Technology is undoubtedly ensuring business continuity across the banking sector especially after the pandemic outbreak. Banks have been employing Artificial Intelligence that has enabled them to boost productivity, save costs while at the same time enhance the overall customer experience. The emergence of voicebots or voice assistants in the BFSI sector has become a game-changer. With the use of speech recognition and Natural Language Processing (NLP) technology boosted by AI-powered business algorithms, voicebots help banks accelerate the customer engagement strategy. In an exclusive conversation with Harshal Desai of Elets News Network (ENN), Ratan Kesh, EVP & Head of Retail Operations and Service at Axis Bank shared his views on AXAA – an AI-powered conversational Voicebot.
To what extent is Eva, Tia, Alexa, Siri (digital voice applications) going to have an impact on the financial institutions?
Starting in the 2000s, voice response became more common and cheaper to deploy. Banks have launched voice-activated consumer banking sometime in early 2014, which has slowly progressed around the world and it has been providing convenience to the customer. Today most of the simple queries can be resolved through the IVR (Interactive Voice Response) without reaching out to the agent directly.
However, 2018 has been a hallmark year in “Voice – Activated Banking” with more consumers using Alexa and other voice services for daily banking. Powered by Artificial intelligence (AI) and Machine Learning (ML), these solutions enabled more sophisticated voice-led banking services. Consumers started using Voice-based searches even in Google using Alexa and Siri and a host of banks started pilot or full-scale launches for simple query resolution using voice-activated banking solutions. Since then, the technology has evolved, the range of services has grown rapidly and many more companies in the BFSI space have integrated Voice-BOTinto traditional IVR to deliver a seamless delivery seamless experience to the customers.
In a traditional IVR, if you want to reach out for any specific service you will have to go through the IVR tree. For example, to change the pin of a debit card you will have to traverse through multiple layers in the IVR tree starting with choosing the language (English or Hindi), account type (Saving or Current), Card type (Debit or Credit), Issue type (PIN Related) and then choose the specific requirement e.g Generate PIN. Therefore, by the time you generate a pin for your debit card, you would have already gone through a large part of the IVR tree with multiple clicks. However, with voice-activated IVR or VoiceBOT, you can simply ask, “I want to generate pin for my debit card” and the BOT would directly take you to the right node and help in generating the PIN. This means, the IVR traversal times come down, BOT understands the intent and thereby deliver superior experience to the customer. My belief is that most financial institutions and other service industries wanting to provide convenience to customers would find such solutions quite useful and therefore, adoption of such technologies will grow rapidly.
Will regional dialects, different languages, restrict the growth of voice assistants?
The answer is yes and no. India is a land of diversity with 22 languages and 160 plus dialects. It is quite complex to build BOT technology which can seamlessly understand all these languages and dialects. However, the majority of Axis Bank’s customers use Hindi and English and to a large extent Hinglish (a combination of Hindi and English) for day-to-day banking requirements. Our voice assistant has the capability to understand and communicate in Hindi, English and Hinglish. Weare developing solutions in few other prominent Indian languages.
The voice interfaces use Natural Language Processing (or NLP) to decipher words which are also termed as Speech-to-Text, NLU (Natural Language Understanding). Like in Siri or Alexa, this transcription system is used to teach the Voice Bot to understand the “Intent” of the customer e.g what exactly is the customer’s query or request. Similar technology is also used in automated subtitling, smart homes, and in-car systems. Having understood the “Intent”, Voice Bot refers to the solution library to provide replies using Text-to-Speech (TTS) technology. TTS is an AI-driven simulation of human speech using a deep learning technique. The advantage of TTS or speech synthesis is that it spontaneously generates a human voice from raw text instead of playing pre-recorded files.
The good news about Voice Bot is that it has the ability to learn from both success and failure and has the power of acceptance i.e., the ability to understand that an error has been made and next time will not repeat the error.
Clearly, we are seeing customers feeling happy about interacting with the voicebot because it helps them converse and get response to their queries in a language that they are comfortable in. They do not have to ask the Agent to talk in their preferred language or get the call diverted to another agent who can understand the language.
What are the major areas where voice assistants are giving 100% results?
We launched the VOICEBOT in two phases with underlying NLU capabilities (Natural Language Understanding). The first phase was to understand the intent and route the call to right agent, the right node for resolution, therefore. In that journey, we got 90% plus level of accuracy wherein BOT is able to correctly understand the customer’s requirements. Subsequently, we launched query resolution across 17 basic services e.g account balance, changing of address, nearest ATM, applying for a chequebook, etc. In these 17 basic services too we got high degree of accuracy (above 90%) and thus customers are finding it more convenient to use VOICE BOT. However, I must mention that we are far away from declaring victory. Voice Bot cannot handle objections or complex queries raised by a customer. These complex queries are handed over to an agent. In addition, in the current form, the Voice Bot does not understand and decipher sarcasm or sentiment and tonality of customers’ voice. However, we are on the path of improving technological capabilities.
What would happen to human agents then? Do we still, require them?
Voice Bot has the ability to handle basic queries and deliver a superior experience to customers. However, as soon as it comes to the state where a customer requires advice on complex banking matters or would like to raise a dispute, human agent or expert bankers have to come into the picture.
Our belief is that the VoiceBot will not replace humans instead of humans and Humanoid (BOT) will work together side by side to deliver the best experience to the customer. BOT will handle most of the simpler, routine query and requests; however, in areas where BOT is unable to handle a customer’s requirement, the same would be seamlessly handed over to an Agent who will ensure subsequent request fulfilment. This is where it is very important for companies like Axis Bank to integrate relevant technologies. For example, when a call gets transferred from BOT to a human agent, the context and background of the previous conversation should be transferred so that the customer does not need to repeat the entire concern. Technology will play a big role in ensuring recording of calls handled by BOT, transfer the transcript along with relevant knowledge portal using predictable AI so that human agent is empowered to handle the call in a more effective manner.
On top of the voicebot, we are building real-time speech analytics that has the ability to convert speech to text and understand the context of the call. If the conversation is not going smoothly between the agent and the customer, then it has the ability to give certain instant triggers and guidance to the agents so that to do course correction. Alternatively, it also has the ability to step in between the conversation and transfer the customer to a subject-matter expert (SME) who can handle the conversation better. Hence, a technology stack and integration between VoiceBot, speech analytics, Knowledge Portal and Agent’s Unified-Desk will deliver a superior customer experience.
How can the overall customer experience be amplified to deliver an Omnichannel experience and what role would voicebot play in that?
I have already spoken about the three capabilities namely the VoiceBot, Knowledge portal, and Speech analytics. The underlying framework is all about having a strong knowledge management portal so that if a customer is asking any type of question, we have the answer in the knowledge portal or knowledge bank. For a simple question, we can use an API to take the response from the knowledge management portal and relay the response directly to the customer or via an agent. If the customer comes to the website for certain queries, it should be available at the click of a button. Assuming it was not readily available and the customer clicks the Chat option, the ChatBot or Chat Agent should be able to resolve the query using the Knowledge Bank; Customers, however, can have the option to Click to connect with the Call Center for more information. At the call centre, the voicebot will resolve the query. If the voicebot is not able to resolve the query then the human agent will take over. What we are essentially doing is providing the customer to reach out to the bank in any mode of preference. We are also working towards creating a seamless integration of Email Bot, WhatsApp Bot, Voice Bot, Chat Bot using the power of underlying NLP capabilities, AI, ML and Knowledge portal. On top of all this, we are building what we call a service data layer, which brings together all the conversations across various channels. All this data will be organised for each customers (customer 360 degree) and made available to front-facing agents to enable then handle the customer queries proactively including that of matching conversations with profile and persona of customers.
In other words, voicebot alone will not be able to deliver omnichannel experience, unless email bot, chatbot, speech analytics, knowledge portal, and service data layer connect together to deliver a complete ecosystem on which voicebot will flourish.
Axis Bank’s philosophy of customer engagement is Dil Se Open. How does voicebot fulfill that? Isn’t there a dichotomy?
Dil Se Open means that we at Axis Bank will demonstrate human empathy and openness in every interaction with our customers.
Let me give you an example from the peak of Covid induced lockdown. At the beginning of lockdown in March-April 2020, we were forced to operate our branches with 40% capacity to maintain social distancing norms while maintaining all the hygiene standards.
During this time, most of our employees were working from home and they reached out to the customers over call just to enquire well-being of customers and if any support is needed. Of course, we ensured that our digital channels are working fine. We created a large number of work from home call centre customer service capabilities to help customers in times of need. We created Covid Helpline for employees and extended employees and their family members.
All these factors, which were largely driven around being sensitive to the customers’ needs, had a significant positive impact on customer loyalty. We found that our customer experience score (rating in Top 2 boxes) improved from 67% to 75% in just about six weeks which is a phenomenal improvement. This is one of the reflections of Dil Se Open where we could show empathy towards customer requirements. On the other hand, we were perhaps the first bank to create 500/ 1000 and 1500 Work From Home (WFH) Call Centre Capability with suitable endpoint security and voice recording capability. In the middle of the COVID-19, we launched Voice BOT to support customers. All these initiatives are driven out of the philosophy of keeping customer as the centre of attention – that is our approach to demonstrating “Dil Se Open” to our customers.
Specific to Voice Bot interactions – it has two dimensions; firstly, time is precious and most customer is in a hurry and merely looking for a quick response to their queries. Why would you want a customer to wait in the call center queue to get an answer to a simple query, which can be answered by a Voice Bot instantly and accurately? That to me is Dil Se Open and the voicebot has the capability to resolve the queries 24×7. Secondly, wherever the Voice Bot is unable to answer a customer query, the customer is being transferred to a human agent seamlessly. The agents are trained to handle complex queries, show empathy and fulfill requirements. So, human and humanoid are working together to deliver the essence of “Axis Bank – Dil Se Open” philosophy.
The COVID-19 outbreak inflated the digital inclination of customers across the banking sector. How is Axis Bank utilizing this opportunity by leveraging technology?
Before COVID-19, the ratio of digital to physical channels was around 40-60 where 40 per cent was digital and 60 per cent was still relying on physical applications. During COVID-19, the ratio almost became 90-10, where 90 per cent was digital adoption. Currently, as we are moving out of the COVID-19 situation and lockdown is withdrawn, we are seeing a 50-50 ratio and fast-moving towards 60-40. Some of the customers who have adopted digital are staying digital because of the convenience it offers.
What is making this happen is the acceleration of STP (Straight-through processing) journeys that we have enabled in insta services like address change, nomination, updating KYC. These are using AI-powered validations, real-time API with various Govt Databases as well internal validations with internal core systems to provide instant fulfillment. Furthermore, the customer journeys are created once and are being rendered seamlessly across all touchpoints – in self-service or assisted mode.
Recently, we have also introduced Video KYC, instant loan disbursal, digital services for SMEs which are helping our customers to avail the products while helping the bank to reduce the cost of acquisition while delivering superior Cx.
Axis Bank has partnered with Vernacular.ai to deploy its multi-lingual voicebot AXAA on its phone banking IVR. How has been the experience so far on this?
One of the reasons why we wanted to partner with Vernacular.ai is because they are more agile. Voicebot capability is easier said than done. More importantly, the ability to connect with Axis Bank’s core technology is complex and therefore we wanted a partner who is willing to be agile, flexible to work with us and develop this product to suit our environment.
I must say that Vernacular.ai has so far shown agility, flexibility to learn and willingness to collaborate for end objective.
I also would like to say that my team at Axis Bank has really worked hard to learn and design the solution with Vernacular.ai, which is benefitting companies like Vernacular.ai to progress in their own journey and acquire more customers. We really worked hard to make sure that the solution is effective and accurate. In essence, what matters is a good partnership between the technology provider as well as Bank’s own business and technology team.
I have already mentioned the value delivered by reducing customer wait time across the 17 specific services. I think our real success will be better understood when we want to apply the voicebot for handling disputes or add more languages for resolving more complex queries. We are hopeful that Vernacular.ai will show similar hunger, agility and keenness to partner with us as we progress towards solving complex problems. The technology stack deployed by them has so far worked well but it has to undergo significant advancements and convergence as we progress further from hereon.
Is this deployment helping you increase the efficiency of your contact center operations?
We receive close to hundred thousand customer calls at our call centres on a daily basis. VoiceBot gives us the capability and flexibility to handle customer calls coming on 24×7 basis. Furthermore, with 17 services being handled using BOT, if I am able to get 12-15% calls being handled by BOT seamlessly, it is a great value for us from an Operational efficiency perspective.
Secondly, voicebot is also helping us integrate with some of the other bots and using the knowledge portal to answer queries correctly. I am able to enrich my knowledge portal, which helps me apply the same to the other social media platforms like WhatsApp Bot, ChatBot, Email Bot. Thirdly, as I said, we will declare a victory when the voicebot is able to handle a dispute or a complex query of the customer. For now, I will say yes, it is has been quite promising so far.
What kind of challenges have you faced, and how has Axis bank and Vernacular.ai come together to address those challenges?
Data is the key. We had to take a huge amount of data for training purposes and that is where, I would say, partnership and resilience come into the picture. Initial thought from some of the team members was to go live and test with 2-3 services. However, our considered view, after a thorough deliberation was that unless we are able to test it for a sizeable volume of call, it would not make a meaningful impact. Thus we decided to go ahead with 15-17 services but that demanded us to work with huge amount of data for a significantly longer period of time. Extracting “Intents”, “Entities” from such large data-set was a task on its own. BOT had to be trained in large number of scenarios. Furthermore, it needed extracting the key features from training data and ultimately create sample conversations between the Customer and BOT to mimic the conversation between Customer and Agent and then refining those using AI-powered engine were some of the key challenges. The team worked together to deliver very high degree of accuracy and it improved from a low of 40% to 90% in 6 weeks. Other challenge that we could successfully address was relating to the platform integration i.e with Core Call Centre platform, knowledge portal and core banking systems. One thing is for sure that challenges will increase as we have more dialects and attempt to handle disputes but I am confident that the team is capable enough to address them.