Business support method, business support system, business support program, and trained model used therefor, generation method thereof, and execution program thereof

Abstract

To provide business support that proposes a sales pitch suitable for a customer.SOLUTION: A business support system 10 outputs sales pitch examples suitable for a customer under negotiation, the business support system 10 including: contract factor prediction models 21, 22 trained by machine learning using training data so as to predict, on receipt of an input of attribute information i1 including age information and sexuality information of the customer under negotiation, a contract factor correlating with the attribute information i1; and predictive pitch generation units 23, 24 which generates sales pitch examples p1, p2 suitable for the customer under negotiation, using a large language model 31, based on the predicted contract factor for the customer under negotiation, the contract factor being predicted by the contract factor prediction models 21, 22 by inputting the attribute information related to the customer under negotiation to the contract factor prediction models 21, 22, and the attribute information i1 used for the prediction.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a sales support method, a sales support system, a trained model used therein, a generation method thereof, and an execution program thereof. [Background technology]

[0002] Conventional sales support systems include those that evaluate the capabilities of salespeople (Patent Document 1), predict sales deals (Patent Documents 2 and 3), and generate sales plans for uncontracted customers (Patent Document 4).

[0003] However, until now, there has been no sales support system that can suggest sales pitches suitable for specific customers.

[0004] In other words, the results of sales activities depend heavily on the abilities of individual salesmen. Therefore, in most sales companies, when a top salesperson leaves, sales drop. This is because the sales talk used to improve the success rate is not retained as an asset for the company. Even if the sales talk remains, it cannot be used effectively unless it is linked to the experience of what kind of customers it was effective for. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2000-222486 A [Patent Document 2] JP 2019-79302 A [Patent Document 3] JP 2020-87023 A [Patent Document 4] Patent Publication No. 2021-64406 [Patent Document 5] Patent Publication No. 2022-69602 Summary of the Invention [Problem to be solved by the invention]

[0006] Therefore, a main object of the present invention is to provide sales support that proposes a sales pitch that is suitable for each customer. [Means for solving the problem]

[0007] The sales support methods that solved the above problems are as follows. <First aspect> A method for supporting sales by outputting examples of sales pitches suitable for uncontracted customers, the method comprising the steps of: The information processing system includes a contract factor prediction model that is machine-learned using training data so as to predict a contract factor correlated with attribute information of an uncontracted customer when the attribute information of the uncontracted customer is input, A step in which the contract factor prediction model predicts the contract factor from the input attribute information of the unconverted customers; the information processing system includes a step of generating a sales pitch example suitable for the uncontracted customer by using a large-scale language model based on the predicted contract factor of the uncontracted customer and the attribute information used for predicting the predicted contract factor; The attribute information includes age information and gender information. A sales support method comprising:

[0008] (Action and effect) If it is possible to accurately predict the closing factors (factors that affect closing, such as concerns and decision points) that differ from customer to customer, it is possible to deliver a sales pitch that addresses these factors, leading to an increased probability of closing a deal. On the other hand, closing factors that differ from customer to customer are correlated with attribute information including age and gender. Therefore, as in this sales support method, by predicting the closing factors of unconverted customers using a closing factor prediction model based on the attribute information of the unconverted customers and generating example sales pitches that take these predicted closing factors into account, even inexperienced salespeople will be able to use effective sales pitches tailored to individual customers.

[0009] <Second aspect> A sales support system that outputs examples of sales talks suitable for uncontracted customers, A sales factor prediction model that is machine-learned using training data so that when attribute information of unconverted customers is input, sales factors that correlate with the attribute information are predicted; a predictive speech generation unit that generates an example of a sales pitch suitable for a non-contracted customer by using a large-scale language model based on a predicted contract factor of the non-contracted customer predicted by the contract factor prediction model by inputting the attribute information on the non-contracted customer into the contract factor prediction model and the attribute information used for predicting the predicted contract factor, The attribute information includes age information and gender information. A sales support system characterized by:

[0010] (Action and effect) This provides the same effects as the first embodiment.

[0011] <Third aspect> The predictive talk generation unit outputs the sales talk example together with the predictive closing factors used to generate the sales talk example. A sales support system according to a second aspect.

[0012] (Action and effect) Although the sales talk generated by the predictive talk generator is sufficient in itself, it is useful to know what the predicted closing factors used in generating the sales talk were when using the generated sales talk. Therefore, it is preferable to output the sales talk example generated in this embodiment together with the predicted closing factors used in generating the sales talk.

[0013] <Fourth aspect> a contract storage unit in which attribute information of a contracted customer who has successfully concluded a contract and a contract factor that caused the contract are stored in association with each customer; and a case-based sales talk generation unit that generates sales talk examples suitable for the uncontracted customers using a large-scale language model based on the input attribute information of the uncontracted customers and the contract factors of the contracted customers having attribute information that matches or is similar to the input attribute information of the uncontracted customers, the case-based sales talk generation unit being read from the contract storage unit. The sales support system according to the second or third aspect.

[0014] (Action and effect) According to this sales support system, when negotiating with a customer who has not yet made a contract, a sales pitch suitable for that customer can be generated by referring to past contract cases. Since there is a correlation between attribute information including age and gender and contract factors (concerns, decision points, etc.), there is a commonality in the contract factors of customers with matching or similar attribute information. Therefore, by continuously storing attribute information of past contract customers and the contract factors of those contract customers in the contract storage section, this system becomes an asset that can be easily used. For example, even an inexperienced salesperson can use an effective sales pitch backed up by past performance by using this sales support system.

[0015] <Fifth aspect> The example-based talk generation unit includes: a search is made for closing factors of closing customers having attribute information that matches or is similar to the input attribute information of unconcluded customers from the closing storage unit, and the sales pitch example is generated using a large-scale language model based on the attribute information of the unconcluded customers and the closing factors that have been searched for as having the matching or similar attribute information thereto; A sales support system according to a fourth aspect.

[0016] (Action and effect) When generating a sales pitch example using a large-scale language model based on the input attribute information of unsold customers and the attribute information and closing factors of closed customers read from the closing storage unit, it is theoretically possible to input all records in the closing storage unit into the large-scale language model, but the more data there is, the slower the processing of the large-scale language model becomes. Also, the current situation is that the number of input tokens of the large-scale language model is limited. Therefore, as in this embodiment, if the closing factors of closed customers having attribute information that matches or is similar to the input attribute information of unsold customers are searched from the closing storage unit, and the closing factors to be referred to are narrowed down, and then a sales pitch example is generated using the large-scale language model, it is possible to reduce the processing load of the large-scale language model and avoid the constraint on the number of input tokens, and it is also possible to easily identify the closed customers, their attribute information, and closing factors to be referred to.

[0017] <Sixth aspect> The example-based talk generation unit outputs the generated sales talk example and the closing factors used to generate the sales talk example. A sales support system according to a fifth aspect.

[0018] (Action and effect) Although the sales talk generated by the example-based talk generator is sufficient in itself, it is useful to know what closing factors were referenced when using the generated sales talk example. Therefore, it is preferable to output not only the generated sales talk example as in this embodiment, but also the closing factors that were referenced.

[0019] <Seventh aspect> The contract factors include at least one of a contract concern and a contract decision point. A sales support system according to any one of the second to sixth aspects.

[0020] (Action and effect) Although closing factors differ depending on the customer, they are correlated with customer attribute information and are factors that affect closing. There are various closing factors, but a typical factor that all customers have is concerns about the contract. Accurately understanding the concerns about the contract and eliminating them as much as possible, in other words, taking into account the customer's concerns (anxieties) and delivering a sales pitch that addresses them, is one of the most important things in sales. In addition, the decision point for closing a contract is also an important closing factor. This is because if the decision point for closing a contract can be predicted, it becomes easier to encourage the customer to make a decision and closing the deal becomes easier. Therefore, it is preferable that closing factors include at least one of the concerns about the contract and the decision point for closing a contract.

[0021] <Eighth aspect> The closing factors include a closing decision point, A response speech output unit outputs a response speech example that corresponds one-to-one to the predicted decision point predicted by the contract factor prediction model. A sales support system according to any one of the second to sixth aspects.

[0022] (Action and effect) Even if the decision point can be predicted, it is not easy for an inexperienced salesman to imagine the sales talk to use at the closing stage. Therefore, it is preferable to predict the decision point of an unsold customer using a decision point prediction model based on the attribute information of the unsold customer, as in this sales support system, and output the predicted decision point and corresponding counter-phrase example. This allows even an inexperienced salesman to use an effective counter-phrase when closing a sale.

[0023] <Ninth aspect> A sales support program that causes an information processing system to output examples of sales pitches suitable for uncontracted customers, The information processing system includes a contract factor prediction model that is machine-learned using training data so as to predict a contract factor correlated with attribute information when the attribute information is input, The information processing system includes: A process in which the contract factor prediction model predicts the contract factor from the input attribute information of the uncontracted customers; and generating a sales pitch example suitable for the uncontracted customer using a large-scale language model based on the predicted contract factors of the uncontracted customer and the attribute information used for predicting the predicted contract factors. The attribute information includes age information and gender information. A sales support program characterized by:

[0024] (Action and effect) This provides the same effects as the first embodiment.

[0025] <Tenth aspect> When attribute information including age and gender information is input, the system is trained using training data to output a contract factor predicted from the attribute information. A trained model characterized by:

[0026] (Action and effect) As described above, in order to increase the possibility of closing a sale, it is important to accurately grasp the closing factors that differ from customer to customer. Furthermore, the closing factors are correlated with attribute information including age, gender, etc. Therefore, by using the trained model of this embodiment, it becomes possible to accurately predict the closing factors of unconverted customers based on the attribute information of the unconverted customers, and to effectively utilize the predictions in sales.

[0027] <Eleventh aspect> Using attribute information including age information and gender information, and training data including contract factors associated therewith, a contract factor prediction model is generated by machine learning, which outputs contract factors predicted from attribute information including age information and gender information when the attribute information including age information and gender information is input. A method for generating a trained model.

[0028] (Action and effect) This provides the same effects as the tenth aspect.

[0029] <Twelfth aspect> Information processing system, Attribute information including age information and gender information is input to the trained model of the tenth aspect, and a contract factor predicted from the attribute information is output. Execute the process, An execution program for a trained model.

[0030] (Action and effect) This provides the same effects as the tenth aspect. Effect of the Invention

[0031] The present invention brings about an advantage that it becomes possible to provide sales support that proposes sales pitches suited to the customer. [Brief description of the drawings]

[0032] [Figure 1] FIG. 1 is a system configuration diagram. [Diagram 2] FIG. 2 is a block diagram of main functions. [Diagram 3] 13 is a flowchart of a predictive talk generation unit. [Figure 4] 13 is a flowchart of a case-based talk generation unit. [Diagram 5] 1 is a flowchart of case registration. [Figure 6] 13 is a flowchart of a counter speech output unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0033] FIG. 1 shows an example of a sales support system 10 that outputs sales pitch examples suitable for uncontracted customers. This sales support system 10 is composed of a talk generation server 20 that outputs sales pitch examples suitable for uncontracted customers when attribute information i1 of the uncontracted customers is input. The sales support system 10 of this example assumes a case where sales support is performed by a large number of people, and assumes a client-server system in which input to the talk generation server 20 is performed via a terminal 40 connected via networks 101 to 103, and the sales pitch examples are output from the talk generation server 20 to the terminal 40, but may also be a stand-alone system.

[0034] That is, in the illustrated example, a sales talk example can be output as follows. That is, when a salesman requests an input screen for generating a sales talk on the terminal 40, the talk generation server 20, which receives the request, generates input screen display data and transmits it to the terminal 40. The terminal 40, which has received the input screen display data, displays the input screen on its display device. The salesman operates the input device of the terminal 40 and inputs attribute information i1 of the unsold customer who is the target of the sale on the input screen of the terminal 40. The attribute information will be described later, but after making a selection in a selection-type item and inputting a sentence in a description-type item, the attribute information is transmitted to the talk generation server 20 by clicking a decision button or the like. When the talk generation server 20 receives the attribute information i1 of the unsold customer, the talk generation server 20 generates output screen display data including a sales talk example x1 based on the received information and transmits it to the terminal 40. The terminal 40, which has received the output screen display data, displays an output screen including the sales talk example x1 on its display device.

[0035] The networks 101 to 103 are generally configured as the Internet 101, an intranet, a LAN 102, a mobile phone communication network 103, or a combination of these networks, but are not particularly limited as long as information can be transmitted and received between the token generation server 20 and the terminal 40. The networks 101 to 103 may be only wired, only wireless, or a combination of wired and wireless.

[0036] The sales support system 10 can be configured by one or more computers 20, 41 to 43 that store various programs and data required for implementing various prediction models, various storage units, and a talk generation unit, etc., which will be described later, and execute processes based on the programs. That is, the hardware and software configuration can be appropriately adopted from known configurations for realizing the sales support system 10. For example, the sales support system 10 can be a client-server system with a general Web three-layer architecture, which includes a Web server (presentation layer / UI layer) for inputting attribute information of uncontracted customers and displaying generated sales talk examples, an application server (application layer / logic layer / business layer) constituting the talk generation server 20, and a database server (data layer) for storing information required for generating sales talk examples.

[0037] The computers 20, 41 to 43 constituting the sales support system 10 are not particularly limited, but may have, for example, a CPU (central processing unit), a main storage device (RAM, random access memory), an auxiliary storage device (hard disk drive, solid state drive, etc.), and a network controller for connecting to a network, all connected via a bus. The sales support system 10 may also have input devices such as a keyboard and a pointing device, and an output device such as a display device, similar to the terminals described below.

[0038] The terminal 40 may be a personal computer 41, 42 that stores various programs required for communication with the sales support system 10 and executes processing based on the programs, or a mobile information terminal 43 such as a smartphone. The terminal 40 is not particularly limited, but may have a CPU (central processing unit), a main storage device (RAM, random access memory), an auxiliary storage device (hard disk drive, solid state drive, etc.), a network controller for connecting to the networks 101 to 103, an input device, and a display device, all connected via a bus. As the input device, a keyboard, a mouse, a pointing device such as a touch panel, etc. may be used.

[0039] (Contract factor prediction model) As shown in Fig. 2, the token generation server 20 includes contract factor prediction models 21 and 22 that are machine-learned to predict contract factors correlated with attribute information i1 of uncontracted customers when the attribute information i1 is input. The contract factor prediction models 21 and 22 can be learned by unsupervised learning, semi-supervised learning, or reinforcement learning, but supervised learning is preferred. In addition, as the algorithm of the prediction model in the case of supervised learning, known algorithms such as decision trees, random forests, support vector machines, neural networks, deep learning, naive Bayes method, and K-nearest neighbor method can be used.

[0040] The training data used in the supervised learning of the contract factor prediction models 21, 22 is a collection of a large number of data in which customer attribute information and contract factors are associated with each customer. As this training data, it is preferable to use both data on customers who have successfully concluded a contract and data on customers who have failed to conclude a contract, but only one of them may be used. In addition, the training data can be additionally stored in a training data storage unit such as a database (not shown) as needed, and machine learning can be performed to update the contract factor prediction models 21, 22 as needed. As the training data, information on some or all of the customers stored in the contract storage unit described later can also be used.

[0041] (Attribute information) The attribute information of the training data includes at least age information and gender information associated with each customer. The age information can be, for example, a range option such as 20s, 30s, 40s, 50s, 60s, 70s, 80s and over, or it can be the age itself. The gender information can be, for example, male, female, other options. These are particularly strongly correlated with the way of thinking about contracts. The more attribute information of the contracted customer, the better, so it is preferable to use, for example, one or more of the following in addition to the above two. When using this additional attribute information, some or all of the additional information may be null (no information input). The training data may also include any information such as personal information of the contracted customer (name, address, telephone number, etc.) and sales representative information in association with the attribute information. Occupation: The customer's occupational information may be, for example, but is not limited to, a selection of options such as manager, professional, company employee, temporary contract employee, part-time worker, student, etc., or any sentence (individually expressed sentence). Annual income: The customer's annual income information can be a range option such as less than 3 million, between 3 million and 5 million, between 5 million and 10 million, between 10 million and 20 million, and over 20 million, but it can also be the annual income amount itself. Disposable income: The customer's disposable income information can be a range option such as less than 3 million, between 3 million and 5 million, between 5 million and 10 million, between 10 million and 20 million, and over 20 million, but it can also be the amount of disposable income itself. Length of service: A customer's length of service reflects the customer's character and the stability of their livelihood, and so naturally affects the contract. Length of service information can be a range of options, such as less than 1 year, 1 to 3 years, 3 to 5 years, 5 to 10 years, 10 years or more, etc., or it can be the number of years itself. Marital status: Needless to say, whether the customer is married or not also has an impact on the contract. Marital status can be, for example, but is not limited to, options such as single, married, divorced single, and others. Children: Whether or not the customer has children also has an impact on the contract. Child information can be the number of children, such as 1, 2, 3, 4, 5 or more, or simply the presence or absence. It can also be the number of children by gender. Parents: The customer's parents may also affect the contract. Parent information can be, for example, whether they live together or not, or whether they are alive or deceased. · Color of clothing: The color of the clothing worn represents the customer's personality and correlates with their attitude toward the contract. Color information for clothing can be, for example, two options: monotone / dark and colorful / warm / patterned, but it can also be more detailed options or arbitrary text. Facial Expressions: Customer facial expressions reflect the customer's personality and therefore affect the contract. Facial expression information can be, for example, two options: stiff / cool / stern and calm / rich / smiling, but it can also be more detailed options or arbitrary sentences. · Pace of conversation: The pace of conversation reflects the customer's personality and therefore affects the contract. The pace of conversation information can be, for example, a binary option of fast or slow, but it can also be a more detailed option or an arbitrary sentence. · Intonation: Intonation of a conversation represents the personality of the customer and therefore affects the contract. Intonation information of a conversation can be, for example, two options, plain / monotone and intonation / high tone, but it can also be more detailed options or arbitrary sentences. Tone of conversation: Tone of conversation reflects the customer's personality and therefore influences the contract. Tone of conversation can be, for example, a binary choice between assertive / declarative and interrogative / questioning, but it can also be more specific or arbitrary.

[0042] (Contract Factors) Although the closing factors differ depending on the customer, they are correlated with the attribute information of the customer and are factors that affect the closing of the contract. There are various closing factors, but a typical factor that all customers have is the concern about the contract. Accurately grasping the concern about the contract and removing it as much as possible, that is, considering the customer's concerns (anxieties) and giving a sales pitch that addresses them, is one of the most important things in sales. In addition, the decision point of the closing is also one of the important closing factors. If the decision point of the closing can be predicted, it becomes easier to encourage the customer to make a decision and to close the deal. Therefore, it is preferable that the closing factors of the training data include at least one of the concern about the contract and the decision point of the closing. When predicting a plurality of closing factors, the closing factor prediction models 21, 22 can be provided according to the number of closing factors. In the illustrated example, there is provided a first contract factor prediction model 21 that predicts contract concerns and a second contract factor prediction model 22 that predicts decision points for contracting, but if only one of the contract factors is to be predicted, only one corresponding prediction model will be provided.

[0043] (Contract concerns) Contract concerns as contract closing factors can be defined as appropriate, but according to the inventor's knowledge, concerns can be classified into the following six categories; therefore, one or more of these can be associated with customer attribute information to create training data, or the six concerns can be associated with further subdivided concerns to create training data.

[0044] Concern 1: Money Concerns related to money include those directly related to the price of the product or service, such as high prices, insufficient budgets, and difficulty in continuing payments when paying in installments, as well as concerns about environmental factors, such as reduced income, increased child support costs, increased nursing care costs, and no savings, and concerns about payment methods, such as lifetime payments or fixed-term payments.

[0045] Concern 2: "Necessity of consulting a third party" This corresponds to a case where a customer cannot decide whether to enter into a contract without consulting a third party such as a close relative or an interested party. In this case, the relationship of the third party to the customer is important, so this can be made a sub-point of concern.

[0046] Concern 3: "Timing" This includes worries about "children's exams" and "job transfers." In this case, the timing is important in relation to the customer, so for example, weddings, births, entrance exams / schooling, job changes, birthdays, and other family events can be subdivided into concerns.

[0047] Concern #4: Reliability This applies to concerns about the product or service being sold, or the company that provides it. In this case, the concerns can be subdivided into, for example, the company's reputation, the reputation of the product or service, the person in charge, after-sales support, and the amount of money.

[0048] Concern 5: "Possibility of the existence of more advantageous products or services" This corresponds to the fear that there may be other products or services with better conditions (cheaper, higher quality, more thorough after-sales support, etc.) than the product or service being sold. In this case, the concerns can be subdivided into, for example, price, person in charge, after-sales support system, quality / performance, added value, etc.

[0049] Concern #6: "Fear of results, continuity, and change" These concerns vary by industry. For example, in gyms and classes, it is anxiety about whether or not you will see results or whether or not you can continue, and in cosmetics and insurance, it is anxiety about the change from what you are currently using or signing up for to the new one. Since it is difficult to break these concerns down into options, they can be written as any sentence.

[0050] (Decision Point) Decision points as factors for closing a sale can be determined as appropriate, but according to the inventor's knowledge, decision points can be classified into the following eight categories, for example. Therefore, training data can be created by associating one or more of these with customer attribute information, or by associating the eight decision points with further subdivided points or with the written text of the customer or salesperson itself.

[0051] Decision point 1: Try it out This is the case when a customer tries out a product or service and realizes its effectiveness, quality, value, and sustainability, and then concludes a purchase.

[0052] Decision point 2: Price This is the case when the customer is satisfied with the price and a deal is made.

[0053] Decision point 3: Educational advice This is when the salesperson helps the customer make the best choice, leading to a sale.

[0054] Decision point 4: Third-party consultation This is the case when a customer believes it is necessary to consult a third party, but a deal is concluded with or without consultation.

[0055] Decision point 5: Feasibility This is the case when the customer believes that it is not impossible to achieve and goes ahead with the deal.

[0056] Decision Point 6: Timing in Life This is when a customer believes that now is the time to purchase a product or service and concludes the deal.

[0057] Decision point 7: Future gains versus future losses This is the case when a customer makes a purchase while thinking about the future regrets they may have had if they had not made the purchase.

[0058] Decision point 8: Third party opinion This is when a customer hears the story of someone who was in the same situation as them and made a deal, and then goes on to make a sale.

[0059] (Predictive Talk Generation Unit) As shown in Fig. 2, the talk generation server 20 includes a predictive talk generation unit 23 that generates sales talk examples suitable for unconcluded customers using a large-scale language model (LLM) 31 based on the predicted contract factor m1 of unconcluded customers predicted by the contract factor prediction model 21 by inputting attribute information i1 related to unconcluded customers into the contract factor prediction model 21 and the attribute information i1 used to predict the predicted contract factor m1. In the illustrated example, the predictive talk generation unit 23 corresponds to the first contract factor prediction model 21 that predicts concerns about a contract, and does not include a second predictive talk generation unit corresponding to the second contract factor prediction model 22 that predicts decision points for contracting, but may be the opposite, or a single or separate predictive talk generation unit corresponding to both may be included.

[0060] More specifically, as shown in FIG. 3, the predictive talk generation unit 23 creates a prompt p1 for generating a sales talk example suitable for an unsold customer based on the predicted closing factor m1 of the unsold customer predicted by the closing factor prediction model 21 and the attribute information i1 used to predict it, and provides the prompt p1 to the large-scale language model 31 to obtain a sales talk example t1 output from the large-scale language model. The talk generation server 20 generates, for example, output screen information x1 (it is more preferable to use the predicted closing factor m1 used to generate the sales talk example t1) including the sales talk example t1 output from the large-scale language model 31 and transmits it to the terminal 40, and the terminal 40 that receives it can display the received output screen information x1 on the screen. At this time, if there are a sales talk example t3 or a counter-talk example t2 generated by the case-based talk generation unit 26 described later, the output screen information x1 including them can be generated and transmitted to the terminal 40, or the output screen information can be divided to generate a plurality of pieces of output screen information and transmitted to the terminal 40 in sequence.

[0061] The predictive talk generation unit 23 can output a plurality of sales talk examples each based on the same predicted closing factor. The predictive talk generation unit 23 can also use a large-scale language model to generate the prompt p1.

[0062] The prompt p1 for generating a sales talk example by the predictive talk generation unit 23 can be appropriately determined. As an example, the following shows an example of the prompt p1, a predicted contract factor (in this example, a concern), and a generated sales talk example t1 when a prototype of the sales support system 10 is used and example customer attribute information is input. <Example of prompt p1> "You are a top salesperson. Please output an appropriate sales pitch for proposing an insurance plan to a customer with the following input attribute information and the following predictive factors for purchase." <Input attribute information> "Gender: Male, Age: 30s, Occupation: Company employee, Married: Married, Children: 2, Parents: Separated, Annual income: 5 million to 10 million, Disposable income: 5 million to 10 million, Years of employment: 10 years or more, Color of clothing: Monotone, dark, Pace: Slow, Expression: Calm, rich, smiling, Intonation: Intonation, high tone, Tone of speech: Questioning, asking questions" <Prediction factors for closing: concerns> "Money, Reliability" <Sales pitch example t1> "Hello, customer. I understand your name and the details of your inquiry. I understand that you are married and have two children. From now on, it is important to place emphasis on insurance to protect your family. Today, I would like to propose suitable insurance products to address your concerns about "money" and "concerns about the company and the system." First, in terms of "money," I will propose suitable life insurance and medical insurance for your family, taking into consideration that your annual income is less than 10 million yen and you have been employed for more than 10 years. Life insurance for your family will ensure that your family will receive a comfortable amount in the event of an emergency, while medical insurance can cover medical expenses such as hospitalization and surgery due to illness or injury. Regarding the "concerns about the company or system," we also introduce insurance with a job change support service for company employees who are in the situation where their company goes bankrupt or is restructured. With this service, a job change agent will support the customer, increasing the chances of finding a more stable job. As you can tell from their conversation style and facial expressions, they have a high level of flexibility and communication skills. For such customers, we believe it is most important to propose insurance products that fit their future life plans. Let's work together to find the insurance product that best suits your needs and situation. If you have any questions or concerns, please feel free to contact us.

[0063] The large-scale language model 31 used in the talk generation server 20 is not particularly limited. The large-scale language model 31 can be a publicly known model as long as it is capable of input and output (generally equipped with an API), and may be implemented in another computer on the LAN 102, implemented in the talk generation server 20, or implemented in an external computer 30 accessible via the Internet 101 (that is, the sales support system 10 may or may not have the large-scale language model 31 as long as it can use the large-scale language model 31). The large-scale language model 31 is not limited by a learning dataset, but is preferably fine-tuned using a dataset for question answering tasks (for example, a dataset consisting of pairs of prompts and answer examples for creating sales talk examples based on predicted closing factors and attribute information). In addition, if the large-scale language model 31 is capable of setting parameters related to sentence generation, it is preferable to appropriately adjust the parameters (Temperature, Top_p, etc.).

[0064] (Case-Based Talk Generation Unit) The predictive talk generation unit 23 described above is based on prediction of specific closing factors as a general tendency, and does not consider specific similar cases, so it has the advantage of being able to output effective sales talk examples from the beginning of the system's use. However, if there are sufficient accumulated cases, there will be successful cases with high similarity, and considering the closing factors at that time is particularly effective for inexperienced salesmen. Therefore, as shown in FIG. 2, it is preferable to provide a closing storage unit 27 in which attribute information of closing customers who have succeeded in closing and closing factors n1 that caused the closing are stored in association with each customer, and a case-based talk generation unit 26 that generates a sales talk example t3 suitable for an uncontracted customer using a large-scale language model 31 based on the input attribute information i1 of an uncontracted customer and the closing factors n1 of a closing customer that have attribute information that matches or is similar to the input attribute information i1 of the uncontracted customer and that is read from the closing storage unit 27. The case-based talk generation unit 26 can function alone (i.e., without the predictive talk generation unit 23) if there are a sufficient number of accumulated cases in the closing storage unit 27.

[0065] (Deal Memory Section) The contract storage unit 27 stores attribute information of a customer who has successfully made a contract and the contract factor n1 associated with each customer. The attribute information stored in the contract storage unit 27 may be the same as the attribute information of the teacher data described above, as long as it includes at least age information and gender information associated with each customer, or may include only some items or additional items not included in the teacher data. The contract factors n1 stored in the contract storage unit 27 may be individually expressed sentences or options. For example, the contract factors n1 may include the above-mentioned concerns about the contract, decision points, and conversation tactics used, which may be obtained by interviewing or questionnaires with the customer, or the salesperson's thoughts may be recorded. The contract storage unit 27 may also be configured to store any information, such as personal information (such as name, address, and telephone number) of the customer who made the contract, and salesperson information, associated with the attribute information.

[0066] A known database application can be suitably used as the deal storage unit 27, but it may also be a set of files stored in a one-record, one-file format. For example, the deal storage unit 27 may be a database that stores records including text data and can output some or all of the records in response to a query. In particular, in terms of being able to quickly and easily search for similar cases, it is preferable that the deal storage unit 27 is a vector database that vectorizes and stores the text data of case information and efficiently searches for records based on the similarity of the vectors (cosine similarity, etc.). In this case, the character string data (chunks) for each customer who has made a deal, including attribute information and deal factors, is converted (embedded) into a high-dimensional vector representation by a tokenizer, and the vector is stored in the vector database, enabling the search for similar cases.

[0067] When the case-based talk generation unit 26 generates a sales talk example using the large-scale language model 31, it is theoretically possible to input all the information in the contract storage unit 27, the input attribute information i1 of uncontracted customers, and a command (a command to refer to the contract factor of a contracted customer having attribute information that matches or is similar to the input attribute information of the uncontracted customer) to the large-scale language model 31, but the more data there is, the slower the processing of the large-scale language model 31 becomes. Therefore, as shown in FIG. 4, it is preferable that the case-based talk generation unit 26 searches the contract storage unit 27 for the contract factor n1 of a contracted customer having attribute information that matches or is similar to the input attribute information i1 of the uncontracted customer, and generates a sales talk example t3 using the large-scale language model 31 based on the attribute information i1 of the uncontracted customer and the contract factor n1 searched for as having attribute information that matches or is similar to it.

[0068] When a vector database is used as the closing storage unit 27, the latter process is, for example, as follows. That is, the case-based talk generation unit 26 converts the input attribute information i1 of the unsold customer into a high-dimensional vector representation by a tokenizer, and instructs the vector database to search for a record of a vector similar to this vector. After acquiring a similar vector from the vector database, the case-based talk generation unit 26 converts the similar vector back into a character string by a tokenizer, and acquires the closing factor n1 of a closing customer having attribute information that matches or is similar to the input attribute information i1 of the unsold customer. At this time, the vector database can output only one record of the most similar vector, or output multiple vector records in descending order of similarity. After that, the case-based talk generation unit 26 acquires a sales talk example t3 output from the large-scale language model 31 by passing a prompt p3 for generating a sales talk example to the large-scale language model 31 together with the input attribute information i1 of the unsold customer and the closing factor n1 acquired from the vector database as having attribute information that matches or is similar thereto.

[0069] The rest is the same as in the case of the predictive talk generation unit 23. That is, the talk generation server 20 generates output screen information x1 including the sales talk example t3 output from the large-scale language model 31 (preferably including similar case information such as attribute information of similar cases used in generating the sales talk example t3 and closing factor n1) and transmits it to the terminal 40, and the terminal 40 that receives it can display the received output screen information x1 on the screen. At this time, if there are sales talk examples t1 and exchange speech examples t2 to be described later generated by the predictive talk generation unit 23 described above, the output screen information x1 including them can be generated and transmitted to the terminal 40, or the output screen information can be generated by dividing them and transmitted to the terminal 40 in sequence. Also, the case-based talk generation unit 26 can output multiple sales talk examples from the same predicted closing factor. Also, the case-based talk generation unit 26 can use a large-scale language model to generate the prompt p3.

[0070] As an example, when a prototype of the sales support system 10 is used and an example of customer attribute information (similar to the example in the predictive talk generation units 23 and 24) is input, similar cases retrieved by the case-based talk generation unit 26, their closing factors n1, a prompt p3 input to the LLM, and a generated sales talk example t3 are shown below. Note that the closing storage unit is a vector database, and one or more of the six concerns mentioned above are stored as closing factor concerns, and any sentence thought up by the salesman is stored as the closing factor decision point. <Example of prompt p3> "You are a top salesperson. Please output an appropriate sales pitch to propose an insurance plan to a customer with the following input attribute information who has the predicted closing factors (concerns and decision points) of similar cases 1 to 3 below." <Input attribute information> "Gender: Male, Age: 30s, Occupation: Company employee, Married: Married, Children: 2, Parents: Separated, Annual income: 5 million to 10 million, Disposable income: 5 million to 10 million, Years of employment: 10 years or more, Color of clothing: Monotone, dark, Pace: Slow, Expression: Calm, rich, smiling, Intonation: Intonation, high tone, Tone of speech: Questioning, asking questions" <Attribute information of Similar Case 1> "Gender: male. Age: 40s. Occupation: company employee. Married. Has one child. Parents live together. Annual income: 5 million to less than 10 million yen. Disposable income: 3 million to less than 5 million yen. Length of employment: 5 to less than 10 years. Clothing colors: monotone, dark. Walks at a slow pace. Expressions: calm, rich, smiling. Speech is calm and monotonous. Tone of speech: questioning, inquiring." <Factors behind the success of Similar Case 1: Concerns> "The concerns are about money, the need to consult with a third party, and the possibility that there may be more advantageous products or services available." <Similar Case 1: Decision Factors> Specifically, enthusiasm. <Attribute information of Similar Case 2> "Gender: male. Age: 50s. Married. Has four children. Annual income: between 5 million and 10 million yen. Disposable income: between 5 million and 10 million yen. Length of employment: over 10 years. Clothing colors: monotone and dark. Walks at a slow pace. Facial expression: stiff, cool, and stern. Inflection: modulated and high-pitched. Speaks in a questioning, interrogative tone." <Factors behind the success of similar case 2: points of concern> "The concerns are about money, the need to consult with a third party, timing, and the possibility of there being a more advantageous product or service." <Similar Case 2: Decision Factors> "I decided to tell them about the future that can be gained and the future that can be lost. Specifically, even if something were to happen, the family would be able to pay the expenses without being in trouble. In the case of a bank, there is a possibility of freezing the deposits." <Attribute information of Similar Case 3> "Gender: male. Age: 30s. Occupation: company employee. Married. Parents: living separately. Annual income: 5 million to less than 10 million yen. Disposable income: 3 million to less than 5 million yen. Length of employment: 10 years or more. Clothing: monotone, dark colors. Walks at a slow pace. Facial expression: calm, rich, smiling. Speech: plain, monotonous. Tone: questioning, inquiring." <Factors behind the success of similar case 3: points of concern> "The concerns are about money and the possibility of there being a more advantageous product or service." <Similar Case 3: Decision Factors> "Specifically, we will explain the details of the current insurance coverage and the maturity payments that will be received in the future." <Sales pitch example t3> “Dear customer, from what I have heard from you, you are a family with two children and parents who live apart. I understand that you have a stable job and income, but are feeling anxious about your future. First, regarding money, our insurance plans provide appropriate coverage in the unlikely event of an emergency, so that your family will not be in financial difficulty. Some plans also include maturity benefits that can be received in the future, so you can use them to meet financial challenges other than emergencies. Next, regarding concerns about the company or system, our insurance company has a long history of experience and not only proposes the best plan for our customers, but also provides comprehensive support and services for the future. We have a system in place that can flexibly respond to our customers' requests and needs, so you can use our services with peace of mind. From your lifestyle and facial expressions, we can see that you value your insurance and your future. We believe that by using our insurance plans, you will be able to protect your precious family and assets and build a brighter future. If you have any questions or concerns, please feel free to let us know. We will answer your questions sincerely.

[0071] The large-scale language model 31 used by the case-based talk generation unit 26 is not particularly limited, and may be the same as or different from the one used by the predictive talk generation units 23 and 24 as in the illustrated example. The tokenizer used for input / output of the vector database may be the same as or different from the one used for input / output of the large-scale language model 31.

[0072] Each time a salesperson makes a deal, or when a certain number of deals have been made, the salesperson can store the case information i2 including the attribute information and closing factors associated with each closing customer in the deal storage unit 27 as shown in FIG. 5. In the example shown in FIG. 1, the closing cases can be registered in the system as follows. That is, when the salesperson requests an input screen for recording closing cases on the terminal 40, the talk generation server 20 that receives the request generates input screen display data and transmits it to the terminal 40. The terminal 40 that receives the input screen display data displays an input screen on its display device. The salesperson operates the input device of the terminal 40 and inputs the case information i2 including the attribute information and closing factors of the closing customer on the input screen of the terminal 40. In these cases, the salesperson makes a selection in the selection type items, and inputs a sentence in the description type items, and then transmits the registration information including the attribute information and closing factors of the closing customer to the talk generation server 20 by clicking the decision button or the like. 2, the token generation server 20 has a case accumulation unit 28 that receives the case information i2 and passes it to the agreement storage unit 27 for storage, and this case accumulation unit 28 converts the case information i2 into an appropriate format as necessary and stores it in the agreement storage unit 27. The case accumulation unit 28 then generates output screen display data including the registration completion information and transmits it to the terminal 40. The terminal 40 that has received the output screen display data displays an output screen showing the registration completion information on its display device.

[0073] On the other hand, even if the decision point m2 can be predicted by the closing factor prediction model 22, it is not easy for an inexperienced salesman to imagine the sales talk at closing. Therefore, as shown in Fig. 2 and Fig. 6, it is preferable to provide a second closing factor prediction model 22 that predicts the decision point for closing, and a back-and-forth speech output unit 24 that outputs a back-and-forth speech example t2 that corresponds one-to-one to the predicted decision point m2 based on the predicted decision point m2 predicted by the second closing factor prediction model 22. This allows even an inexperienced salesman to use an effective back-and-forth speech example t2 at closing.

[0074] Examples of response discourse examples t2 corresponding to the eight decision point examples mentioned above are as follows:

[0075] Response example 1 Below is an example of a response tactic for decision point 1 (try it out) mentioned above. Customer: "I'm not sure if I can keep a pet because I'm not at home much." Salesman: "Okay, then why don't you try it out for a week?"

[0076] Response example 2 Below is an example of a rebuttal regarding decision point 2 (price) mentioned above. Customer: "I'd like to do this sales training, but it's too expensive." Salesman: "President, how many people would you like to take this course? 50 people. Understood. This is the cost of my training program. If you divide it by 50, that's the amount per person. The training is four hours long, so this is the amount per hour. If you invest this amount per hour in your employees and they turn into people who can sell, wouldn't that be a great return?" This is the case when the customer is satisfied with the price and a deal is made.

[0077] Response example 3 Below is an example of a response tactic for decision point 3 (educational advice) above. Salesman: "I've been in this industry for over 10 years now, so I know the difference between those who have been successful in increasing their sales and those who have difficulty increasing their sales. The truth is, those who get results and are successful have a different way of making decisions. When their soul trembles, when they think this is a good idea, they act according to that emotion. On the other hand, those who are unsuccessful are those who make excuses like "I don't have time" or "I have to ask for advice," block off their ears to the voice of their soul, and live their lives doing only what they can. So if your soul is trembling right now, why not give it a try?"

[0078] Response example 4 Below is an example of a response tactic for decision point 4 (third-party consultation) mentioned above. Customer: "I can't decide until I've had a discussion." Salesman: "Please feel free to consult with us. By the way, if you don't mind, please tell me who you plan to consult with." Customer: "My father." Salesman: "Your father. What do you think he would say if I asked him?" Customer: "They'll probably tell me to stop using stuff I don't understand." Salesman: "I see, I understand. By the way, what will you do if your father objects to the purchase?" Customer: "Probably not." This can then be followed by Response Example 3 or Response Example 6, which will be described later.

[0079] Response example 5 The following relates to decision point 5 (feasibility) above. Salesman: "Hey Mr. XX, I have a question. Are you unsure about paying 50,000 yen a month?" Customer: "No way, without a second thought." Salesman: "Thank you. How about 35,000 yen?" Customer: "Hmm, maybe it's still too hard." Salesman: "Thank you very much. How about 20,000 yen?" Customer: "That would be quite confusing." Salesman: "Thank you. Let's start with the conclusion. Throw out any plans that cost more than 35,000 yen. Doing so will probably just stress you out. So how about we create a plan that costs around 28,000 yen a month? What if that were available?" Customer: "Well, I guess I'll give it a try."

[0080] Response example 6 The following relates to decision point 6 (timing of life) mentioned above. Client: "I would like to do it, but if I talk to my family about it and they are against it, I will give up." Salesman: "You know it's a good idea to do it, so why give up now by consulting with people you know will be against it? Why not just give it a try, then report back to us after you've succeeded and be happy about it!"

[0081] Response example 7 The following relates to decision point 7 mentioned above (contrasting the future you gain with the future you lose). Customer: "I'm worried because I don't know if things will go well after I sign the contract." Salesman: "Okay. May I tell you about a customer I worked with some time ago who had the same problem?" Customer: "Yes." Salesman: "Although this person really wanted to adopt the system, his wife was against it and he couldn't go ahead with it. At the time, I thought it would be a bother for him, so I backed off. After a while, he contacted me to say that he was thinking about it again. However, at that time, he was in a situation where we couldn't recommend him to sign the contract due to compliance reasons."

[0082] Response example 8 The following relates to decision point 8 (third-party voice) mentioned above. Salesman: "Listening to your story, I remembered a case of a person exactly like you. I would like to tell you a story about a typical customer, in the hope that it might be of some help to you." [Industrial Applicability]

[0083] The present invention can be used in a sales support method, a sales support system, a trained model used therein, a method for generating the trained model, and a program for executing the trained model. [Explanation of symbols]

[0084] 10...sales support system, 20...talk generation server, 21, 22...sales factor prediction model, 21...first sales factor prediction model, 22...second sales factor prediction model, 23...predictive talk generation unit, 24...rebuttal speech output unit, 26...case utilization talk generation unit, 27...sales memory unit, 28...case accumulation unit, 31...large-scale language model, 40...terminal, 101-103...network, p1, p3...prompt, t1, t3...sales talk example, t2...rebuttal speech example.

Claims

1. A method for supporting sales by outputting examples of sales pitches suitable for uncontracted customers, the method comprising the steps of: The information processing system includes a contract factor prediction model that is machine-learned using training data so as to predict a contract factor correlated with attribute information of an uncontracted customer when the attribute information of the uncontracted customer is input, A step in which the contract factor prediction model predicts the contract factor from the input attribute information of the unconverted customers; the information processing system includes a step of generating a sales pitch example suitable for the uncontracted customer by using a large-scale language model based on the predicted contract factor of the uncontracted customer and the attribute information used for predicting the predicted contract factor; The attribute information includes age information and gender information. A sales support method comprising:

2. A sales support system that outputs examples of sales talks suitable for uncontracted customers, A sales factor prediction model that is machine-learned using training data so that when attribute information of unconverted customers is input, sales factors that correlate with the attribute information are predicted; a predictive speech generation unit that generates an example of a sales pitch suitable for a non-contracted customer by using a large-scale language model based on a predicted contract factor of the non-contracted customer predicted by the contract factor prediction model by inputting the attribute information on the non-contracted customer into the contract factor prediction model and the attribute information used for predicting the predicted contract factor, The attribute information includes age information and gender information. A sales support system characterized by:

3. The predictive talk generation unit outputs the sales talk example together with the predictive closing factors used to generate the sales talk example. The sales support system according to claim 2.

4. a contract storage unit in which attribute information of a contracted customer who has successfully concluded a contract and a contract factor that caused the contract are stored in association with each customer; and a case-based sales talk generation unit that generates sales talk examples suitable for the uncontracted customers using a large-scale language model based on the input attribute information of the uncontracted customers and the contract factors of the contracted customers having attribute information that matches or is similar to the input attribute information of the uncontracted customers, the case-based sales talk generation unit being read from the contract storage unit. The sales support system according to claim 2.

5. The example-based talk generation unit includes: a search is made for closing factors of closing customers having attribute information that matches or is similar to the input attribute information of unconcluded customers from the closing storage unit, and the sales pitch example is generated using a large-scale language model based on the attribute information of the unconcluded customers and the closing factors that have been searched for as having the matching or similar attribute information thereto; The sales support system according to claim 4.

6. The example-based talk generation unit outputs the generated sales talk example and the closing factors used to generate the sales talk example. The sales support system according to claim 5.

7. The contract factors include at least one of a contract concern and a contract decision point.

5. The sales support system according to claim 2 or 4.

8. The closing factors include a closing decision point, A response speech output unit outputs a response speech example that corresponds one-to-one to the predicted decision point predicted by the contract factor prediction model.

5. The sales support system according to claim 2 or 4.

9. A sales support program that causes an information processing system to output examples of sales pitches suitable for uncontracted customers, The information processing system includes a contract factor prediction model that is machine-learned using training data so as to predict a contract factor correlated with attribute information when the attribute information is input, The information processing system includes: A process in which the contract factor prediction model predicts the contract factor from the input attribute information of the uncontracted customers; and generating a sales pitch example suitable for the uncontracted customer using a large-scale language model based on the predicted contract factors of the uncontracted customer and the attribute information used for predicting the predicted contract factors. The attribute information includes age information and gender information. A sales support program characterized by:

10. When attribute information including age and gender information is input, the system is trained using training data to output a contract factor predicted from the attribute information. A trained model characterized by:

11. Using attribute information including age information and gender information, and training data including contract factors associated therewith, a contract factor prediction model is generated by machine learning, which outputs contract factors predicted from attribute information including age information and gender information when the attribute information including age information and gender information is input. A method for generating a trained model.

12. Information processing system, Attribute information including age information and gender information is input to the trained model according to claim 10, and a contract factor predicted from the attribute information is output. Execute the process, An execution program for a trained model.

Images