Businesses oriented towards creating a top-of-the-line customer experience need a competent and comprehensive 24/7 customer support. In the past, offering a reliable phone number with an informed customer service team was enough to satisfy customers with questions and problems. But because of shifting expectations, companies are starting to introduce a chat option as a way of instant contact with their clients. The chat bots are artificial intelligence (AI) enabled programmes (literally meaning ‘robots which can interact with humans through chats’) that simulate interactive human conversations by using key pre-calculated user phrases and auditory or text-based signals. They allow more freedom for customers to start a conversation whenever and wherever they want to, and for as long as they need to.
Noida Power, a distribution company of RP-Sanjiv Goenka Group, with more than 93,000 live connections for supplying power to Greater Noida region, has created a smart, instant, and interactive chat bot that lets customers text questions like: “How can I apply for a new connection?” or "When is my next bill due?" and so on. The chat bot called Saathi is live 24/7, and has been conceptualised, developed, and implemented by NPCL’s in-house IT team. It took approximately two months, after the finalisation of the business requirements, for the team to create the prototype, test, and make the final product go-live on NPCL's website and Android version of the app. The working model responds to real questions asked by customers, and its AI component also allows it to learn as it goes along. NPCL happens to be the among the first distribution companies in India to have implemented a live and fully-functional chat bot interfacing with its consumers.
Meeting Customer Expectations
Customers are looking for newer ways to interact with companies, thanks to several converging technology trends. Foremost, customers want to engage in a way that is most convenient for them, and many now prefer texting instead of using the phone or visiting the brick-and-mortar office. The utility companies are therefore aiming at smart and effective process management.
No standing in queues, or waiting for the call to be answered by the customer care executive, or rummaging through the FAQ segment of the website to get a solution.
The AI chat bot, Saathi, works as a smart-conversation assistant and helps NPCL to deliver convenience and personalisation according to the customer’s preference. Natural language processing, artificial intelligence, and machine learning capabilities enable the chat bot to be relevant and practical for the target audience – the existing and the potential customers.
The Technology Underneath
Built on RASA open source framework, the chat bot’s machine learning capability continually monitors and adapts to what people expect and how they ask questions. With little data to train the models in the beginning, RASA framework was the best option to start the development with. RASA, being open source, offers scope for a high degree of customisation as compared to other chatbot products offered by major players. It can be deployed on-premise and hence allows one to avoid data security-related issues, inherent to solutions hosted on the cloud. It also comes with good support and has been adopted by a wide customer base. Among other conversational AI platforms, ChatterBot and Haptik were also evaluated. Apart from that, a solution from an external vendor, called Yellow Messenger, was considered before opting to go for in-house development on the RASA framework. Even though RASA has minor setbacks like support for analytics is not available with the free version, or it does not offer a dashboard to review the performance, the overall advantages it offered by far outweighed the limitations.
In NPCL website or app, once the user clicks on the Saathi chat window, there is a pre-designed menu that prompts the user to click on segments like: New Connection, Bill & Payment, Complaints, General Query, Download Forms, and Contact Us. The navigation is ‘stateless’, giving the users the freedom to navigate from an iteration to another smoothly. There is also a window, where the customer can ask unscripted questions that might be included or not included in the menu. The response is instant in both cases. The overall accuracy, including scripted and unscripted queries is more than 91% and the accuracy for unscripted queries is around 64%. While the accuracy level of the scripted menu is higher, the unscripted section enables the breadth of queries to be much wider. The development team of NPCL is continuously working to train the chat bot to increase the accuracy levels further.
Saathi also has a voice-assisted chat function. The words spoken in English after activating the voice option in Saathi get automatically converted to English text and thereafter the chatbot responds similarly like other text inputs.
If the system does not respond exactly the way it was expected to, those errors are recorded. Data is analysed through the backend data analytics system, and then the chat-bot is made to learn and adapt to the queries. So that the next time somebody asks the same question, NPCL Saathi is able to throw up accurate results.
The Technology Underneath
Built on RASA open source framework, the chat bot’s machine learning capability continually monitors and adapts to what people expect and how they ask questions. With little data to train the models in the beginning, RASA framework was the best option to start the development with. RASA, being open source, offers scope for a high degree of customisation as compared to other chatbot products offered by major players. It can be deployed on-premise and hence allows one to avoid data security-related issues, inherent to solutions hosted on the cloud. It also comes with good support and has been adopted by a wide customer base. Among other conversational AI platforms, ChatterBot and Haptik were also evaluated. Apart from that, a solution from an external vendor, called Yellow Messenger, was considered before opting to go for in-house development on the RASA framework. Even though RASA has minor setbacks like support for analytics is not available with the free version, or it does not offer a dashboard to review the performance, the overall advantages it offered by far outweighed the limitations.
In NPCL website or app, once the user clicks on the Saathi chat window, there is a pre-designed menu that prompts the user to click on segments like: New Connection, Bill & Payment, Complaints, General Query, Download Forms, and Contact Us. The navigation is ‘stateless’, giving the users the freedom to navigate from an iteration to another smoothly. There is also a window, where the customer can ask unscripted questions that might be included or not included in the menu. The response is instant in both cases. The overall accuracy, including scripted and unscripted queries is more than 91% and the accuracy for unscripted queries is around 64%. While the accuracy level of the scripted menu is higher, the unscripted section enables the breadth of queries to be much wider. The development team of NPCL is continuously working to train the chat bot to increase the accuracy levels further.
Saathi also has a voice-assisted chat function. The words spoken in English after activating the voice option in Saathi get automatically converted to English text and thereafter the chatbot responds similarly like other text inputs.
If the system does not respond exactly the way it was expected to, those errors are recorded. Data is analysed through the backend data analytics system, and then the chat-bot is made to learn and adapt to the queries. So that the next time somebody asks the same question, NPCL Saathi is able to throw up accurate results.
The Benefits Accrued
1. Scalable - NPCL Saathi chat bot currently handles 100 queries per day. Even with the current capabilities, the chat bot can handle up to 3600 queries per second.
2. Available 24/7 - The chat bot is like a virtual assistant that is available at all times to provide support to the customers. No waiting for the customer service agent to finish with another call, or listening to repetitive call-waiting music or messages, or navigating through a sea of interactive voice responses (IVR), to get to the required solution.
3. Reduced Manual Effort - A key element of Saathi is the reduction of the manual effort required for human interaction. One way this has been achieved is by making sure that most of the possible options are provided by the chat-bot itself through a pre-programmed menu, with the user just needing to select the right option. This greatly reduces the time and effort required to interact with the chat bot. It also helps the chat-bot to respond accurately to the users by eliminating the grammatical and spelling errors in the inputs, or provide service to the users who are not comfortable with typing in English.
4. Speed - Saathi works with immense computing power and speed. It returns responses within milliseconds, whereas an average call-centre call lasts 5 minutes. Saathi can handle 3600 queries/second. Saathi perfectly complements NPCL’s call centre and customer care office, to reduce the manpower bottleneck, especially during peak times.
5. Paperless - Saathi has helped in saving more than 200 paper tokens in the Customer Care office, and also led to the daily saving of approximately 120 printed forms related to bill or new connection. This has led to saving of more than 300 pages per day.
6. Reliable and Correct - While the overall accuracy is currently at 91%, for the unscripted text inputs the reliability is about 64%. The percentage of accuracy is constantly increasing by constantly training the chat bot by mapping the correct responses.
7. Cost Saving - Human staff takes time and money to hire and train. On an average, a company spends in excess of INR 30,000 to hire a customer service representative. That cost keeps piling up through the course of their career. A programmed AI chatbot, like Saathi, saves a lot of cost as it is a one-time expense. No additional training or development is required. Saathi works with little reconfiguration and regular updates to incorporate new learning. There is practically zero downtime.
8. Sentence Recommendation's- If users are unsure of what they are asking for or if the conversation goes into a loop that makes exiting difficult, sentence recommendations help users to understand what the chat bot can do. It also helps to ease the conversation. Sentence recommendation function has increased the level of accuracy significantly.
NPCL Saathi is capable of handling 3600 queries per second. It has reduced 11% footfall in the customer care offices, amounting to reduction of usage of resources.
Future Scope
1. Scalable: By simply adding more servers, memory, and computing power, the chat bot can be scalable to handle much more volume, to ease out the congestions in call centres or customer care offices. At present, the focus is on expanding the range of content served through the Chatbot as well as on improving the accuracy level of response. Subsequently, it can be extended to providing responses in local language as well as to voice bot.
2. Replicable: The chat bot application can be replicated for other functions which require a fair amount human intervention yet are standard queries in nature. For instance, IT support, HR support etc. The chat bot can also be implemented in other companies of the Group, which require interaction with the customers.
3. Personalisation: The chat bot application can be replicated for other functions which require a fair amount human intervention yet are standard queries in nature. For instance, IT support, HR support etc. The chat bot can also be implemented in other companies of the Group, which require interaction with the customers.
4. Regional Languages: Adding regional languages to the repertoire of the chat bot can make it user-friendly and accessible to users comfortable in regional languages. This is expected to increase the number of users manifold.
There is a lot of potential for AI chat bots. Companies like NPCL are relying on this technology to know and provide service to their customers better. This is true for the service as well as support functions. With voluminous data to process, this type of technology will always be in favour. There are different notions of developing algorithms to train bots faster and making them more human-like. With so much on the anvil, it is safe to say the AI chat bot technology is here to thrive and grow.
Customer Feedback
The information about Saathi was circulated through press releases in local newspapers and messages were sent out to consumers to encourage them to use Saathi for customer service. The response has been very encouraging. A user-feedback survey was also done.
Saathi’s responses have been very fast, I was able to find answers to most of my queries accurately
I am very happy with most of the features provided in Saathi
For more information on NPCL Saathi, or for any assistance in AI chat bot implementation, please get in touch with:
Arijit Mitra,
Vice President – IT
Noida Power
E: arijit.mitra@rp-sg.in
M: +91 99119 98204
