge can use three means of communication: a live call, a robot call, and SMS. The person in charge may or may not respond to the call from the person in charge. Just because the person in charge responds to the call from the person in charge does not necessarily mean that they will make the overdue payment. Conversely, the person in charge may make the overdue payment even if they do not respond to the call from the person in charge.

[0023] For example, an administrator who manages the responsible parties will plan which parties to contact, when, and by what means of communication to use in order to collect as much debt as possible. However, costs are incurred depending on the means of communication when the responsible party contacts the parties. In this embodiment, the costs are costs for the reminders. The costs are burdens incurred by the card company. For example, the costs may be monetary costs, time costs, the number of reminders, hardware resource costs, software resource costs, or other costs.

[0024] For example, if collection is carried out by a human call, the costs include labor costs and telephone charges. If collection is carried out by a robot call, the costs include telephone charges. If the card company uses a paid robot call service, the costs include the fee for using the robot call service. If collection is carried out by SMS, the costs include the fee for using the SMS. The costs that a card company can afford to incur vary depending on its budget, human resources, hardware resources, software resources, and other factors. Collection staff need to collect debts efficiently within cost constraints.

[0025] However, even if the administrator tries to plan which recipients to contact, when, and by what means, there are an infinite number of combinations. Hereinafter, these combinations will be referred to as "collection combinations." Because there are an infinite number of collection combinations, it is extremely difficult for the administrator to create an efficient plan. Even if debt collection from a specific recipient is successful, if debt collection is not streamlined across the entire credit card service, the card company will incur losses by having to compensate the recipients for the overdue amounts.

[0026] Therefore, the estimation system 1 of this embodiment estimates a combination of reminders so that debt collection can be made more efficient across the entire credit card service, within the cost constraints allowed by the card company. The person in charge makes reminders to the recipient based on the reminder combination estimated by the estimation system 1. This enables the estimation system 1 to make reminders to the recipient more efficient. Details of the estimation system 1 will be explained below.

[0027] [3. Functions realized by the estimation system] FIG. 2 is a diagram illustrating an example of functions realized by the estimation system 1. In this embodiment, an example is taken of a case where the main functions of the estimation system 1 are realized by a server 10. For example, the server 10 includes a data storage unit 100, a probability estimation model storage unit 101, an effect information acquisition unit 102, a cost information acquisition unit 103, a constraint information acquisition unit 104, and an estimation unit 105. Each of the data storage unit 100 and the probability estimation model storage unit 101 is realized by the storage unit 12. Each of the effect information acquisition unit 102, the cost information acquisition unit 103, the constraint information acquisition unit 104, and the estimation unit 105 is realized by the control unit 11.

[0028] [Data storage section] The data storage unit 100 stores data necessary for estimating a reminder combination. For example, the data storage unit 100 stores a target person database DB.

[0029] FIG. 3 is a diagram showing an example of a subject database DB. The subject database DB is a database in which various data related to subjects is stored. For example, the subject database DB stores subject IDs, credit card numbers, personal information, demographic information, delinquent payment information, and usage information. Any data may be stored in the subject database DB. The data stored in the subject database DB is not limited to the example of FIG. 3. For example, the subject database DB may store data on users who are not delinquent in payments (i.e., users who do not fall under the category of subjects).

[0030] The subject ID is an example of subject identification information that can identify the subject. Therefore, the words "subject ID" can be read as "subject identification information." The subject identification information may be information other than the subject ID. The subject identification information may be information that can identify the subject in some way. For example, the subject identification information may be the subject's email address, telephone number, or other information. A credit card number may be used as subject identification information. The subject ID may also be used as a login account for a credit card service.

[0031] The personal information stored in the subject database DB is personal information about the subject. For example, the personal information indicates the subject's name, address, telephone number, email address, or a combination of these. Demographic information is information about the characteristics of the subject. Personal information may correspond to demographic information, and demographic information may correspond to personal information. For example, demographic information indicates the subject's gender, age or age group, occupation, annual income, family composition, or a combination of these. Each of the personal information and demographic information may be publicly known information.

[0032] Delinquent payment information is information relating to a payment that the subject has defaulted on. For example, the delinquent payment information is the amount of the delinquent payment, a breakdown of the payment that the subject has defaulted on, the withdrawal date of the payment that the subject has defaulted on, or a combination of these. The delinquent amount is the total amount of the payment that the subject has defaulted on. The breakdown of the payment that the subject has defaulted on is the settlement details indicated by the payment. For example, the breakdown of the payment that the subject has defaulted on is the settlement date of the credit card used by the subject, the payment amount, the payment location (e.g., a store, etc.), or a combination of these. When a subject defaults on a payment, the server 10 updates the delinquent payment information associated with the subject's subject ID.

[0033] The usage status information is information about the subject's credit card usage. The usage status information may also include information about payments that the subject has not defaulted on. The usage status information may be a credit card statement. For example, the usage status information indicates the payment date, payment amount, payment location (e.g., store, etc.) of the credit card used by the subject, or a combination thereof. The usage status information may indicate installment payments, bonus payments, or other usage status of the credit card. The usage status information may also indicate the subject's payment amount for each month. When a subject uses a credit card service, the server 10 updates the usage status information associated with the subject's subject ID.

[0034] The data stored in the data storage unit 100 is not limited to the target database DB. The data storage unit 100 may store any data necessary for estimating a reminder combination. For example, the data storage unit 100 may store a training database necessary for learning a probability estimation model M1, which will be described later. The data storage unit 100 may also store a management tool for card company personnel.

[0035] [Probability estimation model memory section] The probability estimation model storage unit 101 stores a probability estimation model M1 that has learned the relationship between subject characteristic information regarding the characteristics of a training subject and the probability that an effect will be obtained from the training subject. The training subject may be an actual subject or a fictitious subject. The training subject may also be a subject of a credit card service other than the credit card service for which debt collection is planned.

[0036] The subject characteristic information is information relating to the characteristics of the subject. The subject characteristic information indicates characteristics that are correlated with the probability of achieving results when dunning is carried out. The probability estimation model M1 learns this correlation. The subject characteristic information may also include information that has no particular correlation. For example, the subject characteristic information may be the subject's credit card number, personal information, demographic information, delinquency information, usage information, past dunning history, or a combination of these. The subject characteristic information may also include past debt collection history, credit-related information, or other information. In this embodiment, an example is given in which the subject characteristic information is demographic information.

[0037] The effect is the success of the dunning. In other words, the effect is the achievement of the purpose of the dunning. For example, the effect is that a debt is collected through the dunning. The effect may be the amount collected through the dunning. The effect may differ depending on the situation in which the estimation system 1 is used. The effect may be contact with the subject, rather than collecting the debt. For example, if the predetermined action is contact for a purpose other than dunning (e.g., a telephone appointment), the effect is that the subject is contacted. If the predetermined action is presenting an advertisement, the effect is that the subject views the advertisement, selects the advertisement, or purchases the advertised product or service. If the predetermined action is sales, the effect is that the sales is successful. For example, the effect is that a contract is concluded.

[0038] The probability estimation model M1 is a model that estimates the probability of achieving an effect. For example, the probability estimation model M1 is a model that estimates the probability that a target person will make a payment when a reminder is sent at a certain time and by a certain means. The probability estimation model M1 is a model created using a machine learning technique. Various known machine learning techniques can be used. The machine learning technique may be any of supervised learning, semi-supervised learning, and unsupervised learning. For example, the probability estimation model M1 may be a neural network, a support vector machine, a large-scale language model, or other models. There are various theories about the meaning of the term machine learning, but the machine learning in this embodiment includes various known meanings. For example, the machine learning in this embodiment also includes AI (artificial intelligence) and deep learning. The meaning of these machine learning also applies to other models other than the probability estimation model M1 (for example, the initial value estimation model M2 described below).

[0039] For example, the probability estimation model storage unit 101 stores a trained probability estimation model M1 that has learned the relationship between subject characteristic information and the probability of obtaining an effect. In this embodiment, the server 10 trains the probability estimation model M1, but a computer other than the server 10 may also train the probability estimation model M1. For example, the staff terminal 20 may train the probability estimation model M1. The probability estimation model M1 includes a program that performs a series of processes, such as calculating embedded representations, and parameters referenced by the program. The parameters of the probability estimation model M1 are adjusted through training. The program and parameters of the probability estimation model M1 may be similar to known programs and parameters. For example, the program of the probability estimation model M1 may include multiple layers, such as an input layer that accepts data input, an intermediate layer that calculates embedded representations, and an output layer that outputs estimation results. The parameters of the probability estimation model M1 may be weights and biases.

[0040] FIG. 4 is a diagram showing an example of the relationship between the input and output of the probability estimation model M1. The data formats of the input and output of the probability estimation model M1 are basically the same during estimation and learning. The output numerical value in FIG. 4 is the probability of obtaining an effect. In this embodiment, an example is taken of a case where a probability estimation model M1 is prepared for each contact means. For example, if three contact means are prepared, such as a person call, a robot call, and SMS, three probability estimation models M1 are prepared, such as a probability estimation model M1 for person calls, a probability estimation model M1 for robot calls, and a probability estimation model M1 for SMS, as shown in FIG. 4.

[0041] For example, the data storage unit 100 stores a training database in which training data necessary for training the probability estimation model M1 is stored. The training data includes an input portion that is input to the probability estimation model M1 during training and an output portion that becomes the correct answer during training. The input portion of the training data has basically the same format as the input data that is input to the probability estimation model M1 during estimation. The output portion of the training data has basically the same format as the output data that is output from the probability estimation model M1 during estimation. Note that the input portion of the training data may have a slightly different format from the input data that is input to the probability estimation model M1 during estimation. Similarly, the output portion of the training data may have a slightly different format from the output data that is output from the probability estimation model M1 during estimation.

[0042] For example, the input portion of the training data includes subject characteristic information for training. In this embodiment, the input portion of the training data also includes the timing at which a prompt is issued to the training subject. The output portion of the training data indicates the probability that an effect will be obtained from a subject with characteristics indicated by the subject characteristic information, which is the input portion of the training data (the probability of obtaining a correct answer during learning). The training data may be created by an administrator of the estimation system 1 or may be created using a known tool. The server 10 trains the probability estimation model M1 so that the output portion of the training data is output when the input portion of the training data is input. The server 10 trains the probability estimation model M1 by adjusting the parameters of the probability estimation model M1 based on the training data.

[0043] The algorithm for training the probability estimation model M1 may be a known algorithm used in the field of machine learning. For example, the server 10 may train the probability estimation model M1 on training data based on an algorithm such as gradient descent or backpropagation. The loss function used during training may also be a known loss function. Based on the loss function, the server 10 calculates a loss, which is the error between the output portion of the training data and the output from the probability estimation model M1 during training. The server 10 completes the training when the loss becomes small to a certain extent. Upon completing the training, the server 10 records the trained probability estimation model M1 in the probability estimation model storage unit 101.

[0044] As shown in FIG. 4 , when a probability estimation model M1 is prepared for each contact means, the probability estimation model M1 for a certain contact means learns the relationship between the subject characteristic information of the training subject for that contact means and the probability obtained from that training subject. Therefore, training data is prepared for each contact means on which the probability estimation model M1 for that contact means learns. The data format of the training data learned by the probability estimation model M1 for each contact means may be the same as the data format of the training data learned by the probability estimation model M1 for the other contact means. In this embodiment, three probability estimation models M1 are prepared: a probability estimation model M1 for human calls, a probability estimation model M1 for robot calls, and a probability estimation model M1 for SMS. Therefore, the server 10 trains the probability estimation model M1 for that contact means based on the training data for each of these three contact means.

[0045] [Effect information acquisition section] The effect information acquisition unit 102 acquires effect information relating to the effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects. In this embodiment, the predetermined means is a contact means for reminding, and the predetermined action is reminding, so the effect information acquisition unit 102 acquires effect information relating to the effect obtained when a reminder is performed by a predetermined contact means at a predetermined timing for each of a plurality of subjects.

[0046] The effect information is information somehow related to the effect obtained from the subject. The effect information may be a binary value indicating whether or not there is an effect. However, in this embodiment, the effect information is not a binary value, but rather an example in which the effect expected from the subject is indicated by three or more levels of numerical values. For example, the effect information acquisition unit 102 acquires effect information for each of multiple subjects based on the probability of obtaining an effect when a reminder is sent to each of the subjects at a predetermined timing via a predetermined communication means, and the degree of effect obtained from the subject. Note that the effect information may be expressed in other formats, such as characters or symbols, instead of numerical values.

[0047] The degree of effect is the magnitude of the effect. In this embodiment, the effect of debt collection is large when debt collection is successful from a subject with a large amount of delinquent debt, so an example is given in which the amount of delinquent debt corresponds to the degree of effect. The degree of effect may differ depending on the situation in which the estimation system 1 is used. For example, if the specified action is contact for a purpose other than reminder (e.g., telephone appointment), the effect is the sales or profit obtained from the subject through the contact. If the specified action is presentation of an advertisement, the advertisement is the price of the product or service that is the subject of the advertisement. If the specified action is sales conducted, the effect is the sales or profit obtained through the sales.

[0048] FIG. 5 is a diagram illustrating an example of the processing of each of the effect information acquisition unit 102, the cost information acquisition unit 103, the constraint information acquisition unit 104, and the estimation unit 105. In this embodiment, the effect information acquisition unit 102 acquires a probability that an effect will be obtained from each of multiple subjects based on the effect information of the subject and the probability estimation model M1, and acquires effect information based on the probability for each of the multiple subjects. For example, the effect information acquisition unit 102 acquires subject characteristic information for each of the multiple subjects. In this embodiment, a case is exemplified in which demographic information is used as the subject characteristic information, so the effect information acquisition unit 102 acquires demographic information for each of the multiple subjects from the subject database DB as the subject characteristic information of the subject. The effect information acquisition unit 102 may acquire the subject characteristic information for each of the multiple subjects from a database other than the subject database DB, a computer other than the server 10, or an external information storage medium.

[0049] In this embodiment, it is assumed that a planning period for which debt collection is planned is determined in advance. The effect information acquisition unit 102 identifies multiple timings within the planning period. In the example of FIG. 5, it is assumed that the planning period is from February 5, 2024 to February 9, 2024. Therefore, the effect information acquisition unit 102 identifies five days within the planning period as multiple timings. The planning period may be of any length. For example, the planning period may be several days, one week, two weeks, or one month. It is assumed that data indicating the planning period is stored in the data storage unit 100. The planning period may be specified by a person in charge or a manager.

[0050] For example, the effect information acquisition unit 102 inputs input data including target person characteristic information and timing within the planned period to the probability estimation model M1 for each target person. In the example of FIG. 5, there are five timings, and therefore five pieces of input data for each target person. In this embodiment, a probability estimation model M1 is prepared for each contact means, and therefore the effect information acquisition unit 102 inputs input data to each of the multiple probability estimation models M1. In the example of FIG. 5, the effect information acquisition unit 102 inputs input data for each target person to the probability estimation model M1 for human calls, the probability estimation model M1 for robot calls, and the probability estimation model M1 for SMS.

[0051] For example, when input data is input, the probability estimation model M1 calculates an embedding representation of the input data based on parameters adjusted by learning. The embedding representation is information indicating the characteristics of the input data. The embedding representation is also called a feature. The embedding representation may be in any format, such as a vector format, an array format, a matrix format, multiple numerical values, a single numerical value, or other format. If the probability estimation model M1 is a large-scale language model, the probability estimation model M1 divides the input data into tokens and then calculates an embedding representation of each token. The probability estimation model M1 outputs a probability according to the embedding representation calculated from the input data.

[0052] In this embodiment, a probability estimation model M1 is prepared for each contact means, and the effect information acquisition unit 102 acquires a probability for each contact means based on the probability estimation model M1 for that contact means. The processing performed by each probability estimation model M1 may be similar. The parameters of each probability estimation model M1 become different values ​​through learning, so even if the input data is the same, different output data is acquired. In the example of FIG. 5, the effect information acquisition unit 102 inputs the input data of each target person to each of the probability estimation model M1 for human calls, the probability estimation model M1 for robot calls, and the probability estimation model M1 for SMS, and acquires the probability indicated by the output data output from each of these three probability estimation models M1.

[0053] For example, the effectiveness information acquisition unit 102 may calculate, for each target person, the probability that an effect will be achieved if reminders are sent to that target person via each of multiple contact means, and calculate an expected recovery amount, which is the value obtained by multiplying the delinquent amount, which indicates the degree of effect, by the probability. The expected recovery amount is an example of effectiveness information. The expected recovery amount can also be said to be the degree to which an effect is expected. In the example of Figure 5, the delinquent amount of target person A is 100,000 yen. The effectiveness information acquisition unit 102 calculates the expected recovery amount by multiplying the delinquent amount of target person A, 100,000 yen, by the probability that an effect will be achieved if reminders are sent via each of the three contact means. The delinquent amount of target person B is 200,000 yen. The effectiveness information acquisition unit 102 calculates the expected recovery amount by multiplying the delinquent amount of target person B, 200,000 yen, by the probability that an effect will be achieved if reminders are sent via each of the three contact means.

[0054] The effect information acquired by the effect information acquisition unit 102 is not limited to the example of this embodiment. For example, the effect information acquisition unit 102 may calculate an expected recovery amount by multiplying the probability estimated by the probability estimation model M1 by a weighting coefficient and then multiplying the result by the delinquent amount, and acquire the result as effect information. The effect information acquisition unit 102 may calculate an expected recovery amount by multiplying the probability estimated by the probability estimation model M1 by the delinquent amount multiplied by a weighting coefficient, and acquire the result as effect information. The effect information acquisition unit 102 may acquire a probability instead of an expected recovery amount as effect information. The effect information acquisition unit 102 may acquire the delinquent amount instead of the expected recovery amount as effect information. In this case, the effect information acquisition unit 102 does not need to calculate a probability. The server 10 does not need to include the probability estimation model storage unit 101.

[0055] [Cost Information Acquisition Department] The cost information acquiring unit 103 acquires cost information regarding the costs required when a reminder is sent to each of a plurality of target persons at a predetermined timing via a predetermined communication means. The cost information is information indicating the degree of cost. The cost information may indicate monetary costs, time costs, hardware or software resource costs, or other costs. For example, the cost information indicates a numerical value of the cost. The cost information may be expressed not only by numerical values ​​but also by characters or symbols. In this embodiment, an example is given in which the cost information is stored in the data storage unit 100. Therefore, the cost information acquiring unit 103 acquires the cost information from the data storage unit 100. The cost information may be stored in a computer other than the server 10 or in an external information storage medium. In this case, the cost information acquiring unit 103 may acquire the cost information from the other computer or the external information storage medium.

[0056] In this embodiment, the cost differs depending on the contact means. Furthermore, the cost of each contact means is assumed to be a fixed value. In other words, an example will be given in which the cost of a contact means is predetermined and does not vary depending on factors other than the contact means. A mode in which the cost varies depending on other factors will be described in a modified example below. In this embodiment, the data storage unit 100 stores cost information for each of multiple contact means. The cost information acquisition unit 103 acquires cost information for each of the multiple contact means from the data storage unit 100. If there is only one contact means, the data storage unit 100 stores cost information for that one contact means. The cost information acquisition unit 103 only needs to acquire cost information for that one contact means. Even if there are multiple contact means, the cost information may be common to the multiple contact means.

[0057] In the example of FIG. 5, three contact methods are available: a human call, a robot call, and SMS, so the cost information acquisition unit 103 acquires cost information for a human call, cost information for a robot call, and cost information for SMS. The cost information for a human call indicates the cost when a human call is made. For example, the cost information for a human call indicates the total labor costs and telephone charges for a human call. The cost information for a robot call indicates the cost when a robot call is made. For example, the cost information for a robot call indicates the telephone charges for a robot call. The cost information for SMS indicates the cost when SMS is made. For example, the cost information for SMS indicates the usage fee for SMS. The cost information may indicate the cost per contact using these contact methods, or may indicate the total cost of multiple contacts.

[0058] [Constraint information acquisition part] The constraint information acquisition unit 104 acquires constraint information regarding the cost constraints allowed for collection. The cost constraint is an upper limit of the cost. The cost constraint can also be referred to as an allowable range of the cost. The cost constraint may be a constraint at an individual timing, or may be a constraint for the entire planning period for which debt collection is planned. In this embodiment, an example is given in which the constraint information is stored in the data storage unit 100. Therefore, the constraint information acquisition unit 104 acquires the constraint information from the data storage unit 100. The constraint information may be stored in a computer other than the server 10 or in an external information storage medium. In this case, the constraint information acquisition unit 104 may acquire the constraint information from the other computer or the external information storage medium.

[0059] In this embodiment, the constraints differ depending on the contact means. Furthermore, the constraints for each contact means are assumed to be fixed values. In other words, an example will be given in which the constraints do not change depending on factors other than the contact means, and there is a one-to-one relationship between the contact means and the constraints. A mode in which the constraints change depending on other factors will be described in a modified example below. In this embodiment, the data storage unit 100 stores constraint information for each of multiple contact means. The constraint information acquisition unit 104 acquires constraint information in which constraints are defined according to the contact means. In other words, the constraint information acquisition unit 104 acquires constraint information for each of the multiple contact means from the data storage unit 100. If there is only one contact means, the data storage unit 100 stores the constraint information for that one contact means. The constraint information acquisition unit 104 only needs to acquire the constraint information for that one contact means. Even if there are multiple contact means, the constraint information may be common to the multiple contact means.

[0060] In the example of FIG. 5, three contact methods are provided: a human call, a robot call, and SMS. Therefore, the restriction information acquisition unit 104 acquires restriction information for human calls, restriction information for robot calls, and restriction information for SMS. The restriction information for human calls indicates restrictions when a human call is made. For example, the restriction information for human calls indicates the allowable amount for labor costs and telephone charges for human calls. The restriction information for human calls may indicate an upper limit on the number of calls allowed for human calls (for example, an upper limit on the resources of hardware or software used for human calls). The restriction information for robot calls indicates restrictions when a robot call is made. For example, the restriction information for robot calls indicates the allowable amount for telephone charges for robot calls. The restriction information for robot calls may indicate an upper limit on the number of calls allowed for robot calls (for example, an upper limit on the resources of hardware or software used for robot calls). The restriction information for SMS indicates restrictions when SMS is made. For example, the restriction information for SMS indicates the allowable amount for SMS usage fees. The SMS restriction information may indicate an upper limit on the number of messages allowed in an SMS.

[0061] [Estimation part] The estimation unit 105 estimates a combination of targets to be reminded, a predetermined timing, and at least one of a predetermined contact means, based on the effect information and cost information for each of the multiple targets and the constraint information, so that an overall efficient effect is obtained within the constraints. In this embodiment, a case will be described in which the estimation unit 105 estimates a combination of targets, predetermined timing, and predetermined contact means, but the estimation unit 105 may estimate a combination of targets and predetermined timing without estimating the predetermined contact means. The estimation unit 105 may estimate a combination of targets and predetermined contact means without estimating the predetermined timing. As described above, in this embodiment, the combination is a reminder combination.

[0062] The range of the constraint means that the cost is less than or equal to the cost indicated by the constraint. In other words, the range of the constraint means that the cost does not exceed the cost indicated by the constraint. When the total cost according to the combination of reminders estimated by the estimation unit 105 is less than or equal to the cost indicated by the constraint, this corresponds to being within the range of the constraint. The whole refers to the entire group of multiple targets. In other words, the whole refers to the entire credit card service. Achieving an efficient effect means that the overall effect is relatively high. In other words, having a high effect relative to the cost required for reminders corresponds to achieving an efficient effect. In other words, achieving an efficient effect means that the cost performance of reminders is good. Achieving an efficient effect means that the cost-effectiveness is high.

[0063] In the example of Figure 5, a variable value of 1 corresponds to a reminder being sent. A variable value of 0 corresponds to a reminder not being sent. For example, there is a variable for each combination of target person, timing, and contact method. If the number of target people is x (x is a natural number), the number of timings is y (y is a natural number), and the number of contact methods is z (z is a natural number), then there are x × y × z variables. In the example of Figure 5, the timings are assumed to be five days from February 5, 2024 to February 9, 2024. There are three contact methods: human call, robot call, and SMS. Therefore, there are 5 × 3 = 15 variables per target person. If there are 1,000 target people, there are 1,000 × 15 = 15,000 variables. In the example of Figure 5, two variables are set to 1. For example, the first variable set to 1 indicates that a robot call will be sent to target person A on February 5, 2024. The second variable, which is set to 1, indicates that an SMS will be sent to subject B on February 5, 2024.

[0064] For example, the estimation unit 105 determines at least one combination of candidate combinations for which a reminder will be issued as a combination for which a reminder will actually be issued. In the example of FIG. 5, the estimation unit 105 determines combinations of variables of 0 and 1. In the example described above, the estimation unit 105 determines 15,000 combinations of variables of 0 and 1. A reminder combination with a variable of 0 is not adopted, so no cost is incurred, but a reminder combination with a variable of 1 is adopted, so a cost is incurred. The estimation unit 105 estimates a reminder combination so that the total cost value of a reminder combination with a variable of 1 is within the constraint range, and the total expected collection efficiency value of a reminder combination with a variable of 1 is relatively higher than other combinations.

[0065] For example, the estimation unit 105 estimates a combination for which a reminder is to be performed based on a metaheuristic method. A metaheuristic method can also be referred to as a method for solving a combinatorial optimization problem. Combinatorial optimization does not need to identify an exact solution, as long as a solution that is appropriate to a certain extent is identified. For example, the metaheuristic method may be a genetic algorithm, particle swarm optimization, ant colony algorithm, or simulated annealing. In this embodiment, an example is given in which the estimation unit 105 determines a combination for which a reminder is to be performed based on a genetic algorithm, but the estimation unit 105 may estimate a combination for which a reminder is to be performed based on another metaheuristic method.

[0066] In this embodiment, the estimation unit 105 calculates an index related to the efficiency of the effect obtained from each of a plurality of subjects based on the effect information and cost information of the respective subjects, and estimates a combination of reminders based on the index for each of the plurality of subjects. In the example of FIG. 5, the estimation unit 105 calculates the expected collection efficiency for each subject based on the expected collection amount, which is an example of the effect information for the subject, and the expected cost, which is an example of the cost information for the subject. For example, the estimation unit 105 calculates the expected collection efficiency for each subject by dividing the expected collection amount for the subject by the expected cost for the subject. The estimation unit 105 calculates the expected collection efficiency for each reminder combination.

[0067] The expected recovery efficiency is an example of an efficiency index. An efficiency index can be said to be the degree of effect per unit cost. An efficiency index may be an index calculated by substituting effect information and cost information into a calculation formula. An efficiency index is not limited to the expected recovery efficiency. For example, the estimation unit 105 may multiply the expected recovery amount or the expected cost by a coefficient appropriate for the subject and then divide the result, rather than simply dividing the expected recovery amount for each subject by the expected cost for the subject.

[0068] In this embodiment, constraints are set according to the contact means, and the estimation unit 105 estimates a combination of reminders within the scope of the constraints according to the contact means. For example, for each contact means, the estimation unit 105 may estimate a combination of a target person to whom an action will be taken and at least one of timing and contact means, so that an overall efficient effect is obtained within the scope of the constraints according to the contact means. For each contact means, the estimation unit 105 estimates a reminder combination such that the total cost of a reminder combination with a variable of 1 is within the scope of the constraints of the contact means, and the total expected collection efficiency of a reminder combination with a variable of 1 is relatively high. In this case, too, the estimation unit 105 may estimate a reminder combination based on the metaheuristic method described above.

[0069] The estimation unit 105 may estimate a combination of reminders so that there is only one combination of reminders for a given target person in which the variable is 1. If the total cost of reminder combinations for all targets or a certain number of targets in which the variable is 1 is within the constraint range and there is room up to the upper limit of the constraint, the estimation unit 105 may estimate a combination of reminders so that there are two or more combinations of reminders for a given target person in which the variable is 1. The estimation unit 105 may estimate a combination of reminders that is efficient and effective overall, based on a method other than the metaheuristic method (for example, a machine learning method).

[0070] [4. Processing performed by the estimation system] Fig. 6 is a diagram showing an example of processing executed by the estimation system 1. In the present embodiment, a case is taken as an example in which the main processing of the estimation system 1 is executed by the server 10. The processing of Fig. 6 is executed by the control unit 11 executing a program stored in the storage unit 12. It is assumed that learning of the probability estimation model M1 is completed before the processing of Fig. 6 is executed.

[0071] 6, the server 10 acquires subject characteristic information for each of a plurality of subjects based on the subject database DB (S1). For each subject, the server 10 calculates the probability that an effect will be obtained from the subject when a reminder is sent to the subject at a given time by a given contact means based on the subject characteristic information of the subject, each of a plurality of timings, and a probability estimation model M1 for each of a plurality of contact means (S2).

[0072] For each subject, the server 10 calculates the expected recovery amount for that subject based on the delinquent amount indicated by the delinquent amount stored in the subject database DB and the subject's probability calculated in S2, and acquires this as effectiveness information for that subject (S3). For each contact method, the server 10 acquires cost information indicating the expected cost of that contact method stored in the storage unit 12 (S4). For each subject, the server 10 calculates the expected recovery efficiency based on the subject's effectiveness information acquired in S3 and the cost information acquired in S4, and acquires this as an index related to efficiency (S5). The calculation of information other than the variables shown in FIG. 5 is completed by the processes of S1 to S5.

[0073] For each contact method, the server 10 acquires the constraint condition information for that contact method stored in the storage unit 12 (S6). Based on the indicators acquired in S5 and the constraint condition information acquired in S6, the server 10 estimates a combination of reminders that will achieve an efficient overall effect within the constraints of each of the multiple contact methods (S7). In S7, the server 10 estimates a reminder combination based on the metaheuristic method described above such that the total cost of the reminder combination with a variable of 1 is within the constraints of the contact method, and the total expected collection efficiency of the reminder combination with a variable of 1 is relatively higher than other combinations. The server 10 transmits reminder combination data indicating the reminder combination estimated in S8 to the responsible party terminal 20 (S8), and this process ends.

[0074] FIG. 7 is a diagram showing an example of a screen displayed on the terminal 20 of the person in charge. For example, when the terminal 20 of the person in charge receives the dunning combination data sent by the server 10 in S8, it displays the screen SC of FIG. 7 on the display unit 25 based on the dunning combination data. The combination data may show not only the combination of dunning requests for which the variable becomes 1, but also the total value of the expected collection efficiency obtained by the dunning requests. The total value is calculated by the server 10. The person in charge can understand their own debt collection plan by checking the screen SC. The person in charge may modify the plan displayed on the screen SC by operating the operation unit 24.

[0075] [5. Summary of embodiments] The estimation system 1 of this embodiment acquires effect information for each of multiple subjects. The estimation system 1 acquires cost information for each of multiple subjects. The estimation system 1 acquires constraint information. Based on the effect information and cost information for each of multiple subjects and the constraint information, the estimation system 1 estimates a combination of subjects to be reminded, timing, and / or contact method, so that an overall efficient effect is achieved within the constraints. This allows the estimation system 1 to estimate a combination of reminders that satisfies cost constraints and achieves an overall efficient effect, thereby making reminders for subjects more efficient. For example, the estimation system 1 can achieve high cost-performance (cost-effectiveness) reminders overall, rather than only successfully collecting payments from certain subjects. The estimation system eliminates the need for personnel and managers to plan reminders, thereby increasing convenience for personnel and managers.

[0076] Furthermore, the estimation system 1 acquires effect information for each of a plurality of subjects based on the probability of obtaining an effect when a reminder is sent to each of the subjects at a predetermined timing via a predetermined means of communication, and the degree of effect obtained from the subject. This allows the estimation system 1 to estimate a reminder combination based on effect information that comprehensively takes into account the probability of obtaining an effect and the degree of effect, thereby improving the accuracy of the estimation of the reminder combination. As a result, the estimation system 1 can make reminders for subjects more efficient.

[0077] Furthermore, the estimation system 1 calculates an index related to the efficiency of the effects obtained from each of the multiple subjects based on the effect information and cost information of the multiple subjects, and estimates a combination of reminders based on the index for each of the multiple subjects. This allows the estimation system 1 to estimate a combination of reminders based on the index related to the efficiency of the effects obtained from each of the multiple subjects, thereby improving the accuracy of estimating the combination of reminders. For example, the index allows the estimation system 1 to easily apply metaheuristic techniques. As a result, the estimation system 1 can make reminders for the subjects more efficient.

[0078] Furthermore, the estimation system 1 acquires constraint information that defines constraints according to the means of communication. The estimation system 1 estimates a combination of reminders within the scope of the constraints according to the means of communication. As a result, even if the constraints differ for each means of communication, the estimation system 1 can estimate a combination of reminders within a feasible scope by estimating a combination that will be effective overall and efficiently within the scope of the constraints of each of the multiple means of communication. Since the person in charge can make reminders that efficiently combine multiple means of communication, the estimation system 1 can make reminders to the target person more efficient.

[0079] Furthermore, the estimation system 1 acquires the probability that an effect will be obtained from each of the multiple subjects based on the effect information of each of the multiple subjects and the probability estimation model M1, and acquires effect information based on the probability for each of the multiple subjects. This allows the estimation system 1 to improve the estimation accuracy of the probability that an effect will be obtained from each of the multiple subjects by using the probability estimation model M1. As a result, the estimation accuracy of the combination of reminders is improved, and the estimation system 1 can make reminders to the subjects more efficient.

[0080] Furthermore, each of the multiple subjects is a defaulter who has fallen behind on his / her credit card payments. The predetermined means is a means of contacting the defaulter to urge him / her to pay. The predetermined action is urging him / her to pay. The cost is the cost for urging him / her to pay. The estimation system 1 can improve the efficiency of urging him / her to pay his / her credit card payments. The estimation system 1 can realize efficient debt collection for the entire credit card service.

[0081] [6. Modifications] The present disclosure is not limited to the above-described embodiments, and may be modified as appropriate without departing from the spirit of the present disclosure.

[0082] 8 is a diagram showing an example of functions realized in the modified example. For example, the server 10 includes a future risk information acquisition unit 106 and an initial value estimation model storage unit 107. The future risk information acquisition unit 106 is realized by the control unit 11. The initial value estimation model storage unit 107 is realized by the storage unit 12.

[0083] [6-1. Variation 1] For example, depending on the contact method, the cost may vary depending on whether or not the target responds. If the contact method is a person call, if the target responds to the call from the person in charge, the person in charge will have to talk with the target, which takes time, and the cost will be high. If the target does not respond to the call from the person in charge, the person in charge will not have to talk with the target, which takes less time, and the cost will be low. For this reason, instead of the cost being a fixed value as in the embodiment, the cost may be calculated based on the probability of the target responding.

[0084] The cost information acquisition unit 103 of Modification 1 calculates the probability of a response when a reminder is sent to each of a plurality of subjects at a predetermined timing via a predetermined contact means, and acquires cost information based on the probability for each of the plurality of subjects. Modification 1 takes as an example a case in which the probability of a response is calculated using the probability estimation model M1, but the probability of a response may also be calculated using a model other than the probability estimation model M1. The output data from the probability estimation model M1 of Modification 1 indicates the probability of a response, rather than the probability of a payment being made. Therefore, the output portion of the training data also indicates the probability of a response, rather than the probability that a training subject will make a payment.

[0085] For example, the cost information acquisition unit 103 inputs, for each subject, input data indicating a combination of subject characteristic information of the subject and a predetermined timing to the probability estimation model M1. The input data may be the same as in the embodiment. The probability estimation model M1 calculates an embedded representation of the input data and outputs output data indicating the probability of a response according to the embedded representation. The cost information acquisition unit 103 acquires the output data to obtain the probability of the subject's response. Although this embodiment differs from the embodiment in that the output data indicates the probability of a response, the processing executed by the probability estimation model M1 may be the same as in the embodiment.

[0086] For example, the cost information acquisition unit 103 calculates the cost of each of multiple subjects so that the higher the probability of response for that subject, the higher the final cost required for that subject, and acquires cost information. The cost information acquisition unit 103 acquires cost information indicating the final cost required for each subject by substituting the probability of that subject and the cost of the means of contact into a predetermined calculation formula for each subject. The calculation formula defines the relationship between probability and cost. The cost information acquisition unit 103 may acquire cost information indicating the final cost required for each subject by multiplying the probability of that subject by the cost of the means of contact for that subject. The cost information acquisition unit 103 may perform this multiplication based on a predetermined weighting coefficient. Although the method of acquiring cost information differs from the embodiment, other processes may be similar to those of the embodiment.

[0087] The estimation system 1 of the first modification calculates the probability of a response when a reminder is sent to each of a plurality of subjects at a predetermined timing by a predetermined means of communication, and acquires cost information based on the probability for each of the plurality of subjects. This allows the estimation system 1 to estimate a reminder combination based on cost information corresponding to the subject's response probability, thereby improving the accuracy of estimating the reminder combination. As a result, the estimation system 1 can make reminders to subjects more efficient.

[0088] [6-2. Variation 2] For example, if the person in charge contacts the target person but the target person does not make a payment, the person in charge may send another reminder to the target person. In this case, the person in charge will have to send multiple reminders, and the total cost will be high when future costs are also taken into account. For this reason, the cost may be calculated based on the probability of obtaining a benefit from the target person.

[0089] The cost information acquisition unit 103 of the second modification calculates the probability that an effect will be obtained when a reminder is sent to each of a plurality of subjects at a predetermined timing by a predetermined means of communication, and acquires cost information based on the probability for each of the plurality of subjects. For example, the cost information acquisition unit 103 calculates the probability that an effect will be obtained from each subject based on the probability estimation model M1 described in the embodiment. The method of calculating the probability may be the same as in the embodiment.

[0090] For example, the cost information acquisition unit 103 calculates the cost of each of a plurality of subjects so that the lower the probability of obtaining an effect from each of the subjects, the higher the final cost required for that subject, and acquires cost information. The cost information acquisition unit 103 acquires cost information indicating the final cost required for each subject by substituting the probability of obtaining an effect from that subject and the cost of the means of communication into a predetermined calculation formula for each subject. The calculation formula defines the relationship between probability and cost. The cost information acquisition unit 103 may acquire cost information indicating the final cost required for each subject by multiplying the value obtained by subtracting the probability of obtaining an effect from that subject from a predetermined value (e.g., 1) by the cost of the means of communication. The cost information acquisition unit 103 may perform this multiplication based on a predetermined weighting coefficient. Although the method of acquiring cost information differs from the embodiment, other processing may be similar to the embodiment.

[0091] The estimation system 1 of the second modification calculates the probability that an effect will be obtained when reminders are sent to each of a plurality of subjects at a predetermined timing by a predetermined means of communication, and acquires cost information based on the probability for each of the plurality of subjects. This allows the estimation system 1 to estimate a reminder combination based on cost information corresponding to the probability that an effect will be obtained from each subject, thereby improving the estimation accuracy of the reminder combination. As a result, the estimation system 1 can make reminders to subjects more efficient.

[0092] [6-3. Variation 3] For example, some subjects may not only be in arrears with past payments, but may also have applied for large installment or bonus payments. While future installment or bonus payments are not currently in arrears, there is a possibility that the subject may fall behind on future installment or bonus payments. Therefore, future installment or bonus payments can be considered a future risk. Variation 3 provides an example of a case in which such future risks are taken into consideration when determining the combination of reminders.

[0093] The estimation system 1 of variant example 3 includes a future risk information acquisition unit 106. The future risk information acquisition unit 106 acquires future risk information regarding the risk that a reminder will be required in the future for each of a plurality of subjects. The future risk information is information related to the need for reminders. The future risk information may be information that directly indicates a risk, or may be information that is correlated with a risk. In variant example 3, an example is given in which the future risk information is represented numerically, but the future risk information may also be represented by letters or symbols other than numerically.

[0094] For example, the future risk information indicates the amount of installment payments or bonus payments of the subject. The amount of installment payments or bonus payments of the subject is assumed to be indicated in the usage status information stored in the subject database DB. Therefore, the future risk information acquisition unit 106 acquires the future risk information by acquiring the usage status information of the subject from the subject database DB. The future risk information acquisition unit 106 may acquire the future risk information from a database other than the subject database DB, a computer other than the server 10, or an external information storage medium.

[0095] In addition, the future risk information may indicate other amounts that the subject may become delinquent in in the future, in addition to the amount of the installment payment or bonus payment of the subject. For example, the future risk information may indicate the amount of cash advance of the subject. When a model for estimating future risk is prepared, the future risk information may be estimated based on the subject's subject characteristic information and the model. The future risk information acquisition unit 106 may acquire future risk information using a known model such as a credit model. The future risk information acquisition unit 106 may use credit information stored by an external credit institution as future risk information.

[0096] The estimation unit 105 of the third modification estimates a combination of reminders further based on future risk information. For example, the estimation unit 105 multiplies the expected collection efficiency based on a weighting coefficient corresponding to the future risk information. For example, the higher the future risk indicated by the future risk information, the larger the weighting coefficient. The estimation unit 105 calculates the expected collection efficiency of a target person so that the higher the future risk of the target person, the larger the expected collection efficiency of the target person, thereby making the target person more likely to be selected as a reminder target.

[0097] The estimation system 1 of the third modification acquires future risk information for each of a plurality of subjects. The estimation system 1 estimates a reminder combination further based on the future risk information. This allows the estimation system 1 to estimate a reminder combination taking into account the future risks of the subjects, thereby improving the accuracy of the reminder combination estimation. As a result, the estimation system 1 can make reminders for subjects more efficient.

[0098] [6-4. Variation 4] For example, in Modification 3, the future risk information acquisition unit 106 may acquire future risk information based on the probability that a risk will occur for each of a plurality of subjects and the degree of the risk. In Modification 4, the future risk information acquisition unit 106 calculates the probability that a risk will occur based on demographic information for each of a plurality of subjects. For example, the future risk information acquisition unit 106 inputs the demographic information of the subjects into a credit model that calculates the probability of a risk occurring, and acquires the probability output by the credit model. The probability may be calculated based on a model other than the credit model. The probability may be calculated by a program for calculating probability, rather than a machine learning model in particular.

[0099] The degree of risk is the magnitude of the risk. For example, the amount of the installment payment or bonus payment of the subject corresponds to the degree of risk. The degree of risk may also be the amount of cash advance of the subject. The degree of risk is assumed to be stored in the subject database DB. The future risk information acquisition unit 106 acquires the degree of risk of the subject from the subject database DB. The future risk information acquisition unit 106 may acquire the degree of risk of the subject from a database other than the subject database DB, a computer other than the server 10, or an external information storage medium.

[0100] For example, the future risk information acquisition unit 106 may acquire future risk information for each of a plurality of subjects by multiplying the probability of a risk occurring for each of the plurality of subjects by the degree of the risk for that subject. The future risk information acquisition unit 106 may acquire future risk information by multiplying the probability of a risk occurring for each of the plurality of subjects and the degree of the risk for that subject based on a weighting coefficient associated with at least one of these. The future risk information acquisition unit 106 may input the probability of a risk occurring for each of a plurality of subjects and the degree of the risk for that subject into a model for estimating future risk information, and acquire future risk information for that subject output by the model.

[0101] The inference system 1 of the fourth modification acquires future risk information based on the probability that a risk will occur to each of a plurality of subjects and the degree of the risk. This allows the inference system 1 to improve the accuracy of estimating future risk information. As a result, the inference system 1 can improve the accuracy of estimating reminder combinations, thereby making reminders for subjects more efficient.

[0102] [6-5. Variation 5] For example, the method of acquiring future risk information is not limited to the examples of Modifications 3 and 4. For example, when multiple subjects each use a predetermined service, the future risk information acquisition unit 106 may acquire future risk information based on the past usage of the service by each of the multiple subjects. The usage status is indicated in usage status information stored in the subject database DB. For example, the future risk information acquisition unit 106 determines whether each of the multiple subjects has been in arrears with payments in the past based on the past usage status of each of the multiple subjects. The future risk information acquisition unit 106 may identify the amount of past arrears of each of the multiple subjects based on the past usage status of each of the multiple subjects.

[0103] For example, the future risk information acquisition unit 106 acquires future risk information so that the risk of a subject who has previously defaulted on payments is higher than the risk of a subject who has not previously defaulted on payments. The calculation formula required to acquire the future risk information is stored in the data storage unit 100. The calculation formula indicates the relationship between the presence or absence of delinquency and risk. The future risk information acquisition unit 106 acquires future risk information indicating the risk depending on the presence or absence of delinquency based on the calculation formula.

[0104] For example, the future risk information acquisition unit 106 may acquire future risk information such that the greater the amount of past due payments that a given subject has made in the past, the higher the risk of that subject. In this case, the above formula indicates the relationship between the amount of past due payments and the risk. The future risk information acquisition unit 106 acquires future risk information indicating the risk corresponding to the amount of past due payments based on the formula. The future risk information acquisition unit 106 may acquire future risk information based on a model that estimates future risk information based on the subject's usage status.

[0105] The estimation system 1 of the fifth modification acquires future risk information based on the past usage of services by each of a plurality of subjects. This allows the estimation system 1 to improve the accuracy of estimating future risk information. As a result, the estimation system 1 can improve the accuracy of estimating reminder combinations, thereby making reminders for subjects more efficient.

[0106] [6-6. Variation 6] For example, the estimation unit 105 may estimate a combination of reminders so that reminders are given priority to subjects for whom the effect is relatively large. In the example of FIG. 5, the estimation unit 105 increases the weighting coefficient used in calculating the expected collection efficiency as the expected collection amount increases. The relationship between the expected collection amount and the weighting coefficient is assumed to be predetermined in the data storage unit 100. The estimation unit 105 may acquire a weighting coefficient corresponding to the expected collection amount based on the relationship. The estimation unit 105 calculates the expected collection efficiency by multiplying the expected collection amount by the weighting coefficient and dividing the result by the expected cost. The estimation unit 105 may calculate the expected collection efficiency in the same manner as in the embodiment, and then multiply the weighting coefficient corresponding to the amount of the expected collection amount.

[0107] For example, the estimation unit 105 may increase the weighting coefficient used in calculating the expected collection efficiency as the amount of arrears of the target person increases. The relationship between the amount of arrears and the weighting coefficient is assumed to be predetermined in the data storage unit 100. The estimation unit 105 may obtain a weighting coefficient according to the amount of arrears based on the relationship. The estimation unit 105 may calculate the expected collection amount by multiplying the amount of arrears by the weighting coefficient and then multiplying the result by the probability. Even when performing such a calculation, the estimation unit 105 can estimate a combination of reminders so that reminders are given priority for targets for whom the effect is relatively large.

[0108] The estimation system 1 of the sixth modification estimates a combination of reminders so that actions are preferentially taken on subjects who will receive a relatively large effect. This makes it easier for the estimation system 1 to select subjects who will receive a relatively large effect, thereby increasing the effect obtained when reminding the subjects is successful.

[0109] [6-7. Variation 7] For example, the constraint information acquisition unit 104 may acquire constraint information in which constraints are defined according to a predetermined timing. The relationship between the timing and the constraint is assumed to be stored in the data storage unit 100. In the seventh modification, the data storage unit 100 stores constraint information for each of a plurality of timings. The constraint information acquisition unit 104 acquires constraint information for each of the plurality of timings from the data storage unit 100. When there is only one timing, the data storage unit 100 stores the constraint information for that one timing. The constraint information acquisition unit 104 only needs to acquire cost information for that one timing. Even when there are multiple timings, the cost information may be common to the multiple timings.

[0110] The estimation unit 105 of the seventh modification estimates a combination of reminders within a constraint corresponding to the timing. For example, the estimation unit 105 may estimate, for each timing, a combination of a target person to whom an action is to be taken and at least one of a timing and a contact method so as to obtain an overall efficient effect within the constraint corresponding to the timing. The constraints may include conditions other than cost. For example, the constraints may include, as a condition, that the total value of the expected collection efficiency is equal to or greater than a predetermined value. In this case, satisfaction of the condition corresponds to being within the constraints. The predetermined value may differ for each timing. For example, the predetermined value may be a value corresponding to the amount collected in the same period in the past. Although the method of acquiring the constraints used when estimating a reminder combination is different from that of the embodiment, the estimation of the reminder combination itself may be the same as that of the embodiment.

[0111] The estimation system 1 of the seventh modification acquires constraint information in which constraints according to predetermined timings are defined. The estimation system 1 estimates a combination within the range of the constraints according to the predetermined timings. As a result, even if the constraints differ for each timing, the estimation system 1 can estimate a combination that will efficiently obtain an effect overall within the range of the constraints for each of the multiple timings.

[0112] [6-8. Variation 8] For example, estimation by the estimation unit 105 may be completed quickly or may take a long time depending on the combination used as the initial value. The initial value is the combination when estimating the combination begins. In the example of FIG. 5, the initial value of the combination of variables 0 and 1 corresponds to the initial value of Modification 8. Therefore, if the estimation system 1 can increase the accuracy of the initial values ​​used in the estimation by the estimation unit 105, it will be able to quickly estimate an overall efficient dunning combination. Therefore, Modification 8 will take as an example a case where a model for increasing the accuracy of the initial values ​​is prepared.

[0113] The estimation system 1 of the eighth modification includes an initial value estimation model storage unit 107. The initial value estimation model storage unit 107 stores an initial value estimation model M2 in which the relationship between the training effect information and cost information and the training reminder combinations has been learned. In the eighth modification, as in the embodiment, the server 10 is used to learn the initial value estimation model M2, but a computer other than the server 10 may also learn the initial value estimation model M2. For example, the staff terminal 20 may also learn the initial value estimation model M2. The initial value estimation model M2 includes a program that performs calculations such as embedded representations and parameters referenced by the program. The parameters of the initial value estimation model M2 are adjusted by learning. The program and parameters used as the initial value estimation model M2 may be similar to known programs and parameters. For example, the parameters may be weights and biases.

[0114] For example, the data storage unit 100 may store a training database in which training data necessary for training the initial value estimation model M2 is stored. The training data includes an input portion that is input to the initial value estimation model M2 during training and an output portion that becomes the correct answer during training. The input portion of the training data has basically the same format as the input data that is input to the initial value estimation model M2 during estimation. The output portion of the training data has basically the same format as the output data that is output from the initial value estimation model M2 during estimation. Note that the input portion of the training data may have a slightly different format from the input data that is input to the initial value estimation model M2 during estimation. Similarly, the output portion of the training data may have a slightly different format from the output data that is output from the initial value estimation model M2 during estimation.

[0115] For example, the input portion of the training data includes training effect information. The input portion of the training data may also include the timing at which a prompt is sent to the training subject. The input portion of the training data may include training cost information. This information may be the same as the information described with reference to FIG. 5. The output portion of the training data is a combination of prompts corresponding to the effect information, etc., which is the paired input portion (a combination of prompts that is the correct answer during learning). The training data may be created by an administrator of the estimation system 1 or may be created by a training data creation tool.

[0116] For example, since an optimal solution may be obtained with a relatively small number of training data, the output portion of the training data may be an optimal solution obtained from a relatively small number of training data. The optimal solution may be obtained by the metaheuristic method described above. The optimal solution may be specified within the range of constraints used for generating the training data. The server 10 trains the initial value estimation model M2 so that the output portion of the training data is output when the input portion of the training data is input. The server 10 trains the initial value estimation model M2 by adjusting the parameters of the initial value estimation model M2 based on the training data.

[0117] The algorithm for training the initial value estimation model M2 may be a known algorithm used in the field of machine learning. For example, the server 10 causes the initial value estimation model M2 to learn training data based on an algorithm such as gradient descent or backpropagation. The loss function used during training may also be a known loss function. Based on the loss function, the server 10 calculates a loss, which is the error between the output portion of the training data and the output from the initial value estimation model M2 during training. The server 10 completes the training when the loss becomes small to a certain extent. Upon completing the training, the server 10 records the trained initial value estimation model M2 in the initial value estimation model storage unit 107.

[0118] The estimation unit 105 of Modification 8 acquires initial values ​​for a reminder combination based on the effect information and cost information of each of a plurality of subjects and the initial value estimation model M2, and estimates a reminder combination based on the initial values. For example, the estimation unit 105 inputs the effect information and cost information of each of a plurality of subjects into the initial value estimation model M2. The initial value estimation model M2 calculates embedded expressions of the effect information and cost information of each of the plurality of subjects, and outputs initial values ​​corresponding to the embedded expressions. The estimation unit 105 acquires the initial values ​​output from the initial value estimation model M2. The estimation unit 105 estimates a reminder combination based on the initial values. Although the method of acquiring the initial values ​​for the reminder combination differs from that of the embodiment, the other points are as described in the embodiment.

[0119] The estimation system 1 of Modification 8 stores an initial value estimation model M2 that has learned the relationship between training effect information and cost information and training combinations. The estimation system 1 obtains initial values ​​for combinations based on the effect information and cost information of each of multiple subjects and the initial value estimation model M2, and estimates combinations based on the initial values. This allows the estimation system 1 to quickly estimate an overall efficient reminder combination.

[0120] The estimation system 1 of Modification 8 may not include the functions described in the embodiment. The estimation system 1 may have only a configuration for performing a predetermined estimation based on initial values ​​estimated based on the initial value estimation model M2 stored in the initial value estimation model storage unit 107. In this case, the estimation system 1 may have only a configuration for performing a predetermined estimation based on initial values ​​estimated based on the initial value estimation model M2 stored in the initial value estimation model storage unit 107, without including a configuration for estimating a combination of reminders based on effect information, cost information, and constraint information. This configuration improves the accuracy of the initial values, thereby speeding up the completion of estimation when estimating a combination. Such a configuration (e.g., a configuration that solves the problem described in this paragraph without solving the problem described in the "Problem to be Solved by the Invention" section) is also within the scope of the present disclosure.

[0121] [6-9. Other variations] For example, the above modifications 1 to 8 may be combined.

[0122] For example, a function described as being realized by the subject terminal 30 may be realized by the server 10, the person in charge terminal 20, or another computer. A process described as being realized by the subject terminal 30 may be shared among multiple computers. A process described as being realized by the server 10 may be realized by the subject terminal 30, the person in charge terminal 20, or another computer. The main functions of the estimation system 1 may be shared among multiple computers.

[0123] [7. Notes] For example, the estimation system can be configured as follows. (1) an effect information acquisition unit that acquires effect information regarding an effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects; a cost information acquisition unit that acquires cost information regarding the cost required when the action is performed by the means at the timing for each of the plurality of subjects; a constraint information acquisition unit that acquires constraint information regarding constraints on costs allowed for the action; an estimation unit that estimates a combination of the subjects to whom the action will be performed and at least one of the timing and the means, based on the effect information and the cost information of each of the plurality of subjects and the constraint information, so that the effect will be efficiently obtained overall within the scope of the constraints; An estimation system including: (2) the effect information acquisition unit acquires the effect information for each of the plurality of subjects based on a probability that the effect will be obtained when the action is performed by the means at the timing and a degree of the effect obtained from the subject; The estimation system according to (1). (3) The estimation unit calculating an index relating to the efficiency of the effect obtained from each of the plurality of subjects based on the effect information and the cost information of the subjects; estimating the combination based on the indicators of each of the plurality of subjects; The estimation system according to (1) or (2). (4) the cost information acquisition unit calculates a probability of a response when the action is performed by the means at the timing for each of the plurality of subjects, and acquires the cost information based on the probability for each of the plurality of subjects. The estimation system according to any one of (1) to (3). (5) the cost information acquisition unit calculates a probability that the effect will be obtained when the action is performed by the means at the timing for each of the plurality of subjects, and acquires the cost information based on the probability for each of the plurality of subjects. The estimation system according to any one of (1) to (4). (6) The estimation system further includes a future risk information acquisition unit that acquires future risk information regarding a risk that the action will be required in the future for each of the plurality of subjects; the estimation unit estimates the combination further based on the future risk information. The estimation system according to any one of (1) to (5). (7) the future risk information acquisition unit acquires the future risk information based on the probability that the risk will occur to each of the plurality of subjects and the degree of the risk; (6) The estimation system according to (6). (8) Each of the plurality of subjects uses a predetermined service, the future risk information acquisition unit acquires the future risk information based on a past usage status of the service by each of the plurality of subjects; The estimation system according to (6) or (7). (9) the estimation unit estimates the combination such that the action is preferentially performed on the subject for whom the effect is relatively large. The estimation system according to any one of (1) to (8). (10) the constraint information acquisition unit acquires the constraint information in which the constraint according to the timing is defined; the estimation unit estimates the combination within the range of the constraint according to the timing. The estimation system according to any one of (1) to (9). (11) the constraint information acquisition unit acquires the constraint information in which the constraints corresponding to the means are defined; the estimation unit estimates the combination within the range of the constraints according to the means; The estimation system according to any one of (1) to (10). (12) the estimation system further includes an initial value estimation model storage unit that stores an initial value estimation model in which a relationship between the effect information and the cost information for training and the combination for training has been learned; the estimation unit acquires initial values ​​of the combination based on the effect information and the cost information of each of the subjects and the initial value estimation model, and estimates the combination based on the initial values. The estimation system according to any one of (1) to (11). (13) the estimation system further includes a probability estimation model storage unit that stores a probability estimation model in which a relationship between subject characteristic information regarding characteristics of the subject for training and a probability of obtaining the effect is learned; the effect information acquisition unit acquires a probability that the effect will be obtained from each of the plurality of subjects based on the effect information of each of the plurality of subjects and the probability estimation model, and acquires the effect information based on the probability for each of the plurality of subjects. The estimation system according to any one of (1) to (12). (14) Each of the plurality of subjects is a delinquent who has fallen behind on his / her credit card payments, The means is a means of contacting the defaulter to urge them to pay, the action is the reminder; The cost is the cost for the reminder. The estimation system according to any one of (1) to (13). [Explanation of symbols]

[0124] 1 Estimation system, 10 Server, 11, 21, 31 Control unit, 12, 22, 32 Memory unit, 13, 23, 33 Communication unit, 24, 34 Operation unit, 25, 35 Display unit, 20 Person in charge terminal, 30 Subject terminal, DB Subject database, M1 Probability estimation model, M2 Initial value estimation model, N Network, SC Screen, 100 Data storage unit, 101 Probability estimation model storage unit, 102 Effect information acquisition unit, 103 Cost information acquisition unit, 104 Constraint information acquisition unit, 105 Estimation unit, 106 Future risk information acquisition unit, 107 Initial value estimation model storage unit.

Claims

1. an effect information acquisition unit that acquires effect information regarding an effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects; a cost information acquisition unit that acquires cost information regarding the cost required when the action is performed by the means at the timing for each of the plurality of subjects; a constraint information acquisition unit that acquires constraint information regarding constraints on costs allowed for the action; an estimation unit that estimates a combination of the subjects to whom the action will be performed and at least one of the timing and the means, based on the effect information and the cost information of each of the plurality of subjects and the constraint information, so that the effect will be efficiently obtained overall within the scope of the constraints; An estimation system including:

2. the effect information acquisition unit acquires the effect information for each of the plurality of subjects based on a probability that the effect will be obtained when the action is performed by the means at the timing and a degree of the effect obtained from the subject; The estimation system of claim 1 .

3. The estimation unit calculating an index relating to the efficiency of the effect obtained from each of the plurality of subjects based on the effect information and the cost information of the subjects; estimating the combination based on the indicators of each of the plurality of subjects; The estimation system according to claim 1 or 2.

4. the cost information acquisition unit calculates a probability of a response when the action is performed by the means at the timing for each of the plurality of subjects, and acquires the cost information based on the probability for each of the plurality of subjects. The estimation system according to claim 1 or 2.

5. the cost information acquisition unit calculates a probability that the effect will be obtained when the action is performed by the means at the timing for each of the plurality of subjects, and acquires the cost information based on the probability for each of the plurality of subjects. The estimation system according to claim 1 or 2.

6. The estimation system further includes a future risk information acquisition unit that acquires future risk information regarding a risk that the action will be required in the future for each of the plurality of subjects; the estimation unit estimates the combination further based on the future risk information. The estimation system according to claim 1 or 2.

7. the future risk information acquisition unit acquires the future risk information based on the probability that the risk will occur to each of the plurality of subjects and the degree of the risk; The estimation system according to claim 6 .

8. Each of the plurality of subjects uses a predetermined service, the future risk information acquisition unit acquires the future risk information based on a past usage status of the service by each of the plurality of subjects; The estimation system according to claim 6 .

9. the estimation unit estimates the combination such that the action is preferentially performed on the subject for whom the effect is relatively large. The estimation system according to claim 1 or 2.

10. the constraint information acquisition unit acquires the constraint information in which the constraint according to the timing is defined; the estimation unit estimates the combination within the range of the constraint according to the timing. The estimation system according to claim 1 or 2.

11. the constraint information acquisition unit acquires the constraint information in which the constraints corresponding to the means are defined; the estimation unit estimates the combination within the range of the constraints according to the means; The estimation system according to claim 1 or 2.

12. the estimation system further includes an initial value estimation model storage unit that stores an initial value estimation model in which a relationship between the effect information and the cost information for training and the combination for training has been learned; the estimation unit acquires initial values ​​of the combination based on the effect information and the cost information of each of the subjects and the initial value estimation model, and estimates the combination based on the initial values. The estimation system according to claim 1 or 2.

13. the estimation system further includes a probability estimation model storage unit that stores a probability estimation model in which a relationship between subject characteristic information regarding characteristics of the subject for training and a probability of obtaining the effect is learned; the effect information acquisition unit acquires a probability that the effect will be obtained from each of the plurality of subjects based on the effect information of each of the plurality of subjects and the probability estimation model, and acquires the effect information based on the probability for each of the plurality of subjects. The estimation system according to claim 1 or 2.

14. Each of the plurality of subjects is a delinquent who has fallen behind on his / her credit card payments, The means is a means of contacting the defaulter to urge them to pay, the action is the reminder; The cost is the cost for the reminder. The estimation system according to claim 1 or 2.

15. an effect information acquisition step of acquiring effect information regarding an effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects; a cost information acquisition step of acquiring cost information regarding the cost required when the action is performed by the means at the timing for each of the plurality of subjects; a constraint information acquisition step of acquiring constraint information regarding constraints on costs allowed for the action; an estimation step of estimating a combination of the subjects to whom the action will be performed and at least one of the timing and the means, so that the effect will be efficiently obtained overall within the scope of the constraints, based on the effect information and the cost information of each of the plurality of subjects and the constraint information; Estimation methods including:

16. an effect information acquisition unit that acquires effect information regarding an effect obtained when a predetermined action is performed by a predetermined means at a predetermined timing for each of a plurality of subjects; a cost information acquisition unit that acquires cost information regarding the cost required when the action is performed by the means at the timing for each of the plurality of subjects; a constraint information acquisition unit that acquires constraint information regarding constraints on costs allowed for the action; an estimation unit that estimates a combination of the subjects for which the action will be taken and at least one of the timing and the means, based on the effect information and the cost information for each of the plurality of subjects and the constraint information, so that the effect is obtained efficiently as a whole within the scope of the constraints; A program that allows a computer to function as a

Images