Raghu Ravinutala was bitten by the entrepreneurial bug long before he co-founded yellow.ai in 2016 with a vision to make customer and enterprise interactions more effective and effortless. Previously, he worked for 16 years with leading tech companies such as Texas Instruments and Broadcom, taking leadership roles in engineering, product management and business development both in the Bay Area and India. He is an alumnus of NIT Warangal, enjoys a game of squash every now and then and believes in practicing an active lifestyle.
We talk about conversation automation platforms and the AI underlying it, and the work yellow.ai has done with ABS (American Bureau of Shipping) and their Digital Fleet offering.
Entrepreneurship requires a lot of mental stamina and a lot of perseverance. Exercise and activity improve your mood and give you a lot more energy to do what you want to do.
Conversation automation platforms actually replace or automate a lot of interactions that happen between an enterprise and its customers and employees. For any business, including maritime businesses, a lot of the experience that the customers receive is through conversations that they have with your company and how employees interact with your company.
What conversational AI does is use software to understand the conversations and then integrate and drive workflows and transactions, without needing a lot of manual work. To give an example, somebody wants to inquire about a maritime schedule. Instead of them waiting on a phone on a call for some time to get the schedule, it's a digital assistant on the phone that answers the query for the customer, enabling a really fast turnaround time 24/7. You never leave a customer waiting just because someone in the response team is unavailable. That’s the core value of conversational AI.
This gets to the core of every business - great customer experience. Conversational AI is available across geographical regions and multiple languages, as well as across apps like WhatsApp and Messenger.
A fundamental difference between traditional user interfaces and conversational interfaces is that your options are unlimited in the sense that you are reacting dynamically to exactly what the customer needs. In traditional interfaces, you have to first ‘guess’ which options the customer will require, with no feedback loop.
You're also not constraining the user on language. Your website could be in English, but the customer could be Danish or Chinese and still have a natural natural language based interaction. Finally, the channels are more ‘natural’ - you can call, use an app, etc. Conversational AI helps automate those scenarios and make those analog calls into digital ones.
Where AI works best is for transactional tasks where the user wants it done really fast. Bots and virtual assistants are great at this, because they can understand and get it done in 15 seconds. Other interactions are more emotional - for example somebody is looking to complain about a bad experience. At this point in time, virtual assistants and chatbots don't actually give that kind of empathy. So for those kinds of interactions, the human experience is still superior. It's about bots and humans working appropriately together on the right set of interactions to give the best customer experience. Any conversational AI platform needs to be capable of orchestrating the interaction across automation and humans and provide that stability.
As an example, take somebody calling to reschedule their voyage. The automation right now understands this transactional task, interacts with the user and integrates with the backend system and gets the task done. The next question the user asks is around COVID - he’s worried about COVID protocols etc. Here, the platform detects a need for human intervention and automatically transfers the conversation to one of the customer support representatives who can see the complete context of the conversation so far, and then provide reassurance on COVID protocols followed and reassure the end customer. This is how the platform enables a seamless transfer between bot to human and then from human to the bot to actually provide that interaction with the end customer.
The yellow.ai platform can detect emotion, and then move the users based on the emotion it perceives to human interaction.
It’s a dialogue, and people like dialogue. In a website and app, you have 100 different options to choose at any single point of time. With conversational AI, there is a response for your specific query or interaction. That increases the chances of the customer continuing the conversation rather than abandoning it. In marketing scenarios, customers interacting with a conversation have better conversion rates than those interacting on a website module, or a form or a survey.
With ABS, yellow.ai is helping the end consumers and their employees monitor equipment, daily fuel consumption, voyage planning, and tracking the fleet. ABS employees can get information without needing to call their agents. This is being deployed on the Digital Fleet website. The plan is to scale the deployment to telephony, mobile apps and Facebook Messenger etc. The goal is to create an automated way of providing information for ABS customers.
To get started with conversational AI automation, you need a starting level of data to train the AI models. This can be obtained in two different ways. It can come from the prior call logs or chat logs, pre-automation. Recorded utterances are used to train a natural language model to understand those interactions. The second way, in cases where such data is not available, is to generate an initial dataset from scratch. The chatbot is launched, trained on this data. As the interactions grow, those interactions are used to train the system to improve accuracy. When good data is available, you can actually start at 80-85% accuracy from day one. When you don't have the data, and you're using manual training, you start at 60-65% accuracy. Once the interactions start happening on the platform, the accuracy reaches about 90 plus percent.
The yellow.ai platform is configured using a low-code approach. You don't need to write a lot of code to set up the platform for specific use cases. It comes with pre-built modules for typical use cases. The data needed for those repeated use cases is already available and they can be used to build a starting version for a lot of customers. For custom requirements, the user can train on available data, integrate their applications and configure for almost any use case, on chat and voice and across multiple different languages.
One might think that large enterprises might decide to build their own conversational AI platform from scratch. But companies like Roche, Schlumberger, and Alestorm use the yellow.ai platform to develop bots. They are looking to use the already built components from platforms like yellow.ai and then develop their custom use cases on top of such platforms.
The yellow.ai founders initially built it as a consumer application, where consumers can come in on a messenger app and choose a business and interact with them. But they discovered that companies really want to power their interactions on their own platforms, or on their own digital properties, rather than third party properties. So yellow.ai pivoted from providing our own application for consumers to actually powering these interactions on enterprise websites and mobile apps.
The second big learning was that there are two ways of implementing a chatbot service. One is just providing information, the second is actually enabling the transaction. To give an example, when somebody is looking to reschedule a voyage, we could just respond saying that, hey, we understand that you want to reschedule. We raise a ticket, and it will be rescheduled by someone later. So you're automating the conversation, but you're not automating the task, originating from the conversation. But if you automate the tasks originating from the conversations, the customer experience is much better. That is why the yellow.ai platform is now a workflow engine that is integrated with a dialogue management system to automate the tasks that originate from these conversations. Integration to third party tools via APIs, however, is left to the customers and a system of partners, as that is outside the core focus of yellow.ai.
Thes pivots were tough decisions, and the company is still learning. Raghu believes these choices will help yellow.ai scale much faster. With rapid customer growth, the company doesn't need to scale to a lot of internal teams. Instead, yellow.ai enables end customers and partners to deliver these integrations and configurations themselves. Yellow.ai partners with some of the largest system integration companies like Accenture, Infosys, Tata Consultancy Services and a lot of boutique partners as well, who are well trained and certified in the platform to deliver these services.
The difference that yellow.ai has seen in the maritime industry vs. other industries is that right now, maritime companies are adopting new technology as pioneers. Digital transformations done by companies like ABS are almost the first in their space. This is, however, rapidly gaining momentum across multiple different companies in this space.
“Conversational AI platforms need to be capable of orchestrating the interaction across automation and humans.” - Raghu Ravinutala.
“It's absolutely important for companies to store voice data even if they're not starting on conversational AI right now.” - Raghu Ravinutala.