The global chatbot market is projected to reach USD 102.29 billion valuation by 2026, registering a CAGR of 34.75% over the forecast period of 2021 – 2026. Today, companies across sectors such as retail, BFSI and healthcare use bots to reduce operational time and improve efficiency. As much as 90% of the businesses reported faster complaints resolution with the bots, according to the MIT technology review.
Advances in Natural Language Processing (NLP) have made chatbots more accurate, intuitive, and self-contained. However, most of the chatbots available in the market are English-first or English-only. According to statista, only 1.35 billion of the world’s population speak English natively or as a second language.
Since 2009, Reverie Language Technologies has been delivering content in multiple languages in real-time to bridge the language divide. Reverie has now diversified into voice-led social commerce activations and Conversational AI for contact centre automation. Recently, the firm partnered with Boonbox (an exclusive rural assisted Commerce platform) to cater to the non-English speaking Indian community.
In an exclusive interview with Analytics India Magazine, Vivekananda Pani, co-founder and CTO, Reverie Language Technologies and Ramachandran Ramanathan, CEO, Boonbox spoke about their plans to build multilingual bots from the ground up.
AIM: How is Reverie building multilingual chatbots from scratch?
Vivekananda Pani: We consult with the businesses to understand “why the bot is needed”, “what languages are needed”, “are users comfortable typing in their language?”
Next, we understand the use-cases to be automated through bot. This is a step-by-step process to understand the desired user journeys for the different actions the BOT is supposed to do. The broad buckets for the same can be either command or transactional. Once we have the blueprint of the desired flows, we (along with our partner) convert the existing process to a conversation flow between the end user and the bot.
On finalising the flow, the deployment is directly worked on the chatbot platform. This bot development includes API integrations with any of the third-party APIs. Conversion of the bot flow development would take into consideration the usage behaviours of an Indian language user and tweak the flow accordingly. Once the bot flows are developed, the intents and entities are then identified for the Natural Language Understanding (NLU) training. It involves creating all the possible variations in which a user might ask in their language.
For example: A user’s intent can be to “check bank account balance”, in this case user might ask the query in following ways:
- English:
- “What is my account balance”
- “How much money is in my account”
- Hindi
- “मेरे खाते में कितने पैसे हैं”
- “मेरा बैंक बैलेंस कितना है”
- Mere bank account mein kitna balance hai
Indian languages NLU training is what gives our technology the edge. It’s trained on the native script, so the NLU is able to understand the context better and provide higher accuracy. Speech-to-text (STT) is also fine-tuned according to the use-case to provide highest accuracy. Once all the components are ready, automated testing is done on all the components to ensure stability and performance. Another part of testing is done by the language users to ensure the bot is able to reply to varying user questions. The cases that fail are then re-trained using a human-in-the-loop approach to make the engines and the bot robust.
Finally, the bot can be seamlessly published on a wide range of channels like IVR Lines, Website, Mobile Applications, WhatsApp, Facebook Messenger and many more.
AIM: How does the Whatsapp multilingual bot work?
Vivekananda Pani: A WhatsApp bot is just like another contact in your phone: Either bot can initiate the conversation, or you can start just by saying “hi” or simple commands like “do you have table fans” (in the case of social commerce interface). At the very first step, the user is asked about their language preferences, the user can either give the input through Menu based buttons, typing in the message and even through the voice. Users can always change the language anytime during the conversation. Once the user starts interacting with the bot, the message starts going through the Reverie’s Voice Suite to provide the replies in the selected language, either through text or voice notes.
AIM: What’s the tech stack behind your products? What languages does Reverie support?
Vivekananda Pani: Reverie’s Voice Suite supports English and 10 Indian languages including Hindi, Marathi, Gujarati, Bangla, Tamil, Telugu, Kannada, Malayalam, Punjabi and Odia.
The tech stack:
- STT/ASR: Reverie’s Speech to Text (STT) module is developed through the advanced deep learning technologies and carefully curated data with coverage of various accents, ages and genders.
- Language Detection: Language Detection is a critical component of the multilingual chatbot as we need to understand which languages are spoken by the user before it could be understood by the NLU engine.
- NLU: Reverie’s Natural Language Understanding (NLU) is developed using the cutting-edge transformers technologies that understand the Indian languages as well as English.
- TTS: Reverie’s Text-To-Speech provides the fast and natural human speech in both the genders with many customisation options to generate perfect responses.
AIM: Which sectors see the highest demand for multilingual bots?
Vivekananda Pani: Major domains where we see highest demands include:
Banking:
- Core Banking – To check account balance, transfer funds, mini statement, cheque book request
- Collections- Loan collections for different loan products via multiple channels like IVR, whatsapp.
- Customer Support- Information about products, ATM/Branch details
- For Employees- Information retrieval from banking process documents for compliance etc.
- Analyse – sentiments, tonality and agent performance.
Insurance:
- Complete customer onboarding journey including payments- Purchase Life & General insurance (Health, Travel, Motor etc.)
- Service Requests- Policy Renewals, change of nominee, address, claim processing, policy etc.
- Analyse – sentiments, tonality and agent performance.
Retail/e-commerce:
- Product Ordering- To pick and order products via multiple channels
- All transaction command
Set Top Boxes:
- Command and search use cases like – open, launch or play/show a program.
- Language and channel switch use cases.
High demand channels for the same are IVR Bots, Voice enabled chatbot, and Whatsapp.
AIM: How important is interacting in the local language in building customer trust?
Ramachandran Ramanathan: Boonbox has sold products to over 2 million customers in rural India. Our customers are assured, aware and desirous of improving their lifestyles. Boonbox is completely ‘local’ and our interactions with our customers are in local language only. Our customers hail from different states and one of the common characteristics is that they do not converse or communicate in English. Conversing with them in their local language means increased interaction and in the process, building trust. Be it verbal communication or through apps, the language is, by default, local.
AIM: What is the scope and potential of ‘voice commands’ with respect to rural and social e-commerce, especially through Whatsapp?
Ramachandran Ramanathan: Voice is intuitive, hence an ideal means of communication with any set of customers. ‘Voice blast’ of select messages to customer phones has always been part of Boonbox’s ‘go to market’ strategy. By and large, most customers in the non-metro cities are uncomfortable typing text, even if they have had school education. It is baffling that businesses have not yet realised the potential of voice in customer interactions. The key here is to make it obvious to the customer that communication through voice is an option. Boonbox connects with its customers and also drives business transactions using Whatsapp, wherein the use of Voice could be a potential game changer.
Vivekananda Pani: Over the years, WhatsApp has become a familiar channel of customer communication and is coming to the fore in the age of social commerce and Direct to Customer (D2C) players.
AIM: What does the partnership with Reverie mean to Boonbox?
Ramachandran Ramanathan: Reverie’s expertise in Indian language technologies and solutions make them a perfect fit for Boonbox where ‘local’ is the default option and the only way to build customer trust. Reverie’s role would cover Boonbox Apps, communication over Whatsapp and the use of Voice technology wherever possible. The Boonbox-Reverie collaboration has the potential to make eCommerce and Social Commerce second nature for Boonbox’s rural footprint.
