DESC:FIELD OF THE INVENTION
[001] The present invention relates to the field of data processing and more particularly to a system for analyzing and predicting customer dissatisfaction and a method therefor.

BACKGROUND OF THE INVENTION
[002] In conventional customer churn or complaint prediction models developed for a specific business, to determine a customer’s characteristics and preferences, a customer’s data is typically modelled as a static n-dimensional feature vector where each feature represents a structured information available about the customer. For instance, the demographic details or count/frequency of an issue faced by the customer, or a transaction done, etc. are typical examples of structured information. The traditional models manually feature engineering to make predictions. Hence, the predictions are many a times inaccurate or flawed.
[003] There are also existing computer systems or techniques for modelling time-series data of a costumer such as transactional history which theoretically are capable of exploiting the temporal information. However, many empirical results indicate that such models are extremely difficult to train and have been unable to capture long term history of the customer or the customer’s experience.
[004] There are also existing computer systems or techniques for modelling time-series data of a costumer such as transactional history which theoretically can exploit the temporal information. However, many empirical results indicate that such models are extremely difficult to train and have been unable to capture long term history of the customer or the customer’s experience.
[005] It has been observed that the existing methods like filtering customers based on fixed rules created using business knowledge or model based on static feature priority are less accurate in predicting customer dissatisfaction due to the subjective manual interpretation of the customer history. Whenever a large amount of input features or data are fed as the training data, the existing systems find it hard to accurately represent interactions that happened long back.
[006] Though existing systems and models are capable to learn during training how much weightage should be given to each characteristic or feature in the training data, but these weightages always stay constant during prediction for different customers and hence feature priority for all the users becomes static. Accordingly, the existing systems are not capable to dynamically predict importance of some of the features in individual cases and hence provide less accurate predictions.
[007] Moreover, it has also been observed that the models which tend to give higher accuracy are unable to show to a user which event or historical experience of the customer had the most impact on the customer’s decision to not continue with the services of a particular business. Owing to the above limitations, the existing models are unable to predict with a high accuracy as to the likelihood of a customer complaining or leaving a particular business’ services.
[008] Thus, there exists a need for an improved system that analyzes and predicts customer dissatisfaction, thereby mitigating the hereinabove mentioned drawbacks.

SUMMARY OF THE INVENTION
[009] In an embodiment, the present invention provides a system for analyzing and predicting customer dissatisfaction. The system comprises a first data storage for storing a customer history data and an aggregator server connected with the first data storage and configured to aggregate the customer history data. The system has a management server connected with the aggregator server, the management server having a processor configured to implement a deep learning model for processing the customer history data and generating a processed data. The system includes an inference server connected with the management server; the inference server is configured to predict the customer dissatisfaction threshold and key reasons impacting customer experience for each customer based on the processed customer history data.
[010] In an embodiment of the invention, the first data storage includes a static data storage for storing the demographic details of the customers, a sequential structured data storage for storing the customer interaction history; and an unstructured data storage for storing the data in the form of the customer experiences including complaints or feedbacks provided by the customer.
[011] In a further embodiment of the invention, the system includes a second data storage for storing the processed customer history data. The inference server is connected with the second data storage of the processed customer history data.
[012] In a further embodiment of the invention, the inference server has a prediction module configured to identify and visualize specific issues with the customer attributable to leave a particular business. The system includes a user device to display the customer dissatisfaction threshold and specific issues with the customer attributable to leave a particular business.
[013] In a further embodiment of the invention, the deep learning model comprises a first module to process the customer demographic data having a first model to process independent features; and a second module to process each unit of sequential interaction in customer 360-degree view, the second module having a second model to process structured data, a third model to process text data and create text embeddings, and a fourth model to process integrated history.
[014] In a further embodiment of the invention, each of the first and second module includes at least one attention layer to implement an action in an artificial neural network for each customer for predicting the customer dissatisfaction threshold.
[015] In a further embodiment of the invention, a method for analyzing and predicting customer dissatisfaction threshold is provided. The method comprises the steps of: obtaining a plurality of customer history data; storing the customer history data in a first data storage; aggregating the plurality of customer history data by an aggregator server; processing the plurality of customer history data through a management server having a processor to generate a processed data by implementing a deep learning model; and predicting and presenting customer dissatisfaction threshold based on the processed customer history data by an inference server.
[016] In a further embodiment of the invention, storing the customer history data comprises the step of storing the customer history data in the first data storage in a classification including (i) a static data i.e. demographics of the customer, (ii) sequential structured data like transactions or usage history and (iii) sequential unstructured data like customer complaints or feedbacks provided by the customer.
[017] In a further embodiment of the invention, the step of aggregating the customer history data is performed by the aggregator server to create a 360-degree view of the customer and process on a management server.
[018] In a further embodiment of the invention, the processing of the customer history data comprises the steps of: processing the customer demographic data having a first model to process independent features in a first module of the deep learning model; processing each unit of sequential interaction in customer 360-degree view in a second module of the deep learning model; the second module having a second model to process structured data; processing text data and create text embeddings in a third model; and processing integrated history in a fourth model.
[019] In a further embodiment of the invention, the step of processing the customer history data includes the step of implementing an action of the attention layer to first and second module in an artificial neural network for each customer to predict the customer dissatisfaction threshold.

BRIEF DESCRIPTION OF THE DRAWINGS
[020] Reference will be made to embodiments of the invention, examples of which may be illustrated in accompanying figures. These figures are intended to be illustrative, but not limiting. Although the invention is generally described in context of these embodiments, it should be understood that it is not intended to limit the scope of the invention to these particular embodiments.
Figure 1 illustrates a system for analyzing and predicting the customer dissatisfaction threshold in accordance with an embodiment of the invention.
Figure 2 illustrates an architecture of a deep learning model for predicting the customer dissatisfaction in accordance with an embodiment of the invention.
Figure 3 illustrates the method steps for analyzing and predicting the customer dissatisfaction threshold in accordance with an embodiment of the invention.
DETAILED DESCRIPTION OF THE INVENTION
[021] The present invention provides a system for analyzing and predicting customer dissatisfaction is provided whereas the system is capable of analyzing and predicting customer dissatisfaction threshold in an accurate manner and thus helps in improving retention rate of customers in a business.
[022] Referring to Figure 1, a system (100) for analyzing and predicting customer dissatisfaction is illustrated. The system (100) has a first data storage (150) for storing a customer history data. The first data storage (150) includes a static data storage (104) for storing the demographic details of the customers; a sequential structured data storage (106) for storing the customer interaction history; and an unstructured data storage (108) for storing the data in the form of the customer experiences including complaints or feedbacks provided by the customer. The system is having an aggregator server (110) connected with the first data storage (150) and configured to aggregate the customer history data.
[023] The system includes a management server (112) connected with the aggregator server. The management server (112) has a processor (not shown) configured to implement a deep learning model for processing the customer history data and generating a processed data. The system further includes a second data storage (114) for storing the processed customer history data.
[024] The system has an inference server (116) connected with the management server (112). The inference server is configured to predict the customer dissatisfaction threshold and key reasons impacting the experience for each customer based on the processed customer history data. The inference server (116) is connected with the second data storage (114) of the processed customer history data. The inference server (116) comprises a prediction module (118). Based on the processed customer history data, the prediction module (118) is configured to identify and visualize specific issues with the customer attributable to leave a particular business.
[025] The prediction module (118) utilizes the generated processed data to provide visualization to the system user by summarizing major issues impacting customers, a list of customers with high probability of attrition and proposed retention strategy specific to a customer or a group of customers, etc.
[026] The system includes a customer alive status storage (102) for storing a data related to the customer alive status for the customers of the specific business. The system also has a user device (120) to display the customer dissatisfaction threshold and specific issues with the customer attributable to leave a particular business.
[027] In an embodiment of the invention, the user device may be but not limited to, a computer which displays the processed data to predict customer dissatisfaction level. Further, the user device can also present or visualize a summary incorporating major issues impacting customers, a list of customers with high probability of attrition and proposed retention strategy, etc.
[028] In an embodiment of the invention, the processing of the plurality of customer history data and generating the processed data includes at least a memory configured to store an executable program, and a processor. The processor is configured to implement the deep learning model when the executable program stored in the memory is executed.
[029] Referring to Figure 2, an architecture of a deep learning model is illustrated. The deep learning model (200) includes a first module (210) to process the customer demographic data having a first model (212) to process independent features. The deep learning model has a second module (220) to process each unit of sequential interaction in customer 360-degree view. The second module (220) is having a second model (222) to process structured data, a third model (224) to process text data and create text embeddings, and a fourth model (226) to process integrated history.
[030] In an embodiment of the present invention, each of the first and second module of the deep learning model includes at least one attention layer (228) to implement an action in an artificial neural network for each customer for predicting the customer dissatisfaction threshold.
[031] Referring to Figure 3, the steps for performing a method for analyzing and predicting customer dissatisfaction threshold are illustrated. The method includes the step of obtaining a plurality of customer history data. The method includes the step of storing the customer history data in a first data storage (150). Thereafter, a step of aggregating the plurality of customer history data is performed by an aggregator server (110). The method includes a step of processing the plurality of customer history data through a management server (112) having a processor to generate a processed data by implementing a deep learning model (200); and predicting and presenting customer dissatisfaction threshold based on the processed customer history data by an inference server (116).
[032] In a further embodiment of the invention, the method step of storing the customer history data includes the step of storing the customer history data in the first data storage (150) in a classification including (i) a static data i.e. demographics of the customer, (ii) sequential structured data like transactions or usage history and (iii) sequential unstructured data like customer complaints or feedbacks provided by the customer. The method further includes a step of aggregating the customer history data by the aggregator server (110) to create a 360-degree view of the customer and process on a management server (112).
[033] In further embodiment of the invention, the steps of processing of the customer history data includes the step of: processing the customer demographic data having a first model (212) to process independent features in a first module (210) of the deep learning model (200); processing each unit of sequential interaction in customer 360-degree view in a second module (220 of the deep learning model (200); the second module (220) having a second model (222) to process structured data; processing text data and create text embeddings in a third model (224); and processing integrated history in a fourth model (226). The processing of the customer history data further includes the step of implementing an action of the attention layer (228) to first (210) and second module (220) in an artificial neural network for each customer to predict the customer dissatisfaction threshold.
[034] In an embodiment of the invention, the fourth model of the deep learning model can be a sequence to sequence deep neural network model like LSTM or a pure attention-based model consisting of positional embeddings to exploit sequences.
[035] In another embodiment of the invention, the processing of the plurality of customer history data and generating the processed data is performed by a processor implemented with a deep learning model. The deep learning model is a type of artificial neural network which is a learning algorithm inspired by the structure and functional aspects of biological neural networks. Computations are structured in terms of an interconnected group of artificial neurons, processing information using a connectionist approach to computation.
[036] In another embodiment of the invention, the artificial neural network utilizes plurality of attention layers to model customer interaction history with the business. The system utilizes the plurality of customer history data to provide a representation of the customer’s preferences. The system then utilizes this representation and learns to focus on specific instances in the interaction history based on its learnings from the behavior of customers with similar experiences and feedbacks. This information is then used to predict the propensity of the customer to complain/churn. Whereas the attention given to each experience is used to determine which of the issues are impacting customers most to make a predicted decision to continue with the business or to leave a particular business.
[037] In another embodiment of the invention, the artificial neural network is trained based on the plurality of customer history data which is substantially treated as a training data whereas the training data is processed via plurality of attention layers and prediction of customer dissatisfaction threshold is made using a set of business rules based on specific heuristics or the domain knowledge.
[038] In a further embodiment of the invention, customer history data can be obtained from different databases or servers or on cloud databases. Once obtained, the customer history data is integrated to create a 360-degree view of customer and then be processed on a computer server.
[039] In a further embodiment of the invention, the processing of the customer data through deep learning model includes applying predefined rule based or heuristics-based data cleaning steps, statistics-based data pre-processing steps and one or more deep neural networks. The deep neural networks consist of at least one attention layer to dynamically select and focus on certain input features, one temporal layer to process sequential interaction data like transactions done or feedback provided or usage history.
[040] The temporal layer can consist of a sequence to sequence model like RNN or sequence to sequence model with memory like LSTM or a pure attention based model like transformer with positional encodings to exploit sequential information. The processor also includes the prediction module to post process results to identify key customer issues includes a means to analyze the deep learning layers to identify input features which were given most attention for the model prediction. It also includes means to use heuristics or business domain knowledge to prioritize the identified issues based on business impact or fixability.
[041] In another embodiment of the invention, the deep learning model includes at least the step of training the deep learning model based on a plurality of weightage of the training data.
[042] In a further embodiment of the invention, several stages of data transformations are applied by the aggregator server for predicting the customer dissatisfaction. A plurality of customer history data is obtained in form of raw data from internal business specific sources and external sources. A raw data related to onboarding application, demographics of the customer, Application usage by the customer, transaction history, purchase history, services used or rendered by the customer, offers claimed by the customer, service requests raised by the customer, escalations moved by the customer, feedback shared by the customer, ratings provided by the customer and images shared by the customer etc. are obtained from the business internally. External customer history data in raw form is obtained from external sources which includes data in form of demographics, social network used by the customer and interest preferences of the customer. In transformation stage of the customer history data, the collected raw data is curated or sorted in classifications of static data, customer journey data, unstructured data and customer success metric. The static data includes the data related to demographics of the customer and the customer preferences. The customer journey data incudes data related to engagement history, transaction history, usage history and support history. The unstructured data includes feedback(s) shared by the customer, images shared by the customer, complaints of the customer and reviews provided by the customer. The customer success metric includes Net Promotor Score (NPS) score, service ratings and relationship live status of the customer. The sorted data is further processed and classified in form of static data, customer journey data, unstructured data, and customer success metric data. The static processed data includes data related to customer categories and customer priority. The processed customer journey data includes the data related to data which is sequentially stitched and data related to customer journey. The processed unstructured data incudes the data in form of cleaned processed text and cleaned processed images. The processed customer success metric data includes data related to integrated customer success score. The processed data is then aggregated and classified in form of integrated customer 360-degree journey and customer success labels. The aggregated customer history data is processed by using Data Science Models i.e. the deep learning model to provide an integrated customer 360-degree journey with customer dissatisfaction threshold, the attrition probability of a business and the key or major issues impacting the experience of the customer.
[043] In a further embodiment of the present invention, a plurality of customer history data ingested to the aggregator server is obtained in various modes i.e. onboarding applications, application usage, transaction history, engagement history, contact center agent and ratings shared through Net Promoter Score. The customer history data collected from onboarding application and the data received through external data enrichment sources is stored in demographic and static customer data storage. The customer data received in form of Application usage, transaction history and engagement history is classified into structured data and image/ text data. The structured data is stored in a sequential structured data storage. The data in image or text form is stored in an unstructured data storage. The customer history data obtained from contact center agent through contact centre portal is stored in a contact centre data storage which is stored in the structured data storage. The customer history data as obtained from contact center agent also includes call recordings which are converted into a text format using speech to text engine to create call transcripts and textual call summary which are stored in unstructured data storage. The customer history data obtained from contact center agent based on service case rating is stored to a customer alive status storage and the unstructured data storage. The system is configured to enroute the customer history data in form of call transcriptions to customer alive status storage based on the service case ratings. The data obtained in form of ratings received by sources such as Net Promotor Score (NPS) is stored in the customer alive status storage. The data stored in structured data storage is also utilized by system to estimate the customer alive status and consequentially this data is stored in the customer alive status storage.

ADVANTAGES OF THE INVENTION
[044] The system of present invention is capable of predicting the customer dissatisfaction threshold more accurately. The present invention can identify and highlighting the specific issues with the customer attributable to leave a particular business. The system has capability to identify the specific customer history data and customer interactions while considering a plurality of customer history data to predict customer dissatisfaction for each customer.
[045] The system of the present invention also handles temporal data better than existing systems and is also capable to retrieve older information like app experience from over the past years. Consequently, the system can predict the customer dissatisfaction much more accurately in comparison to the existing systems and methods. The present system improves the customer retention by providing information of a specific customer history data which was considered important in analyzing and predicting customer dissatisfaction. With this predicted information of a customer dissatisfaction, the businesses can focus on improving their customer experience with a better understanding of their overall journey.
[046] The foregoing description of the invention has been set merely to illustrate the invention and is not intended to be limiting. Since the modifications of the disclosed embodiments incorporating the spirit and substance of the invention may occur to the person skilled in the art, the invention should be construed to include everything within the scope of the disclosure. ,CLAIMS:WE CLAIM:
1. A system for analyzing and predicting customer dissatisfaction, the system comprising:
a first data storage (150) for storing a customer history data;
an aggregator server (110) connected with the first data storage (150) and configured to aggregate the customer history data;
a management server (112) connected with the aggregator server, the management server having a processor configured to implement a deep learning model (200) for processing the customer history data and generating a processed data; and
an inference server (116) connected with the management server (112) and configured to predict the customer dissatisfaction threshold for each customer based on the processed customer history data.

2. The system as claimed in claim 1, wherein the first data storage (150) comprises:
a static data storage (104) for storing the demographic details of the customers;
a sequential structured data storage (106) for storing the customer interaction history; and
an unstructured data storage (108) for storing the data in the form of the customer experiences including complaints or feedbacks provided by the customer.

3. The system as claimed in claim 1, wherein the system includes a second data storage (114) for storing the processed customer history data.

4. The system as claimed in claim 1, wherein the inference server (116) is connected with the second data storage (114) of the processed customer history data.

5. The system as claimed in claim 1, wherein the inference server (116) comprises a prediction module (118), the prediction module configured to identify and visualize specific issues with the customer attributable to leave a particular business.

6. The system as claimed in claim 5, comprises a user device (120) to display the customer dissatisfaction threshold and specific issues with the customer attributable to leave a particular business.

7. The system as claimed in claim 2, wherein the deep learning model (200) comprises:
a first module (210) to process the customer demographic data having a first model (212) to process independent features; and
a second module (220) to process each unit of sequential interaction in customer 360-degree view, the second module (220) having a second model (222) to process structured data, a third model (224) to process text data and create text embeddings, and a fourth model (226) to process integrated history.

8. The system as claimed in claim 7, wherein each of the first and second module comprise at least one attention layer (228) to implement an action in an artificial neural network for each customer for predicting the customer dissatisfaction threshold.

9. A method for analyzing and predicting customer dissatisfaction threshold, the method comprises the steps of:
obtaining a plurality of customer history data;
storing the customer history data in a first data storage (150);
aggregating the plurality of customer history data by an aggregator server (110);
processing the plurality of customer history data through a management server (112) having a processor to generate a processed data by implementing a deep learning model (200); and
predicting and presenting customer dissatisfaction threshold based on the processed customer history data by an inference server (116).

10. The method as claimed in claim 9, wherein storing the customer history data comprises the step of:
storing the customer history data in the first data storage (150) in a classification including (i) a static data i.e. demographics of the customer, (ii) sequential structured data like transactions or usage history and (iii) sequential unstructured data like customer complaints or feedbacks provided by the customer.

11. The method as claimed in claim 9, wherein the method comprises a step of aggregating the customer history data by the aggregator server (110) to create a 360-degree view of the customer and process on a management server (112).

12. The method as claimed in claim 10, wherein the processing of the customer history data comprises the steps of:
processing the customer demographic data having a first model (212) to process independent features in a first module (210) of the deep learning model (200);
processing each unit of sequential interaction in customer 360-degree view in a second module (220 of the deep learning model (200); the second module (220) having a second model (222) to process structured data;
processing text data and create text embeddings in a third model (224); and
processing integrated history in a fourth model (226).

13. The method as claimed in claim in claim 12, wherein the processing the customer history data includes the step of implementing an action of the attention layer (228) to first (210) and second module (220) in an artificial neural network for each customer to predict the customer dissatisfaction threshold.
Dated this 30th Day of January 2021
ADITYA BIRLA MANAGEMENT
CORPORATION PRIVATE LIMITED
By their Agent & Attorney

(Adheesh Nargolkar)
of Khaitan & Co
Reg No IN-PA-1086