October 29, 2020 12:45 AM: Yellow Messenger gathers a billion data points every quarter and uses this to make its chatbots more accurate
One of the biggest problems with chatbots and conversational AI deployed on user applications and websites is their poor contextual response ability and lack of emotional quotient. Bengaluru-based Yellow Messenger (YM) tries to solve these two problems in this crowded space of chatbots and conversational AI. “We gather a billion data points every quarter and use this to make our chatbots more accurate,” says Vartika Verma, marketing director, YM.
Accuracy in chatbots means contextual responses and showing some level of emotional quotient, a trait predominantly pertaining to human interactions. To stand out in this crowded space today it is important to not just give a very human experience to the end-user but also have more than one channel of revenue. There are companies that focus only on enterprises by offering chatbots for enterprises to enable communication and operations within the organisation. One such example is automating an interactive payroll and employee grievance redressal. Another channel is making services directly available to the public via apps and client sites serving a wide purpose—– right from generating leads to aiding marketing campaigns.
Raghu Ravinutala, co-founder and CEO of YM
“We did not want to focus only on one channel. Instead we started building for both which also helped us innovate technically,” says Raghu Ravinutala, co-founder and CEO of YM. Ravinutala says its key customers have reported growth in revenue to the tune of $10 million. Its key customers include Grab, Schlumberger, Accenture, Flipkart, Dominos, Xiaomi, Roche Pharma, JD.ID and Bajaj Finserve. YM also generated a top line of $5 million in 2019.
So, does the rise of chatbots means the loss of jobs and other resources? Ravinutala suggests otherwise. “When we try to improve the workflow of our customers it does not have to mean replacing jobs with bots. It is how we can automate certain processes using the power of bots,” he says
According to YM, developers get a proper platform to solve problems and build solutions that they were not able to do before due to unavailability and limited access to various resources involved in developing. “We began engaging with the developer community in the geographies we operate in with an aim to familiarise them with our platform. As a result, we are able to help them add value to their projects with our developer tools,” says Verma. YM has a bigger aim of opening up its platforms to anyone with or without coding experience so that they can build their own chatbots, starting with training sessions and hackathons. However, there are many tech nuances to master besides just democratising chatbots and conversational AI. For instance, today largely the input is in textual format. There is also voice and video to solve for. With rise in virtual assistants this will have all the more value in future.
“We are simultaneously solving for other forms of input apart from text. This is a function of developments in both algorithms and computational capacities of the devices we have today,” says Ravinutala. Both developers and investors realise that in some key areas the computational capacities of the processors do not match the rate at which applications are getting complex.
While this is addressable in the long run, in the near term scale and revenue will come from integration of regional languages and other global languages. Verma says currently YM supports over 120 languages including Arabic, Bahasa, Bengali, Cantonese, English, French, Hindi, Mandarin, Spanish and Thai and the accuracy has not been compromised on any of these languages. Solving for languages also makes expanding into other geographies easy. “We see the scope of our developer community and multi lingual capabilities resulting in our own marketplace,” says Verma.
At the moment, YM is building its resources to enable the same. The company has raised $24 million in funding to finance its activities. The key to success, however, is with the developers. The more contextual conversations become, more is the risk of bias. This is prevented by the nature of the use cases itself, as the learnings from several enterprises suggest today. The end-user has the ability to flag such biases and the neural models that make such associations, irrespective of computing capabilities available today, will still be supervised by humans, keeping up the trust. Focus will shift to building for the world from India instead, given the growing importance on self-reliance.