WO2020111963A1 - Method, system and device for routing client messages in communications with a contact centre - Google Patents

Abstract

The claimed solution relates to the field of data processing, in particular to a method, system and device for routing client messages in communications with a contact centre. A method for routing client messages in communications with a contact centre is implemented in steps in which: at least one client message is received in a communication channel; the received message is processed, during which message attributes are determined; on the basis of the obtained attributes at least one subject matter of the client message is determined; messages are allocated on the basis of the determined subject matter of the client message and on the basis of the skills of operators for processing the respective messages; and said client is connected to an operator on the basis of the determined skills required for processing the message.

Description

METHOD, SYSTEM AND DEVICE FOR ROUTING CUSTOMER'S APPEALS IN CONNECTION WITH CONTACT CENTER

FIELD OF TECHNOLOGY

[0001] The claimed solution relates to the field of data processing, in particular, to a method, system and device for routing customer requests in connection with a contact center.

BACKGROUND

[0002] At the moment, there are many mechanisms for routing customer calls, using for this both attributes of the call itself and the characteristics of various contact center operators.

[0003] Such solutions are disclosed, for example, in patent documents US6587556, US6611590, US6493695. Solution US6493695 is the closest in essence to the claimed solution and describes a system for routing customer requests depending on the skills of the contact center operator.

[0004] A disadvantage of the known approaches is the lack of tight integration with the automated information system that the client uses before contacting in order to automatically determine the topic and nature of the appeal, which allows more accurately and as soon as possible to process the client’s appeal in the most relevant way.

SUMMARY OF THE INVENTION

[0005] The claimed solution is aimed at eliminating the existing drawbacks in the prior art and is an automated system and method of its operation for processing customer requests from various communication channels, determining call attributes and subsequent routing of calls using the machine learning algorithm for contact center operators, who have the necessary skills to handle specific customer requests.

[0006] The technical effect in the implementation of the claimed solution is to ensure the process of automated processing of multiple calls from various communication channels and their subsequent redirection based on certain attributes of calls to relevant operators with the required set of skills for the speedy processing of incoming calls. [0007] In a preferred embodiment, a method for routing customer requests when communicating with a contact center is claimed, comprising the steps of:

receive at least one client’s request in the communication channel;

process the received appeal, during which the attributes of the appeal are determined;

based on the obtained attributes, at least one subject of the client’s appeal is determined;

perform distribution of appeals based on a specific subject of the client’s appeal and the skills of operators to process the relevant requests; and

initiate the connection of the said client to the operator, based on certain skills required to process the appeal.

[0008] In one particular embodiment of the method, the communication channel is an online channel or a telephone channel.

[0009] In another private embodiment of the method, the online channel is a website, mobile application, social network, email, or messenger.

[0010] In another particular embodiment of the method, clients are additionally identified on the communication channel.

[OOP] In another private embodiment of the method for identified customers, the presence of a history of search queries for goods and / or services is additionally checked.

[0012] In another particular embodiment of the method, keywords are identified based on the appeal attributes that identify the request in the client’s appeal.

[0013] In another particular embodiment of the method, a semantic analysis of the appeal is performed.

[0014] In another particular embodiment of the method, the determination of the available operators is carried out using an analysis module containing a machine learning model that executes a classification algorithm.

[0015] In another particular embodiment of the method, the classification algorithm is one or more neural networks, a gradient boosting algorithm, a genetic algorithm, or a random forest.

[0016] In another particular embodiment of the method, the algorithm is trained on a sample based on customer requests. [0017] In another particular embodiment of the method, when processing a client’s request, the machine learning algorithm analyzes the skills of free operators, the time to wait for the request in the request queue, and the required operator skills to solve the request.

[0018] In another particular embodiment of the method, when a call arrives in the channel of a website or mobile application, the location coordinates of the client are additionally used.

[0019] In another particular embodiment of the method, when initiating a connection to an operator, the client selects an interaction channel.

[0020] In another particular embodiment of the method, the operator is a chatbot or telephone bot.

[0021] The claimed solution is also carried out using a customer call routing system in communication with a contact center, which comprises: a user device forming a customer call in a communication channel; a customer identification module that analyzes the appeal for the content of information about registered customers;

call classification module

processing the received appeal to determine the attributes of the appeal;

determination of at least one subject of the client's appeal;

a module for collecting data about the client’s context, which analyzes and supplements data on the client’s contact with information about the subject of the appeal based on client’s history data for the product and / or service;

call distribution module executing

distribution of requests based on a specific subject of the client’s appeal and the skills of operators to process the relevant requests;

initiating the connection of the said client to the operator, based on certain skills required to process the appeal.

[0022] In a particular embodiment of the system, the call distribution module executes a machine learning algorithm implementing the classification algorithm.

[0023] In another particular embodiment of the system, the algorithm is one or more neural networks, a gradient boosting algorithm, a genetic algorithm, or a random forest.

[0024] In another particular embodiment of the system, the module performs semantic processing of customer requests. [0025] In another particular embodiment of the system, when processing a client’s request, the machine learning algorithm analyzes the skills of free operators, the time to wait for the request in the request queue, and the required operator skills to solve the request.

[0026] The claimed solution is also carried out using a customer call routing device in communication with a contact center, which contains at least one processor and at least one memory that stores machine-readable instructions that, when executed by the processor, implement the above method.

BLUEPRINTS

[0027] FIG. 1A illustrates a general view of the interaction of elements of a claimed solution.

[0028] FIG. 1B illustrates the process of processing customer requests through an automated system.

[0029] FIG. 2A - FIG. 2B illustrate the learning process of a machine learning model.

[0030] FIG. 3 illustrates the sequence of steps for processing customer requests according to the claimed method.

[0031] FIG. 4 illustrates an example device for implementing the claimed solution.

DETAILED DESCRIPTION OF THE INVENTION

[0032] As shown in FIG. 1A, the general scheme (100) of the objects involved in the implementation of the presented solution includes a number of user devices (software) that are connected via a communication channel (130) to a contact center (120) to process their calls with the help of operators (150). The work of the contact center (120) is coordinated using a server (140) containing an algorithm for processing incoming customer requests (110).

[0033] The communication channel (130) may be a telephone line or an online data channel. As an online channel, a website, a mobile application, instant messengers, social networks or any other communication principle that can be used to transmit information to the contact center via the Internet (120) can be used. Online data channel for communication via the Internet can be implemented on the basis of a wide range of protocols, for example, TSRLR, 802.11, GSM / 3G / 4G / 5G, etc.

[0034] The client device (110) may be a personal computer, laptop, cell phone, smartphone, tablet, game console, smart TV, or the like.

[0035] The server (140) is the main computing node that stores the necessary data for implementing the algorithm for processing client requests (110). Server (140) also contains a database that stores information about registered customers (140) who have purchased certain goods or services, their registration information and data browsing information on the Internet. The server database (140) also stores operator skill data (150), which is constantly updated to process incoming customer requests (110).

[0036] In FIG. 1B shows the details of the modules included in the processing system (160) of customer requests (110), which is stored on the server (140). This system (160) contains a customer identification module (161), a case classification module (162), a user context data collection module (163), a case handling evaluation module (164) and a case distribution module (165).

[0037] Client identification module (161) Allows you to identify a client from a common client base by the characteristics of the calls available in a particular channel, which uses the attributes for each channel from which calls are received.

[0038] The user appeal classification module (162) allows to add information about the appeal with attributes based on the appeal characteristics and the identified user (111-113). Traditionally, classification is based on user-side solutions (IVR, various telephones or E-mail addresses for various issues). The proposed solution allows the use of a single point of contact for users (110).

[0039] The user context data collection module (163) allows you to supplement the information about the call with information about the subject of the call based on the history of the client’s interaction (111-113) with the products. For example, the technology for tracking customer actions (111-113) can be applied on a website or in a mobile application when purchasing a product or service. Such technology may be, for example, clickstream.

[0040] The user handling processing evaluation module (164) of the user (111-113) allows several characteristics of the handling processing to be reduced to a binary value, desirable / not desirable. Reduction to a binary value allows you to use machine learning algorithms for predicting the results of future processing of such calls.

[0041] The distribution module of calls (165) uses a poorly structured set of information about the call, static information about the skills of the operators (160) and dynamic information about the employment of the operators (160) to decide whether to transfer the call to a specific operator (151-153).

[0042] In FIG. 2A - FIG. 2B presents an example of the learning process of a machine learning model for the automated distribution of customer requests (110).

[0043] A training set is formed as follows. At the first stage (201), data is collected on customer requests (software). Calls can be presented in text format, as a record of a conversation between an operator and a client (111-113), in the case of a direct telephone call to the contact center, or a log of communication in text mode, for example, chat sessions.

[0044] Next, for each appeal, its attributes are allocated (202). In particular, the attributes can be keywords that describe the purpose of the appeal, for example, the client’s phrase - “How to change train tickets?”. Attributes will be the terms “change” and “train”. Key terms describe the desired action, which is the subject of a client’s appeal (110).

[0045] Next, enrichment of the access data is performed. Data enrichment is carried out depending on the availability of previously stored information on the client’s request, information about which was added to the training set. Such information may be client identification information (110), for example, registration data, phone number, social network account data, etc., history of open spaces and / or interaction with previously purchased products or services, etc. The attribute of available information is taken into account in the characteristics of the appeal.

[0046] If the client has been identified, then a request is made for data from the CRM system. Information from CRM is a list of Web pages and screens of mobile applications with which the client interacted, indicating the transition time to the page / screen and the key attributes with which the data of the page / screens is associated. These attributes are determined by the subject area of the product, which is supported by the contact center. For example, for information pages about a package tour, the key attributes are the dates of the tour, resort / country, hotel, duration of the tour. For the page from the help section "How to exchange a ticket?" the key attribute is "exchange." [0047] Next, the routing data of the call to the particular contact center operator (203) and the nature of their processing are taken into account. In particular, whether switching to another operator, processing time, number of operator phrases, etc., was performed.

[0048] Next, the quality assessment (204), formed by the client (110) based on the communication with the operator (150), as well as on the basis of information received from the quality control service, is taken into account.

[0049] Based on the totality of the above data, a training vector is generated for the machine learning model (205). This approach is repeated for each client call (110) from the test sample.

[0050] Also, based on the parameters of each appeal, such as: customer’s assessment, quality control service’s assessment, processing time, number of client redirects, a binary assessment is generated that defines the quality parameter of the processing of the appeal. Based on the generated vectors and the quality parameter of treatment, a training sample is created for the machine learning model on which its training is performed.

[0051] The machine learning model is stored on a server (140) and is associated with a call processing system (160). A model can be one or more neural networks, a random forest algorithm, gradient boosting, or a genetic algorithm. The machine learning model provides the implementation of the forecasting (prediction) process of the probability of successful handling of treatment with certain attributes by an operator (150) having a given set of skills for a given time period.

[0052] Generation of a binary estimate.

[0053] The essence of the mechanism is to cast the characteristics of the processed call to the value 0 (processed in an undesirable way) or 1 (processed in a desirable way).

[0054] Algorithm Input: Customer Evaluation C (integer from 1 to 5), Quality Control Department Evaluation Q (integer from 1 to 5), call processing time T (number of seconds from contacting the client to closing the application) and number conversions between operators R. The evaluation algorithm is as follows:

[0055] If C <3, then the algorithm ends with a result of 0.

[0056] If Q <3, then the algorithm terminates with a result of 0.

[0057] If R> 1, then the algorithm terminates with a result of 0. [0058] The score is calculated for the processing time of the St. If circulation falls into the quarter of the fastest processed in the sample, then St = 4. In 50% of the fastest processed, St = 3, in 75% of St = 2. For all the others, St = 1.

[0059] The total score for processing the treatment S - (C + Q - 6) + (1 - R) * 4 + St. is calculated.

[0060] If S> 6, then the algorithm terminates with a result of 1, otherwise c ends with a result of 0.

[0061] As shown in FIG. 3, the work of the claimed method (300) for routing customer requests (software) is as follows. The calls arriving at the contact center (120) are aggregated and form a queue for their processing, which is controlled by the server part (140) and the call processing system (160) (step 301).

[0062] The calls received in step (301) are processed by the system (160) for recognition of the subject of the call. If the call comes in via telephone, a speech analysis algorithm can be used to translate voice information into text and perform subsequent semantic analysis. Upon receipt of an appeal from an online channel in the form of a text message or other machine-readable form, its semantic analysis is also performed.

[0063] In step (301), an additional determination of the client (110) generating the call may also be performed. Upon receipt of the appeal, the user is authenticated. For calls received from online chats built into the Web site, the authenticated user login is used. For calls from a mobile application, the user login is also used. For calls from messengers of a social network, a login on a social network is used and an account is searched with an associated social network login. When contacting by e-mail, the sender’s e-mail address is used and the account is checked with the same e-mail address. To contact by phone, a phone number is used and an account is searched with a matching phone.

[0064] If the client’s account (110) is found, then one or more found accounts are attached to the appeal. Next, a search is made for hits with similar contact data in the hit history database stored on the server (140). If previous hits with the same contact details and linked account information are found, then the same account information with the note “found in history” applies to the current hit. If the account is not found, then the appeal is not no identification data is added and its further processing is carried out as usual.

[0065] In step (302), the access attributes are determined. If this is a text call, the text is analyzed for keywords. Method (300) allows you to specify an unlimited set of attributes and related keywords. For example, for the attribute “desired action” there may be options “exchange => (exchange, exchanged, exchanged, exchange) | return => (return, refund, returned, returned). " If there is a match, the corresponding value is written to the attribute. For the text “I want to exchange or return a ticket”, the attribute “desired action” will have a pair of values “exchange”, “return”.

[0066] If this is an appeal from a mobile version of a site or mobile application and the availability of user coordinates, a check is made with the base of geographical objects at airports, railway stations, bus stations, taxis, etc. If there are matches, the attribute “urgent” and “now located in” are added with the value and type of geographical object. This function, as a rule, is implemented when a client (110) addresses through a mobile application that allows you to take into account the geolocation (geographical coordinates) of a client (110). According to geolocation data, one or more points of interest (POIs) can also be determined, which can also play the role of attributes in processing a call.

[0067] By the identified user, a request is made to the server order database (140). If orders are found, then the attribute “buyer” is assigned to such a call and the attribute “type of order” is added in accordance with the type of order with the nearest date of dispatch. If an order with a shipment within the next day is found, the attribute “urgent” is assigned.

[0068] The next step of the method (300) is to enrich the access data (step 303) with client context data (110), for which an analysis of the history of user interaction with the product is performed. The term “product” refers to a Web portal, Web site or mobile application. For example, the Web portal "package tours" or the iOS application "Flights".

[0069] In step (303), data from the CRM system is obtained from the customer identification (software) information. Information from CRM is a list of Web pages and screens of mobile applications with which the client (software) interacted, indicating the time to go to the page / screen and the key attributes with which these pages / screens are associated. These attributes are determined by the subject area of the product, which is supported by the contact center. For example, for Packet tour information pages with key attributes are tour dates, resort / country, hotel, tour duration. For the page from the help section "How to exchange a ticket?" the key attribute is "exchange."

[0070] Next, at step (304), a determination is made of the available contact center agents (150) and their skill set for processing incoming calls. Priorities for handling a call are those that contain the attribute "urgent."

[0071] The distribution of calls by operators (150) in step (305) is performed using a machine learning model that takes into account the set of attributes of each of the calls and the skill set of each of the operators (150) by analyzing the relevance and processing speed of each of the calls.

[0072] Call Routing Mechanism

Suppose that in the contact center (120) the set of skills of operators (150) S consisting of m elements is defined. Each of the n operators (151-153) corresponds to a vector X of dimension m, where each element corresponds to the skill development of the operator (151-153). Each call is described by the vector Y. Also, the function P (X, Y, t) is defined in the contact center (120), which predicts the remaining processing time for the call by Y by the operator (151-153) with skills X and the time t already spent on processing the call.

[0073] Based on the training sample obtained when the contact center (120) works with any starting routing mechanism, a binary classifier B (X, Y, Yu) is constructed that predicts the probability of the desired processing by the operator (151-153) with X handling skills X at a waiting time of Yu. Any algorithm of two-class classification (decisive forest, boosting, genetic algorithms, neural networks) can be used to construct it.

[0074] The current state of the agent (151-153) in the contact center (120) is determined by the vector Z, which consists of his skills X, the characteristics of the job Y, and the time t already spent on processing. Also, at any given time, there are many U calls that require processing to start (calls on the line, letters in the queue, initiated chats). The task of the routing mechanism is to form the distribution of the set of calls U to the operators (150) in such a way as to maximize the average probability of the desired call processing over the whole set U.

[0075] A genetic algorithm is used to form the distribution. A particular is the specific distribution M of the set of calls U to n operators (150) with the current state of Zi. That is, for each operator (151-153), a sequence of 0 or more calls from U is defined so that all calls from U are distributed.

[0076] An individual's viability is the average probability of the desired treatment of calls across the entire set U.

[0077] The calculation of the probability of the desired processing of a specific treatment Y.

In a specific distribution M, the call Y is assigned to the operator (151-153) with skills X. In general, this operator (151-153) already has an active call Y0, which it already processes time Yu, and the call Y in the specific distribution M is preceded by sequence W of 0 or more other hits.

[0078] The prediction algorithm calculates the remaining time tw for processing the current call, tw = P (X, Y0, 10). If the operator (151-153) is already free, then tw = 0. For each call K from W, the processing time is calculated as P (X, K, 0) and the result is added to tw. The probability of the desired treatment Y is calculated as B (X, Y, tw). The average probability is defined as the arithmetic mean between the probabilities of the desired processing of each particular call, calculated according to the algorithm above.

[0079] The mutation operation of an individual is defined as the removal of one random treatment from the distribution and its location in another randomly selected distribution location. Crossing two individuals - the distribution of the first half of calls from U is taken from one parent, and the second half from the second parent. Crossbreeding occurs between randomly selected parents.

[0080] The selection operation is the survival of half a population with high viability rates. The initial population is created by randomly distributing all calls by operators (150).

[0081] The genetic algorithm finishes its work either by timeout (1 second) or when saturation is reached (when, after a cycle, the best result improves slightly).

[0082] After creating the optimized distribution, all calls distributed to free operators (150) go into processing and are removed from the set U. The algorithm repeats its work every second.

[0083] In this case, when routing calls to a particular operator (151-153), the type of client’s call channel (110) can be taken into account. In particular, customers (110) can clarify the preferred principle of communication and the type of interaction with the operator (150). For example, a customer (111) who made a call via telephone can initiate a switch to an online channel, for example, via a Skype call or switch to an automated contact center support system (telephone robot), which will allow you to process a specific call.

[0084] A chat bot can also act as an operator (150), which is advisable if the client’s appeal contains attributes that relate to frequently generated calls and which do not require direct interaction with a live operator (150). Interaction with operators (150) can be carried out on a web site, for example, a chat session or VoIP communication, in a mobile application, telephone communication, using augmented reality tools, etc.

[0085] In FIG. 4 shows an example implementation of a computing device (400), on the basis of which a user device (110), a server part (140), and other devices that can participate in information exchange to implement the claimed solution can be made.

[0086] In the General case, as mentioned above, the device (400) can be selected from a wide range of electronic devices known from the prior art. In the General case, the device (400) contains one or more processors (401) or one or more microcontrollers, RAM (402), means of permanent data storage (403), input / output interfaces (404), input / output devices (405), means of network interaction (406).

[0087] The processor (401) is a basic computing module that provides logical processing of the algorithmic instructions necessary for the device (400) to perform the necessary functions.

[0088] RAM (402) is a standard random access memory designed to store instructions executed by the processor that implement the work of the embedded program logic.

[0089] Means for permanently storing data (403) may include, but are not limited to: a hard disk (HDD), flash memory (NAND, EEPROM, SD cards, etc.), a solid state drive (SSD), optical drives data (CD / DVD / BlueRay discs, etc.).

[0090] I / O Interfaces (404) may include, but are not limited to: ADC / DAC, USB (micro-, Tour C, mini-, etc.), PS / 2, PCI, VGA, RS232, RJ45, FireWire, SATA, IDE, COM, LPT, Audio Jack, HDMI, Display Port, Lightning, etc.

[0091] I / O facilities (405) may include, but are not limited to: a display, a touch screen, a keyboard (mechanical, touch, projection, etc.), a trackball, joystick, touch pad, speakers, microphone, projector, indicator light, buzzer, biometric sensor (fingerprint, retina, iris, voice, palm, vein pattern, etc. scanner), camera, optical sensor, accelerometer, gyroscope, light sensor, proximity sensor, gravity sensor, etc.

[0092] The network communication tool (406) may include, but is not limited to: Bluetooth module, BLE module, NFC, Ethernet card, modem, router, IrDa, GSM modem, GPRS modem, LTE modem, SG modem, WLAN, Wi-Fi module, satellite modem, GNSS receiver, etc.

[0093] The examples of application of the claimed technical solution are not limited to a certain spectrum of a particular sphere of economic activity, since this solution can be used to sell various goods and / or services, for example, in the field of tourism, retail or wholesale, media business, pharmaceuticals, consulting services, etc.

[0094] The presented description of the claimed solution discloses only preferred examples of its implementation and should not be construed as limiting other, specific examples of its implementation, not going beyond the scope of legal protection, which are obvious to a specialist in the relevant field of technology.

Claims

FORMULA 1. A method for routing customer requests when communicating with a contact center, comprising the steps of: - receive at least one client’s request in the communication channel; - carry out processing of the received appeal, during which the attributes of the appeal are determined; - on the basis of the received attributes determine at least one subject of the client's appeal; - perform the distribution of appeals based on a specific subject of the client’s appeal and the skills of operators to process the relevant requests; and - initiate the connection of the client to the operator, based on certain skills required to process the application. 2. The method according to claim 1, characterized in that the communication channel is an online channel or a telephone channel. 3. The method according to claim 2, characterized in that the online channel is a website, mobile application, social network, email or messenger. 4. The method according to claim 3, characterized in that it further identifies customers in the communication channel. 5. The method according to claim 4, characterized in that for identified customers, the presence of a history of search queries for goods and / or services is additionally checked. 6. The method according to claim 1, characterized in that, based on the attributes of the appeal, keywords are identified that identify the request in the client’s appeal. 7. The method according to claim 6, characterized in that a semantic analysis of the treatment is performed. 8. The method according to claim 1, characterized in that the determination of available operators is carried out using an analysis module containing a machine learning model that performs a classification algorithm. 9. The method according to claim 8, characterized in that the classifying algorithm is one or more neural networks, a gradient boosting algorithm, a genetic algorithm or a random forest. 10. The method according to claim 9, characterized in that the algorithm is trained on a sample based on customer requests. 11. The method according to claim 10, characterized in that, when processing a client’s request, the machine learning algorithm analyzes the skills of free operators, the time to wait for the request in the request queue and the required operator skills to solve the request. 12. The method according to claim 4, characterized in that upon receipt of the appeal in the channel of the website or mobile application, the coordinates of the client’s location are additionally used. 13. The method according to claim 1, characterized in that when initiating a connection to the operator, the client selects the interaction channel. 14. The method according to claim 1, characterized in that the operator is a chat bot or telephone robot. 15. A system for routing customer requests when communicating with a contact center, comprising: a user device forming a client’s appeal in a communication channel; a customer identification module that analyzes the appeal for the content of information about registered customers; a call classification module that processes the received call to determine the call attributes; determination of at least one subject of the client's appeal; a module for collecting data about the client’s context, which analyzes and supplements data on the client’s contact with information about the subject of the appeal based on client’s history data for the product and / or service; a distribution module for calls, which distributes calls based on a specific subject of a client’s request and the skills of operators to process the corresponding calls; initiating the connection of the said client to the operator, based on certain skills required to process the appeal. 16. The system according to clause 15, characterized in that the distribution module calls performs the machine learning algorithm that implements the classification algorithm. 17. The system according to clause 16, characterized in that the algorithm is one or more neural networks, gradient boosting algorithm, genetic algorithm or random forest. 18. The system of claim 15, characterized in that the module performs semantic processing of customer requests. 19. The system according to 17, characterized in that when processing a client’s request, the machine learning algorithm analyzes the skills of free operators, the time to wait for the request in the call queue and the required operator skills to solve the request. 20. A device for routing customer requests when communicating with a contact center, containing at least one processor and at least one memory that stores machine-readable instructions that, when executed by the processor, implement the method according to any one of paragraphs. 1-14.