Patent Application
20150237206
1. Technical Field
The invention relates to on-line chat. More particularly, the invention relates to enhancing the customer experience by identifying chats that are redirected to other channels.
2. Description of the Background Art
When a customer is unable to solve a service or product problem using chat with an agent of the service or product provider, in frustration the customer may leave the chat and contact the service or product provider through another channel, such as by a voice call. Redirection or leakage of chats from one channel to other channel is common in a customer service environment. Unfortunately, an increase in the number of leakages can degrade the customer experience, chances of possible sales, and can also lead to customer dropout, where the customer gives up. The most common leakage of chats is leakage-to-voice chat.
The problem of leakage-to-voice from chats is commonly faced by operation managers and chat agents. This may lead to a degradation of the customer experience and productivity of sales team. Embodiments of the invention enhance the customer experience by detecting leakage-to-voice from chats and by providing recommendations to operations, chat agents, and customers. In embodiments of the invention a chat is classified into leakage-to-voice or leakage-to-text chat and actionable recommendations are then provided to operations, chat agents, and customers based on the leakage information. Once a leakage is identified, various other insights are extracted from chats and such insights are fed into a knowledge-base. Such insights also used in agent training and are provided to chat agents as recommendations. This results in a better customer experience.
A fundamental business objective is to provide customers with a level of customer support that meets their needs and expectations, while using the most cost effective techniques. Embodiments of the invention use model-based techniques to analyze the causes of leakage, i.e. leakage of chats from one channel to another channel, and to use the results of the analysis to make recommendations to agents and managers. An increase in the number of leakages is of concern to customer support organizations because such increase can lead to or cause a degradation of the customer experience.
Embodiments of the invention analyze leakages, i.e. transfers, of customer chats on one channel to another channels. Various devices and communications channels are used to establish a chat session between a customer, also referred to herein as a user, and a support agent. The contents of the chat are analyzed, for example, to make recommendations of goods and services to the customer. Otherwise, the customer's chat may be redirected to another channel, such as voice, resulting in leakage. The customer's information is confirmed and retained. A model is generated and used to analyze the chat information and the leakage information.
When a chat leakage is identified, the chat is analyzed to determine factors that may have contributed to the leakage. The data pertaining to the leakage and the analysis results are stored in a knowledge base. The information, and analysis thereof, stored in the knowledge base is then used for the training of agents and to make recommendations to agents and managers with the ultimate objective of improving the customer experience.
Embodiments of the invention enhance the customer experience by classifying chats into leakage-to-voice or leakage-to-text chats. Data about the leakage is used to provide recommendations, and specific training, to chat agents for such purposes as to solve customer queries efficiently, provide insight solutions, and enhance the customer experience.
Internet, a cellular-based communications network, a wireless communications network, a wire line communications network, a Global System for Mobile Communications (GSM) network, a combination thereof, or any other communications network.
The customer 11 uses a customer device to communicate with the chat agent over the communications network. In an example, the customer device can be any of a mobile phone, a handheld device, a tablet, a computer, a portable device, a communicator, or any other device that is capable of communicating with the communications network 13. The communications network 13 can use any suitable communication technology to communicate with the customer 11.
The system allows the customer 11 to access, for example, an application, a Web portal, or a website to communicate with the chat agent over the communications network. For example, when a customer is looking for a product, requires any type of assistance, or has a query, the customer can login or provide specific information to the Web portal to communicate with the chat agent. In various embodiments of the invention, the chat agent can be a voice chat agent, a text chat agent, a video chat agent, or any other chat agent capable of communicating with the customer.
The chat agent can provide insight solutions to customer issues or can redirect the chat to other channels to resolve customer issues. For example, a chat might proceed as follows:
Further, the chat agent can check the information related to the customer. Such information can include, for example, the customer's journey, the customer's communication history, the customer's interests, and any other information associated with the customer. For example, the customer's journey can refer to the chat agents, either a voice or a text chat agent, who interacted with the customer before the customer visited a specific chat agent; the path taken by the customer to reach the chat agent, such as by clicking on a link present on an internal or external site; the customer's Web-log journey; or any other customer information. The customer journey can be any of a virtual journey, a literal journey, an assisted journey, a guided journey, or a combination thereof.
Based on the information related to the customer, the chat agent provides insight solutions to the customer. For example, a chat might proceed as follows:
Or:
Furthermore, the chat agent can store information about the customer in a suitable location.
In an example, as shown in the
In an embodiment of the invention, a chat text 23 is used to build an anchor 24. Production data is actual customer interaction data that is saved in the database at the end of each day. An anchor comprises key phrases and/or words around which the line is centered. For example, extracting a few words around words such as “Department,” “team,” etc. from a text corpus yields a possible list of all the different types of departments and/or teams. Here, the words “Departments” and “team” are referred to as anchor words
In embodiments of the invention, the chat text 23 is the text provided by the chat agent 12, but the chat text could as well be provided by the customer. The channel names referred in the chat text are extracted to build the anchor 24.
One approach to temporary categorization involves the use of edit distance, which is a way of quantifying how dissimilar two strings, e.g. words, are to one another by counting the minimum number of operations required to transform one string into the other. In embodiments of the invention, edit-distance 25, i.e. surface similarity, is used to obtain a temporary categorization of the team/department using positive hits generated by the anchor building process. For purposes of the discussion herein, hits come from the text corpus on which the anchor building process is run. For example, a particular team/department may be grouped with the wrong team/department due to natural language variations, e.g. the ‘callback team’ is grouped in the ‘support team’ because the support team takes any incoming calls; the ‘Billing investigation team’ is grouped in the ‘billing team’ because bill investigation is about disputes and, hence, a separate team.
To address such problem, an initial grouping of some of the teams/departments is performed by business analysts using business sense for the teams/departments and customer needs. Embodiments of the invention provide a service layer, where business understanding is applied to correct and regroup the different departments and/or teams. In the above example, a domain expert corrects these groups manually.
In an embodiment of the invention, a service layer is applied to fine tune mapping and generate a model 22 that includes a final cat file for the team/department names. For purposes of the discussion herein, the cat file contains the categories into which the text corpus is categorized. The cat file can include, for example, a categories grouping of the team/department names based on the edit distance 25. Once the model 22 is generated, an information retrieval (IR) engine 25 can be used to categorize the chat data into the team/department names. In embodiments of the invention, the IR engine is an information retrieval system that includes a method of representing documents as transformed variables, querying them, ranking them based on computation of numerical scores, and retrieving the most relevant documents that match the query.
In embodiments of the invention, the system 10 also applies specific filters 27 for voice and text chat categorization. Generally, these filters 27 and the team/department names occur in a periphery of a predefined number of words. For example, most of the time the chat agent may say “I am transferring the chat to our customer solution team” or “please call our iPhone team.” Instead of taking the entire agent text, the voice and text chat filters 27 create dummy data having a periphery of the predefined number of words. The dummy data is provided with a dummy identifier, for example, [ID21111 1]. The dummy identifier marks the presence or absence of voice and/or chat filters. The identifier is later fed into the priority matrix. This in turn decides the channel of transfer.
In an embodiment of the invention, a priority index is provided to control each of the filters 27. The channel names are identified from anchors and by using a window of words around the anchor, which are referred to herein referred as filters. Different approaches may be used for building these anchors namely, using word contexts, noun extraction, or noun phrase extraction. The type of channel to which leakage has occurred, for example chat or voice, is identified by providing a window of ‘n’ words around the anchors to identify the filters, e.g. transfer, talk to, customer service rep etc.
Once the anchors and filters are identified, the exact channel is identified using a priority matrix. For instance, if a team name is present, i.e. the anchor is present, and voice filter is present, e.g. talk to, then the channel is a voice channel. If the team name is present and the chat filter is present, and the voice filter is also present, then this could possibly be a chat transfer.
The IR Engine 25 includes a grid editor to provide to the priority to the filters 27.
For example, if the voice filter is present and the chat includes a channel name in the classification, then a typical priority is obtained and the chat is most likely a voice transfer to that channel.
In
In an anchor process 33, a concordance of a concordance is a co-occurrence. In a window of ‘n’ preceding or succeeding anchor words, the system looks for co-occuring words The concordance can include direction (left window/right window/around window) and the window size. Based on the direction, positive hits are generated by the anchor building process. In a noun phrase extraction process 34, the pos tagger is used to tag the lines in the chat text based on the pos information. The first noun from the line of chat text is extracted based on the hypothesis that most of the time the product name is mentioned as the first noun.
A stopword removal process 35 is used to remove the stopword from the chat text. The stopword removal process involves removing uninteresting, non-informative, or irrelevant words from the chat text. These stopwords are extracted from a file or a database containing list of words maintained on the hard-drive.
Most of the time, the team/department names do not contain a stopword. A get unique word process 36 is used to remove multiple occurrences of the same team names. This is done by building a new unique set of words from a list, by removing duplicate entries of team names from the list, and using any algorithm for removing duplicates from a list and copying it to any other data structure, such as, set, dictionary, list, etc.
In a surface-similarity process 37, a surface similarity measure is used to determine an edit-distance at the word level or at the character level. The calculation of edit-distance or Levenshtein distance, is done by calculating the effort that is required to change a first word or character to a second word or character and by providing weights for each step change made to the first word or character, where each step change is any one of substitution, deletion, or addition of a character. Therefore, the edit distance or Levenshtein distance, provides a measure to calculate the similarity of any two words. Based on these similar words, team names are grouped in a category of the team/department.
In an embodiment of the invention, if the chat agent's names are also included as a variable, then the system can help evaluate the chat agents' performance. Agents refer customers to speak to a voice agent because they are not able to resolve the customer's issue at their end. This could either mean that the agent is not empowered to handle the query, or that the agent could not find an appropriate resolution and, hence, was asking the customer to call the call center. As a result of the analysis, embodiments of the invention can both identify the skill issue and the empowerment issue.
Leakage to voice results can be linked to agent and the agent performance and relative scoring of agents with the leakage to voice metric can be analyzed. In an embodiment of the invention, the chat agent performance data can be mapped with a customer satisfaction score (CSAT) driver and, as a result, provide actionable recommendations to the operations and the chat agents (see commonly assigned U.S. patent application publication no. 2010/0262,549,
System and Method for Customer Requests and Contact Management, which application is incorporated herein in its entirety by this reference thereto).
Embodiments of the invention can pass contextual information to the voice channel. For purposes of the discussion herein, contextual information is information, such as the plan and/or product that the customer talked about, so that the voice call can resume from that point onwards. For example, a customer X was asking about that iPhone5 in chat and the customer is referred to voice call for an upgrade. As soon as the voice agent picks up the call, the agent can say “How can I help you with iPhone5 upgrade request?” rather than the usual “How can I help you?”
A key aspect of the invention involves passing the insights from one channel to the next channel. If there is a dropout in chat and, for some reason the chat agent is not able to solve the customer problem, then the system can detect the leakage and extract substantially all of the contextual and other relevant information from that chat interaction. The relevant information is summarized and passed to the voice referrals. This removes the need of repeating the process to collect the information and allows the chat agents to communicate intuitively with the customers, thereby resulting in better customer engagement, reduction in drop-offs, and an enhanced overall customer experience.
In an embodiment of the invention, once the leakage in the chat is detected it can be added to a Web-log journey to predict the leakage at the start of the customer journey. A predictive model is built with voice leakage as the response variable and the independent variables are the Web journey, customer historic interactions, and CRM data. A machine learning model can be built to predict whether the customer is likely to end up in an issue that would be referred to voice. Once the likelihood of the customer being referred is high, the customer should not be offered a chat invite or it should be stated upfront, e.g. in case of a button chat, that the issue may not be handled in chat Thus, recommendations can be provided to the chat agents at the start of the chat, based on this detected information. This leads to better engagement of the customer and less leakage of the chats to voice channels, which is relatively costly.
The foregoing description of the specific embodiments can include enhancements of anchor the building process. For example, the grouping of substantially all of the hits in the team category can be performed in other or more intelligent ways, rather than only using the edit distance, including but not limited to, cosine similarity, Levenshtein, Google distance, Bing distance, semantic graph distance, hamming distance, Jaccard, etc.
In
In
The embodiments disclosed herein can be implemented through at least one software program running on at least one hardware device and performing network management functions to control the network elements. The network elements shown in the figures include blocks which can be at least one of a hardware device, or a combination of hardware device and software module.
The computer system 1600 includes a processor 1602, a main memory 1604 and a static memory 1606, which communicate with each other via a bus 1608. The computer system 1600 may further include a display unit 1610, for example, a liquid crystal display (LCD). The computer system 1600 also includes an alphanumeric input device 1612, for example, a keyboard; a cursor control device 1614, for example, a mouse; a disk drive unit 1616, a signal generation device 1618, for example, a speaker, and a network interface device 1628.
The disk drive unit 1616 includes a machine-readable medium 1624 on which is stored a set of executable instructions, i.e. software, 1626 embodying any one, or all, of the methodologies described herein below. The software 1626 is also shown to reside, completely or at least partially, within the main memory 1604 and/or within the processor 1602. The software 1626 may further be transmitted or received over a network 1630 by means of a network interface device 1628.
In contrast to the system 1600 discussed above, a different embodiment uses logic circuitry instead of computer-executed instructions to implement processing entities. Other alternatives include a digital signal processing chip (DSP), discrete circuitry (such as resistors, capacitors, diodes, inductors, and transistors), field programmable gate array (FPGA), programmable logic array (PLA), programmable logic device (PLD), and the like.
It is to be understood that embodiments may be used as or to support software programs or software modules executed upon some form of processing core (such as the CPU of a computer) or otherwise implemented or realized upon or within a machine or computer readable medium. A machine-readable medium includes any mechanism for storing or transmitting information in a form readable by a machine, e.g. a computer. For example, a machine readable medium includes read-only memory (ROM); random access memory (RAM); magnetic disk storage media; optical storage media; flash memory devices; electrical, optical, acoustical or other form of propagated signals, for example, carrier waves, infrared signals, digital signals, etc.; or any other type of media suitable for storing or transmitting information.
Although the invention is described herein with reference to the preferred embodiment, one skilled in the art will readily appreciate that other applications may be substituted for those set forth herein without departing from the spirit and scope of the present invention. Accordingly, the invention should only be limited by the Claims included below.
This application is a continuation of U.S. patent application Ser. No. 14/149,768, filed Jan. 7, 2014, which claims priority to U.S. provisional patent application Ser. No. 61/750,216, filed Jan. 8, 2013, each of which application is incorporated herein in its entirety by this reference thereto.
| Number | Date | Country | |
|---|---|---|---|
| 61750216 | Jan 2013 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14149768 | Jan 2014 | US |
| Child | 14698709 | US |
