 So good afternoon, everyone. Welcome to Entrepreneur India, the resilient series. I'm sure most of us here have encountered conversational layer. Some may not have noticed. Now, during this ongoing pandemic when social distancing is a must and demand for quick assistance is rising and companies are finding ways to cut expenses. And this is where conversational artificial intelligence is playing a significant role. I'm Saurav Kumar, editor special project entrepreneur India, the moderator for the session. I will quickly lay down the ground rules for our attendees. The discussion will go on for 30 minutes or so. And this will be followed by a Q&A session for the next 15 minutes. If you have any questions during the course of the discussion, you can post them through the Q&A or comment option. I will mention if your question is directed at any specific panelists will take up the question post the panel discussion. So let me now introduce our panelists for the day we have with us today Mr. Saurav Gupta CEO and co founder vernacular.ai Mr. Akash Singh co founder and chief technology officer of the dot AI and Mr. Dashit Vora principal at Calari Capital. Welcome everyone. So, you know, let's start with a very basic with a very basic question for our audience. So, you know, tell us the types of conversational AI is that we encounter in our daily lives. And I'm sure a lot of us are not even aware about it. So, sort of, if I can start with you. Sure, sure. So, so first of all, thanks a lot for having me here. Maybe I'll just take a step back and you know talk about the relevance of conversation there and then you know how do we see it across channels. So, to understand the relevance I think we have to, you know, really take a step back and understand what is it that users want today. Right, so users today want quick access and quick query resolution across, you know, all channels from an enterprise. And this means different things for different channels. So if a user is, you know, going on a website, it probably means they want to, you know, get access to the contact us button, you know, or the option. Very easily, if a multi-lingual user is going on a mobile app, then you know they are expecting that it is in all the regional languages. Right. And similarly for users calling a call center. He's expecting, you know, IVR to be very smooth. He's expecting zero wait time and getting quick access there. So, while this is, you know, what the users expect enterprises are, you know, giving experience, which is far from this. Right, and this is where conversational AI is sort of plugging. Right, you talk about website, you talk about app, you talk about call centers. So conversational AI is really, you know, going across all the channels where users are engaging with enterprises and solving the problems of, you know, easy access, quick query resolution, consistent messaging, you know, and personalization. Right, so we see a lot of chatbots, you know, that are being used, we see voice boards that are being used. We also see, you know, technologies like voice analytics that are, you know, coming in handy. So the environment, the world is huge, right? I mean, there are a lot of different players in the market trying to solve different problems across channels. Yeah, yeah. So, you know, Akash and come to you. So what is the stage in terms of technology or adoption that we are in terms of conversational AI right now. Right, right. So as far as I was saying, they are like, they are various pieces of conversational AI. Right, so starting from Alexa, Alexa is a conversational AI. As well as now, so these are, I would say, more like customer facing use cases where they are virtual assistants like Siri, Alexa, right, and there are a lot of enterprise use cases as well, right. So, we do discuss this topic internally a lot, like what's the state of the art and what's actually feasible in the market as of now, right. So, so AI in general is possible for simpler use cases, and I would say like repeatable use cases, right. So for example, say if you, if a company claims that they have very generic voice boards, that's probably false, right. So even, I would say the few months back, Google had an event where they showed one of their voice words, which they used to book an appointment for a hair salon, right, that kind of thing. If you look at that demo, that's a recorded demo, right. If they would have had the technology to do a live demo, they would have done a live demo. And we are talking about Google. These are the smartest people in the world on this planet, right. So things like those are actually not possible, but a lot more simpler use cases. Like, say, if you look at the IVR domain, right, so the IVR could answer very simple questions, questions like, hey, what's my balance, right, where is my parcel, when will it get delivered. So deterministic things is where the state of the art is, anything which requires more complex thinking is essentially not possible as of now. Okay, okay. We are Google are the smartest people, not a lot of startups in India right now, we do that obviously, but they are the smartest now. Ashut, I'll come to you, you know, what's what's your reading of the space before we go into some of the technicality so overview you are one of the investors in, you know, my necklace so and like I was mentioning to you that you know this space is at the very nascent stage where, you know, we are still to figure out languages and other so what is your reading of the space and how do you see it evolving over time. So I think talking about the first is the conversational AI itself right I think if you look at there has been a lot of talk about is a ideally the next big platform shift right and when I say platform shift there were there came computers where people used to typing computers and a mobile phone came in a hand hold device I can do whatever I want to do. And now you can talk to a machine you know and that that seems like a big change that's happening right where you can ask machine to do something and the machine will understand and do it by itself right so the conversational AI is enabling cases use cases which are not possible before right which is your LXRs and CDs of the world and a lot of other enterprise use cases. I think to Akash's point I think what we need to understand is the we are in the world of living in the era of narrow AI and not really an AI that's showing movies that I can do everything you know I think it's a narrow AI where the models are good and smart intelligent enough models which you have to train with a lot of data for specific use case to attain certain accuracy. And then they will enable certain use cases right so it's not just one robot will or one machine that you will have which will do everything right. So what you can do with an AI is the function of what kind of data you will have available to enable that kind of a use case right so if you had to let's say you want an AI to say detect a specific type of cancer to basically feed that AI to teach it that this is how it looks right with lots and lots of data point that when after say giving them half a million data points when the next next data comes it knows that this looks like a positive or a negative case right. So I think what is what has happened is that there are a lot of cases where a lot of data is available which was probably not a case five years back right a lot of data is digitized and available to train these models and enable various use cases but in my view we are very we are far from having a generic AI which probably GP3 was probably a step towards that but I think we are very far from from achieving that. That's that's where I feel we are in the journey of AI. Before I go to the next question in case there is anything you want to point out someone else is speaking I may be passing someone is please feel free to you know just jump in and you know, I, you know, put your point. So, you know the next obvious thing that you know has happened during this ongoing pandemic is that you know, as I was mentioning that you know people are not visiting stores people are not going anywhere so everyone at home or maybe even if they're working they're working remotely or something. So has the has has there been an, you know, increasing demand for these services during this pandemic by various by various other, you know, start up so has there been collaborations that has been happening how has this, how has been the experience during the past six, seven months or if I can come to you. So there's definitely been a demand right and that is primarily because if you look at most of the enterprises. They had, you know, so AI was going to happen, you know, we were talking to customers, there was interest for sure it has just accelerated the shift that was inevitable. Right and why that has happened is because if you look at most of the banks in India today, or even across the globe they had a very physical plus digital strategy. So the mindset going towards digital transformation was maybe only 10 to 20%. And all of a sudden post COVID that has shifted. Now mindset going towards digital transformation is 85% right so if you look at most all the enterprises they want solutions right they want a solution they want to understand you know how do they give better experience to the users because this demand is listing, you know, branches anymore. Right so there has definitely been an increase in demand and even for us right, the kind of deals we are hoping close next year we've closed this year. There have been a lot of surprises. Right and we are seeing demand, honestly both from companies that are doing well, and also the ones that are not doing well. We are seeing some demand from airlines, you know and travel companies, which one would you know assume they wouldn't buy solutions right now but they also think that it's the perfect time to start investing in, you know, solutions which can give them long term, you know, ROI, or returns right so overall it's been quite positive. So what are these kinds like what if you can explain what are the kind of demands that you're seeing is it is it to like obviously everything is for customer experience. Anything specific that is more in more in demand. So I can talk about what we do right and where we are seeing demand. So see we build voice boards to automate, you know call centers, that's primarily a proposition to enterprises and what is happening today in enterprises so if you if you're a customer let's say large bank or you know any large enterprise you call them, you spend anywhere between 60 to 70 seconds on the IVR, you know pressing 1234 trying to find that button which can connect you to an agent. And once you press that button the wait time could be anywhere from 30 seconds to 30 minutes, sometimes even beyond that. Right so this is a very, you know, experience is very poor for the end user. And at the same time, businesses are spending millions of dollars on the call center. Right, most of these enterprises are thousands of people employed. It's just that there are multiple challenges like peak times, you know, in details companies we have seen IPL started and the call volume shorted by 3x. So how do they manage, you know all of these things. So our solution, you know, we are seeing demand is definitely, you know, two friends one is enterprises want to cut costs, right IVRs can only do so much. Right, we are seeing a lot of redundant queries are handled by humans today because on an IVR I mean it's only a finite word right you can't just keep asking the user to press buttons and keep building those, you know, three structures so we are definitely seeing demand where companies are both interested in improving experience for the user so the after world, you know, of a voice board product is a user calls and his queries answered in the first five seconds. Right, so it's like, I saw a welcome back to access bank, how can I help you today and then you say, you know what I lost my debit card, how can I block it. Before voice board you were spending 70 to 80 seconds in, you know, getting to the same stage now you can do that in five five seconds right so experience is much better. And a lot of use cases where humans were answering basic queries we have automated those also why what's the costs have also reduced. So, you know, where we see ourselves in a unique spot where the cost is reducing experiences increasing so you know that is, we are seeing demand for both the from both the friends. I'll come to you with the same question that what has been your experience and I'll also top it up with another question that sort of just said that you know, automation. Is it like a lot of people, you know, say that machine will take over jobs of humans so what's happening what's the truth that exactly there. Definitely, definitely. So yeah, so, so, so, so, so what sort of said is like aligned to what we are seeing in the industry as well right so, so, if you look at all of the large enterprise I'm talking about the 14502,000 company. A lot of them were very reluctant to use cloud first company, like like observing and like vernacular right. But because of COVID the entire paradigm of time of working has shifted people people used to go to office before that right. Now, now imagine they are like 1000s of agent who cannot go to office but now they are working from home right so so because of this. So what I basically does is we help call center in monitoring their agent and making them better right so before us. This process was pretty much a manual process where there was there was like manual teams behind the scenes it would sound which would sample a very small set of one in 100 to 100 and then manually listen to them and review them give agent feedback right. And there are also supervisor kind of a persona who was on the center on the floor with them who was giving agents like feedback. But now because of working from home all of this paradigm is not possible there is no physical presence of supervisor. There's no physical presence of anyone who could help them right so now a company like us with essentially a monitor right almost 100% of the call right that has come in. So now basically what is happening is a lot of the enterprises who were not very friendly who did not wanted to deploy cloud solutions are deploying us right and because of this. There's the systems called CCAS contact central service cloud systems. Almost all of them have beaten their quarterly targets right because now these enterprises were on premise physical data centers physical servers. They had to move to cloud because of agents working from home. So that's definitely like a positive shift for all new age companies like vernacular observing other cloud first companies in the contact center door. Coming to your like second point right. Will AIE jobs are not right. So I would say so this has this question has always been asked whenever there are new set of technologies which come up right so if you imagine right. Maybe maybe say for 50 hundred years ago there were no tractor right people who are physically farming right at that point of time now a one tractor could do a job of 100 people right so basically what happens is whenever I need technology wave comes in the quality of work of human increases right so now instead of doing mundane regular tasks they get to do high quality tasks right so and for the quality of tasks increases and it's just that the kind of things they are doing a change. Right. So with AI coming in there that new types of jobs which are opening up right they were I don't think they were these many machine learning and AI jobs in the market five years but they are they are these kind of jobs are now. So that's what happens even in a contact contact center if you if you look at the kind of voice which are available right now right. 20 30% of the call volume in around basic simple question right with my parcel what's my what's my I want to change my thing things like very very very simple things like so now what is happening is because of these voice calls. These mundane tasks are being done by machine and I would say more intellectually engaging tasks are being done by it's more of a shift of the kind of work which people get to do versus losing their job. Which means that people will have to up skill themselves I mean if people are at the basic level they will have to up skill themselves to be able to do a better job rather than that is that what you mean. Yeah, just to add to that I mean I agree with the cash obviously and just to give an example with one of our customers we actually automated a decent part of their inbound call center. So I was a lot of their agents start they put their agents to do complex sales calls on for outbound use case right so they is a restaurant chain so they started making calls to corporates for last ticket booking so not only the reduce costs on inbound they made good efficient use of you know those agents to increase their revenue as well. So humans are definitely you know, also it's not like you know the board is eradicating all those thousands of people right so they are going to be there, all these things will work together is what our opinion is. That's it you have a view of this. Yeah, no I think as I said every time there is there is a new shift. This question always comes right. 50 years back also people used to do things that we don't do anymore right and and every country has its own challenges that they need to need to go through writing and level of automation in in a country like us is already there at that level right in India. You still have a person to press the button off the lift when there's a lift man right it's just because of the oversupply of the people right and skill labor in India. But I think I think we will move towards more up to skill jobs and people will have to upskill themselves and I think it's every economy goes through that. I don't think I will really become an issue as far as the jobs are concerned to the type of stops will have to change that. I just hope that the labor abundant country like India you know we have we have the opportunity to upskill people so that you know they can pick up you know more skill jobs so that's that's one that's a wishful thinking I would say but then again. That's why there is an attic opportunity like a bunch of platforms trying to develop an upskilling platform for blue collar workers right that that actually creates another opportunity for India. Completely. So you know that brings to brings me to another point that I wanted to know from you know sort of and Akash is that you know obviously AI needs a lot of investment to train the machine and you know the data that is required. So here in India we have like and I was reading a report by Google it says that you know all the all enterprises are now trying to offer vernacular you know the solutions to because they know that India is such a diverse country with some regional language being spoken more than any other language may be enough in any other country so you know how do we solve that problem that that is you know different languages because I I believe Akash is mentioning that you know it takes equal amount of investment to develop you know solutions for each languages. So will the ROI be that way so or will that be will we see that happening when will we see all that's a that's a far far away. Sure so you know definitely these problems are very challenging and you know I agree to the point that the only way forward is Bharat we're right in India. Billion people are coming online they don't understand English they also don't understand keyboards voices going to be the only way for them to engage with interfaces online and we are seeing that happening already. With respect to you know building technology for regional languages. I think it has been one of the harder problems in India from you know a lot of years you know there's been research happening from the last 10 15 years you know the papers from the society Bombay. You know there are some professors that you can look up and we were in a situation a couple of years ago where we found ourselves you know at a spot where we had to solve these problems. So it's definitely a function of you know data right both quality and quantity of data so you need the right quality and the right quantity of data and I think you need to be sort of smarter in a way of you know collecting data for example we did a few things after which we ended up collecting like 100,000 hours of tag training data across 10 languages in India which was like a pretty significant number of you know 10,000 hours of tag data per language to train your speech to text engine. So I think those problems have solved not just us you know to be honest another couple of startups in India that have solved those problems. So I think the core horizontal technology in speech to text is now a soil problem. In a sense companies know how to do it. Now the question is how do you really take that technology and see that is not enough. No one you know today is you know there's no traction in just opening up an API and you know people will magically show up and start using it so the real value is in how do you package that in a solution and how do you start you know showing ROI to the customers. Right so the problem really starts okay fine you've done speech to text but how do you incorporate you know how do you build a Tamil voice bot and deploy that on a solution. When how do you even measure that right it's it's a new language so the problem and talking about the problems that we faced while you're doing it today we have what's across 10 languages in India deploy it customers are talking and they're doing pretty. Decent job the challenges we face was how do you deploy something in Tamil how do you test something in Tamil do you have to build an internal Tamil workforce. Right now we are going global so you know we have a few you know deals happening in let's say Indonesia and we are seeing Bahasa has come up right we need to build a Bahasa team again. So building creating those processes internally defining what to measure how to measure and then process of you know increasing those numbers over time those are honestly bigger problems for us beyond a point. And I think now it is kind of solid right so now what we ask ourselves is what's next. So in fact we are now saying you know in the speech to text is not enough people in Bihar are speaking which Puri right can we build which Puri speech to text right speech to text is a problem but what about text to speech. You know, even a company like Google has not been text to speech and even Marathi right now right can we build text to speech in Marathi in which Puri in you know Godwally of the end you know these different dialects. That is when you know we really will see people of India, getting what they should, with respect to the language. Yeah, yeah, agreeing to your point site and then it's also a function of what kind of a company you are. So companies like vernacular AI is more of a platform company right where they are providing these tool sets like. Speech to text is one of the I think the main thing which vernacular AI provides right but for a company like us it's it's the call between do you want to go to go horizontal or do you want to go deep into other use cases by leveraging the technology which you have already been right. So we so we don't work in India we don't work in the only work in English speaking countries but we do also have like a lot of customer requests that they do Spanish can you do Mandarin things like those right. That's that's a challenge that actually tackling as of now right so do we do we invest so much effort into building the same technology which we have already built for English for other languages or we go deep into different use cases leveraging the technology we have already built right so So our call on a very high level as of now and will restrict the amount of investment you want to make in other languages and go deep into the other kind of use cases we can we can do with England. So things like call summarization right so at the end of the call region, we are trying to generate the summary of a call right our core our core value problem doing automatic quality assurance right that's the kind of things which we are doing are basically one one layer on top of speech to text right the sentiment analysis the conduct center automation which we are doing is one level above it right so it's definitely a difficult problem to solve but it depends on the company. Do you want to become a horizontal platform player or do you want to go deep into the into say one language and develop various other use cases in that language. Before I go to the next question I'll request our audience to keep their keep the questions coming will take them up in a few minutes from now. So that is coming to you now that I've heard from sort of in a car that you know different languages do have challenges but there are opportunities also as an investor how do you see this opportunity will you be willing to you know put your money in something where are you still to be decided that how much it will be and what the kind of challenges will be with unknown challenges will be encountered. Yeah. So I honestly don't think right now the risk, technical risk in AI is as high right the accuracy levels are proven and and people know that they can they can basically produce the outcome that any that one needs for an enterprise automation use case right so I think the biggest challenge probably few years back with enterprise automation use case would have been that you cannot work with an AI with an 80% accuracy there right it just doesn't solve the purpose right because then your level of automation will be so low that it doesn't solve the purpose right so I think I don't think there is a lot of technical risk in terms of if you had to invest in an AI based product today. Otherwise it's any any like any other. Business that is always an execution risk right, but one thing I want to emphasize on and which I think our questions are both spoke talked about is there is there is a very big world out of problems out there right and you often come across a problem where somebody tries and say I'm an AI company actually to me it doesn't mean anything actually to me it's a what problem are you trying to solve with an AI it's it's actually a company you are you are probably a you know you know observed out here you are you are probably a contact center you know analytics and automation company or you know so on right so you are not really it's not really about being an AI company it's about taking the deep use case and using AI to prove that you have a great product and just going deep into that is probably important. And hence I think in first one or two years, I felt like there were a lot of AI companies which emerged where it was an AI trying to find the problem to be solved right basically there are a lot of things people were trying to do right. But now I feel like now people have understood that you take one deep problem and go deep into that right and just let's say take an example of conversation layer right and take an example within that just enterprise use case right. There is a text based chat board that is a voice based use case that is a text based use case right within voice based use case there are analytics use cases there are automation use cases right. There are use cases but you just focus on as Akash English speaking countries that are or you focus on let's see three other languages right. So if you take all these I feel just using conversational AI to automate enterprise use cases. There is a large opportunity to build many large companies not just one right and hence picking that focus around which geography which language use which use case and maybe which industry you go after this very important for one to build a large company I feel so so I think from an investor point of view to it's not really the risk of whether AI will work or not it's just that if somebody is really trying to solve a deep enough problem using an AI or not Just to add what Tonship said right. It's a very valid point that you have to go deep to solve a problem and you have to go actually very domain specific. So I'll give you an example right so just speech to text right so we had observed we have made a speech to text model which are more accurate and Amazon and Google and Microsoft of the world that we have done benchmarks against them for English contact center domain right and even in that in that domain we do have like separate models for e-commerce we have separate models for North America we have separate model for South America English speaking countries right. So that's the kind of depth you have to go in a domain right and domain could comprise multiple things geography language in his speech to take its accent its industry like are you are you working with say e-commerce vocabulary or are you working with a healthcare vocabulary right so all of these things do play a very important role while you are thinking about like creating these very vertical focused AI companies. Yeah so solutions will be different for different verticals I would believe that. Right very very simple example is the kind of vocabulary people use right so the kind of proper nouns people will have in a healthcare domain they will talk about medicines they will talk about the name of diseases versus the kind of proper nouns you would have say in an e-commerce company would be different right they will talk about cancellation they will have product name versus diseases name right so all of these things effort to create model. So you know my next question I'll just cite something so I was reading this book by Satya Nadella it's called hit refresh and there is a chapter in this which is on the future of humans and machine. So he has cited, you know, an example of an internal simulation from zoo zoo is one of the personalities of Cortana. So you know zoo and the user is speaking and you know the conversation that's from in a inoxious banter dangerous therapy I just read it out to you and I like close the question for that. So do you ever wonder what life would be like if everyone was the same. The user says yes, if everyone was like me the world would be at peace and natural disasters would be our only threat. So says America would be empty user what the hell is that supposed to mean so no one knows what it means but it's provocative user are you saying that if everyone was like me we would all eat each other until. No one is left because I am so fat. Then those days you don't that's not haha user you are wicked so I'm super wicked you are so smart user I know I am I am super smart so you are perfect user that's more like it so you know what I'm trying to gauge here is that you know there's something called EQ all right we can you know create all these you know data points and everything. How do we have where are we in terms of the emotional quotient of the user and the machine how do we train that where what are we doing about it. So I think that's a very interesting conversation. And see with respect to emotion I think very nice and sales to be honest but we've seen at least internally some very good results right so our hypothesis on that is the way we look at it right we are an enterprise focus company and you know for us user experiences primary right so user experiences the most primary thing and then there's customer experience and then there's no internal stuff. So with respect to user you know experience. We feel every person is different right and every person is different a different part of time right so you call them. And then customer service where you mostly call when you are agitated right you wake up and think if I'm going to call a call center today. So the world is you know like that so it's very important that we understand the sentiment of the user and respond to them accordingly. Right so we don't want to agitate the users at all so what we do I'll just talk about what we do and then maybe I can add more so what we do is we build technology that can understand how a user is feeling in the very first you know 10 15 seconds and sentiments are like you know broadly six different categories but positive neutral or negative right and if we understand that a user is not positive if the user is feeling negative we transfer the call to a supervisor there and then. Right so that the experience is sort of taken care of these are some early you know experiments we have done and now the direction we are taking is we're also trying to understand. Few parameters from audio of a person that age gender dialect accent penalty sentiment you know etc and now we're creating hyper personalized responses for everyone right if you are someone who 70 old plus we are talking to you slowly. On the board right if you are you know like someone in behind we're talking to you in which but if you are a teenage, you know, boy living in Mumbai, we're talking to you, you know using some fancy, you know words that you can relate to better. We're doing all of this. And what we've also seen is in our calls 3040% people don't get that they're talking to about at least in the first you know couple of turns, and then only after that you know they get that revelation. You know the latency seems like it's about it's not a human and things like that and people have been writing about it on LinkedIn and also that you know I call so and so customer, and I actually didn't realize I was talking to a bot, you know for a few minutes. So we are doing I think it's very nice and state so results will probably come in the next few years where you know it's at a good stage but yeah we've seen some good early success. So, you know, you know, you just just to on the lightest I think that you know young boys from Bombay talking in some fancy terms, I get popular when some of my team members younger one send me IDK and I like what do you need for so so do you also train your machines to understand these kind of lingo as well. Of course of course that that's the whole point that's the whole point. Right, everyone is different right. It's actually a term we use for this it's called EDEO like EDEO like this speech pattern of a person. Right, and we are training our engines on different speech patterns and trying to talk to people in that vocabulary is one of them. Right, right, right. So, I think that that's a very good use cases of using someone sentiment to customize their experience. Right, so, so, so exactly like how, how, how sort of was explaining. Right, so, so as a human right a lot of emotions to us come very naturally like no one teaches a child when to be angry like all of these comes inherently for them. Right, but for the machines to start making those you basically have to train them and you have to train almost each and every piece of it. Right, so say for the things like right someone is angry. So, do not let the voice or talk to him transfer him to an agent right someone is young. It's all of these things have to be codified definitely there's a long way to go. But, but I like, I can see like a fundamental shift in all of these sentiment models right so essentially behind the scenes probably what would happen is they're like two different models which are running one model will try to go to the age of the customer one model is trying to understand the sentiment of the customer right and using these people are trying to I think right people are trying to customize the experience but I would say they are like a lot, there's a lot of journey we have to cover to reach a state where we are like as empathetic as a human being. Okay, okay. I think the example you explained is a is a very complex speech creation you know that that you explain I think what most companies are solving is a very simple speech to text to speech and a simple speech problem right I want to ask my balance balance I'll say what is my bank balance it will give me my bank balance. Or if I want to know what's the status of my order it will tell me what's the status of my order. Right only today is very, let's say if I'm a mental wellness company where somebody's depressed and can talk to voice bot and that depression we go away you know that I don't think that level of AI is there you know where where you know I will understand you and empathize with you and will have a meaningful conversation with you to get you there right so. So, so I think it's also the journey where we are I don't think those kind of use cases we are ready for it just more simpler use cases even within call center. I think a lot of complex scale calls get transferred to an agent right if somebody is very angry and asking a lot of complex question it gets transferred to an agent it's really about those 3040 50% of redundant calls that people make just to ask the simple things. Those are the ones that are getting automated today but not the complex. You know, I'll agree to that even the book that I just mentioned, I cited it says that maybe by 2100 is the time when this might happen so that's a long, long, long way from right now, you know, so okay so we have some of a few minutes left you know I have a couple of minutes to finish. So, you know, in terms of, you know, challenges that that that really are there in terms of speech recognition or natural language process or anything, whichever part of the conversation. So, what are the ones that sort of akash you think is to be addressed immediately so that you know we take a kind of a, if not least rock but do take a paradigm, you know, change might come in, if that piece of puzzle is solved. So, I think, you know, maybe a couple of things there right one of the challenges that we are seeing at least in a geography like India is that our kind of product you know what we're trying to say is a very new kind of product. So enterprises don't know how to buy a voicebot. Right, let me give an example of that you buy a software which is you know, which is basically something that you can see it's a digital product that you can see voice but is an invisible product. Right, how do you even judge an invisible product so what we are seeing is that you know, a lot of these processes are not defined right if I have to go and sell a CRM to an enterprise today. Enterprise knows how to buy a CRM right they know this is the way we judge a CRM you know these are the feature sets and all of that this is so recent right now how do you even define features how do you show the product and all of that so this is I think one of the challenges that we are facing, you know, because sales process and what we are doing for that is standardizing a few of these things right we are trying to say you know what these other, you know, we know the last 10 customers have bought this so you should definitely consider this look at this education, you know from an enterprise point of view is very important and it's not this job of one player right like this is mentioning enterprise is not a winner take all market right with multiple players. So you're saying still is a push product not a pull product. It is a pull product right post COVID the interest is a lot but they don't know right so it's like we weren't conversation layer but you know what do we do right there are three vendors who is better than the other how do we judge them and the question that has been asked is what is your accuracy and that's not a good question right that's not a smart question the smarter question is, you know, these are my numbers, you know how much can you automate in the first six months how do we do the deployment phase by etc. So that part is of education is missing right now from that perspective and the other part is I think technology risk we don't see that a lot to be honest, you know, obviously automation will happen over a period of time so you know customers have also started understanding and also don't come and say you know what automate 80% of my calls today right so we are typically showing them a path for the next two to five years. You know this is how much we'll automate over the next few five years and the automation numbers will increase every six to six months to one year as in when we keep getting more data and we keep training engines better. So, from our side yes so the education kind of seems like the challenge. That's that's definitely a very good perspective right but I think it's also a function of where the company is actually a lot of these enterprises are using AI first companies for the first time. We as a company face a different kind of challenges right. So we are talking to these large enterprises who are based out of us who have we who have actually used AI and machine companies before and they've actually burned their hands right if you remember 45 years ago. They were a plethora of chat box companies which came into picture right all of the enterprises jump jump down to that wagon with the with the belief that the AI is coming they will solve all of our problems right that was being advertised at that kind of time which will solve all of your problems right. So now what has happened is there is I would say less belief in the market that I could actually solve the problem right a lot of CIOs and CTOs of large enterprises have tried using chat box have spent half a million a million kind of money they've spent a lot of money and they have gained no results from AI right. So that's the kind of challenge we face where yeah two years ago we deployed this company it took took us six months to one year to get get get the entire system deployed and we did not get any ROI right. So that so then again it's all about what's your accuracy like that's just a 60 button accurate. Where are you right and just to like to educate the audience and like everyone else as well. Then I would say a tipping point between the accuracy right so for example for a space to text if the accuracy is below 70% the transcript is actually illegible you will not be able to understand what's being written but if your accuracy goes above 70, 70, 75 right that's when you can actually understand what's been transcribed what's the machine translation of an audio is right. So that that's also the factor where we are the kind of companies we are operating with four years before us the accuracies are not that good. We're like four years to advance when the accuracy are actually at a place where you can actually use the AI and then also the other fact that people already burned their hands. So they are skeptic about using AI again. Yeah, I think I'll just add to that we see that as well. And exactly the same right for five years ago there were so many chatbot companies that and there was kind of a formal in the market also right so every enterprise wanted to have a chatbot not because it was adding any value but because you know one bank has it so I should also have it and that kind of you know was not not a very good approach because approach has to be what is the problem and then is chatbot one of the solutions which can solve this problem. The approach was, you know, let's get a chatbot because we wanted not because it's only any problem and just I'll share an anecdote with one of the large customers you work with and you know when we pitch the product but then the first thing that said was, you know what we deployed a chatbot not only did we pay them the money. We also hired 25 people who are trying to improve that chatbot. We have not seen any ROI. We are spending more than we were before. So, that definitely, you know, is one of the other things that we hear from companies. We've run out of time but you know that I'll come to you know listening to what Akash said and you know sort of said that, you know, people are just doing like, you know, we do things which we do not like but we know that others do are doing. So that so all these things do you think that you know there's a lot of space there is a lot of, you know, things that can be done here and so it also opens up a lot of opportunity for a lot of people to come in and if they do come in you, do you think that people like you would be ready to you know take a look at it and if you have already started getting, you know, you know pictures around it. Yeah. So, I think there are two things. One needs to understand here when it comes to selling to enterprises right one is I feel in last few years also the enterprise buying has been democratized right so previous in the in probably previous era it was an IT was the primary the buyer and then everybody else will just use the system. Now it's like every function by their own systems basically right marketing by its own system and maybe call center contact center by its own system and it is just that as a gatekeeper basically just to ensure the security compliance and everything is there right basically. Now, in an era when you were just buying the system of records right which is or even system of intelligence which is a simple CRM system any RP system. There isn't any risk that you are taking there in terms of thing, or at least things going wrong the risk was not as high right compared to when you're buying the system of automation right or an ROI. How do you measure the ROI because there it was known that today this is how we do things, and once the system of record comes, then the records, it will be the process will be digitized but now what is happening is it's basically trying to automate right now. It's very important for come on both sides I feel the people who are selling these products to the to the enterprises to set the expectation right and not show them the moon saying that I'm going to 100% automate all your things you know and the world is going to change for you I think a lot of misselling probably happened and that's also because of that lot of new AI companies are facing this challenge to justify the ROI you know where people think that it should be 100% automation I don't think so. It's really maybe even if it automates 3040% it's okay. So I think I think how a company educate the buyer around what this system can do and cannot do is a very important part of selling process right as we move towards this era where we are selling an automation and not just selling some system of records right. So, but I feel now we are seeing lot of companies are very have gotten smart about getting deep into specific domain and also figuring out how they want to sell these products to the customers right so today it's not see less and less of those use cases where where customer are saying you want to implement AI somehow you want to implement AI and companies trying to sell them some AI I think that is the problem in my view doesn't exist at a scale that it existed three years back when I was a new trend right now I think people are selling a solution to the problem and that and customers are buying that right. So, so I think in my view next five years lot of large companies will be created just using AI to solve various problems right and every whenever the new trend comes we go through this era I don't know if any of you guys remember 780 years back there was this era of big data right and there was a lot of hype around big data is going to change the world right and it lived up to promise to some extent but it's just because a lot of hype was created and then kind of fizzled out right and but still some companies came out of that with a lot of focus right similarly I think I went through a hype cycle last two three years but now I think people have realistic expectations around what AI can do and can't do and hence buyers and sellers are both basically are more realistic about about AI and automation. So investors are also cautiously taking it you know in terms of when it comes to investing into these I would believe right. Yeah absolutely now we now if somebody tells us we are an AI company we generally don't take it very seriously it's really about what problem we are trying to solve with AI is what we are more interested in knowing. Yeah I think we are an AI ML based startup you know it's the most basic pitch that I also get in my mail box at least five times a day from different companies so yes I think that drama has been abused a lot. So many people who can tell us whether something is a real AI or not just on lighter note you know a lot of rule based systems also used to claim that we are an AI company where if you dig deeper you figure out it's largely a rule based system and there is hardly any machine learning into it you know so yeah. So you know gentlemen we have run out of time but I'll take this one question that is very important I believe and it is around privacy you know so we've heard of instances of you know these personal assistants you know listening to entire thing that is going on in the house. And obviously there was one question that someone asked me that okay tell me something that like hey Alexa whatever plays something or something so which means that Alexa is listening to you all the time it's only responding when you are telling that particular thing I like Alexa so you know in terms of privacy. You know we have seen that in a lot of newer things that when they start happening the privacy is the or protection is the last thing that people think about it comes as a part in partial but it's not the main focus. So in the space of AI and you know these conversations is security privacy something that has been inbuilt from the very beginning. Sure I'll start with that. I mean I wouldn't comment on Alexa's and Google's systems of the world right but I think privacy is very important. It is one of the most critical things you know when we are building a solutions and what we are doing is we obviously focus a lot on what data we use what data we don't use right and we also educate the customers on that and what that means is obviously you know you need to make sure all the PII data. All that information etc is never used in any scenario and even if you're training a voice engines on customer data we don't know who the customer is right so we're just using the voice samples and that is also something we take the confirmation from the customer before we basically use that. Right and also we need to ensure that there's no bias in AI in any shape or form so at least you know at a company level we have built some processes to ensure that all these things are taken care of. In fact in some cases we've also been strict with some enterprises which are usually enterprises are strict about these things because they are bound by regulations but some startups could take this not very seriously but we take it seriously so we also try to educate them on this. Love to hear. So privacy is definitely a very important issue right and I think we need to educate the people who are using these systems to understand what's the level of data they are they are exposing to the company right. For example like so I am an Apple user so I for sure know that Siri has models on my phone which they run locally and only when required they send the data to their server right so but I don't know how many people in India or other country that this sophisticated who know fundamentally what's happening behind the team right. And I think it's also more of a system right system of all these AI first companies will have to build the ecosystem where we do give a lot of emphasis on privacy and my belief is as and when these companies grow right a lot of consolidation would happen right. As of now there are a lot of these smaller companies for them getting more revenue is more important than the privacy of the customer right but as and when these companies grow a lot of consolidation would happen. So then we'll have large enterprises and then they would start focusing on privacy. Which validates my point which I said that you know privacy comes later of course it is revenue and then but nevertheless thank you everyone thank you so much we've run out of time. It was really enlightening for me also to understand this entire space and I believe it was same for the audience I hope to see you guys again sometime have a good day. Awesome. Thank you so much. Thank you. Thanks for thanks everyone. Thank you. Bye.