AI Minds #052 | Simrat Singh, Co-Founder at Hooman Labs

About this episode

Simrat Singh is the co-founder of Hooman Labs, a startup building voice agents to help companies communicate with their customers over calls more efficiently. Simrat graduated from IIT Kanpur in India with an engineering degree and then spent a few years at McKinsey and Warburg Pincus before starting Hooman Labs.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of AI Minds, Simrat Singh, Co-Founder of Hooman Labs, joins us to discuss AI-driven voice automation and its impact on customer service.

Simrat shares his journey from consulting and finance to launching Hooman Labs, inspired by a realization during a trip to Vietnam with two college friends.

He explores how AI-powered voice agents streamline customer interactions by integrating transcription, LLMs, and text-to-speech to automate workflows.

The discussion covers the challenges of deploying voice AI in India, addressing linguistic diversity and differences in cost dynamics compared to Western markets.

Simrat highlights how automation enhances efficiency by reducing wait times and improving customer experiences, surpassing traditional IVR systems.

He emphasizes that AI adoption is about solving real business problems rather than just deploying cutting-edge tech. This episode offers insights into the future of AI-driven communication and its potential to revolutionize customer service operations.

Fun Fact: The idea for Simrat Singh and his friends to start something together surfaced during a New Year's Eve celebration in 2022 while they were on a beach in Phu Quoc, Vietnam. They realized all three shared a similar interest in starting their own venture.

Show Notes:

00:00 Consulting to Entrepreneurship Journey

05:16 Automating Call Center Operations

08:59 AI-Driven Conversational Solutions

13:40 Voice Agents: Market Dynamics and Adoption

15:52 Improved Multilingual Support System

18:17 Simplifying AI for Practical Use

More Quotes from Simrat:

Transcript:

Demetrios:

Welcome back to the AI Minds podcast. This is a podcast where we explore the companies of tomorrow. Built AI first. I am your host Demetrios and this episode is brought to you by Deepgram, the number one text to speech and speech to text API on the Internet today. Trusted by the world's top conversational AI leaders, enterprises and startups like Spotify, Twilio, NASA and Citibank. For this episode we're joined by Simrat, the co founder of Hooman Labs. How you doing dude?

Simrat Singh:

Hey Demetrios. Doing well. This is around evening time in here in India, so just about wrapping things up.

Demetrios:

Well, I want to briefly go through your journey because I find it fascinating how diverse it is. You got done with university and you decided to go the route of being a consultant for a little bit before a stint in finance and trading. Can you talk to me about that part of your life?

Simrat Singh:

Yeah, absolutely. Soon, around the time that I was finishing university, I realized that I wanted to explore businesses a little more and how they work, how do you build them? So I was told that a management consulting stint at a company like McKinsey would be a great way to kickstart my career in that direction.

Demetrios:

Did you get to see a bunch of different ways that businesses work?

Simrat Singh:

That's the funny part, because while I thought it would be an insider view into businesses and what makes them work, what doesn't, I realize that consulting for the most part is you are at an arm's length distance from the businesses that you're advising and there's not as much skin in the game. So while I did get a good understanding, especially for somebody soon after college, I think it was a great place to be. But I still did not get the depth that I wanted. So the next step for me was to take a little more ownership of the businesses that I was working with. And investing seemed like the right next step. So I started working with a private equity firm where we evaluated mostly late stage companies, mature businesses, and decided whether they would make a good investment over five to seven years or not. So again, while it was I was closer to the business than I was in consulting, it still wasn't giving me the hands on experience that I had originally set out to get. So then I decided that I want to try my hand at starting up or at least join a young company because that's where I would actually be on the ground floor of getting things done.

Simrat Singh:

Fortunately for me, two of my friends from college, very dear ones, also expressed a similar interest. They were in very different spaces, professionally speaking. But they also wanted to build something from scratch. So the three of us decided to team up and start together.

Demetrios:

So you got the band back together.

Simrat Singh:

And you recognized there's an interesting story there. We actually realized that all three of us were interested in doing similar things on a friends trip to Vietnam. So one night we were, I think it was New year's Eve of 2022 and we were sitting on a beach in Phu Qua around a fire discussing life and things. And that's where we all realized that we wanted to do the same thing. And we decided why not do it together. So that's how things started actually.

Demetrios:

Very cool. And so it was birth in 2022. What did you do to come up with the idea and how did you go out there and recognize that this was the area that you want to play in?

Simrat Singh:

Sure. So 2022 is when we realized that we all wanted to do the same thing. We took six more months to actually do to leave our jobs. So mid of 2023 is when we actually left and started exploring ideas together. We actually looked at a bunch of different spaces. Healthcare edtech were two of the ones that took the most amount of time. But eventually we decided to do something that was building on a technology that seemed to be creating a lot of noise and seemed like it would be the next big thing. Which is why we started thinking about AI speaking to people in our networks.

Simrat Singh:

And a few conversations that we had actually brought up the idea of automating voice operations. These founders and senior folks at startups that we were speaking to mentioned that running call centers for their companies was a big headache for a bunch of different reasons. There's high attrition in this space. The workforce is not the most highly skilled workforce. So outputs are often very variable. Training and quality assurance is a huge time and effort sink. So they were looking at ways to automate this as much as possible. And when we did a little digging around, we realized that the technology also seemed to be at a place where it could start making a difference to these kinds of workflows.

Simrat Singh:

With LLMs and text to speech models becoming more and more capable. So we decided we'll commit ourselves to this idea. We started building the technology alongside. We also were speaking to a bunch of different companies who had shown some interest. And that's how we got started building voice agents.

Demetrios:

What was the difference? There must have been some kind of a delta between when you were sloshing around looking at ed tech startups or looking at health tech startups and trying to figure out if you wanted to start in that space versus what you eventually ended up doing in the voice AI space and why was it that it called you so much and you felt that you had enough confidence to decide on this as where you wanted to dedicate the time and effort?

Simrat Singh:

Sure. So the thing that differentiated this particular idea building voice agents from what we were looking at earlier was how badly the consumers want this. So just to give an idea, when we were looking at health care, one of the ideas that we were considering was ERPS for hospitals. We went and spoke to a bunch of hospital administrators and we realized that that's not something that they are actually looking to solve immediately. Life still goes on for them with pen and paper records and it doesn't really come in the way of them delivering care. Whereas for some of the companies that we spoke to for voice agents, they were actually having a really tough time managing customers experience because they didn't have enough people. There were long wait times for their customers. Holding on call, resolving each issue was taking seven to 10 minutes and all of that was bringing down the NPS for their customers.

Simrat Singh:

So they were really looking to solve that problem for themselves. And that's what was the aha moment for us that yeah, it seems like we would be really solving a pain point for our customers if you build something that works.

Demetrios:

Makes sense. So now talk to me about building out the product and what you've done so far. The inspiration is clear. There was a pain. You saw that customers were very excited about it, I imagine when you started to float the idea by them. What exactly have you built so far and how have you not tried to boil the ocean? Because I'm sure there's so many different pieces to this that you could start building for folks certainly.

Simrat Singh:

So the core technology is transcription fed into an LLM which then translates the text to speech. That's the basic technology. But the magic really happens when you try to solve a use case for a customer and that's where most of our effort has gone into. On the conversational experience side, how natural does the conversation feel and how well can it solve a customer query, which is a combination of prompting the LLM well and making the right integrations with tools and other third party applications which the teams currently use to do their to complete their workflows? And coming to whether we have boiled the ocean or not, I think we are being very focused on what problems we want to solve and only when we feel like there is we can make a significant difference to the customers operations, do we actually go and start building something for them? So just to put some numbers to that, if somebody has say a 2 - 3 member calling team, while they might be excited to try a new technology, we know that it won't really help them unlock a lot of productivity gains or cost savings. So we don't go down that route. Only when somebody has a significant calling team which is doing more than say thousand 1500 calls a day, is when our product can actually help them. And that's when they'll also see the benefits and are likely to stick on and continue using the product.

Demetrios:

What exactly are you doing with the product? I know you mentioned like the core technology is the speech to text, then LLM and then text to speech. Is it replacing call center agents that you get and they're able, you're also able to take action inside of these different systems or is it just helping someone before they get to an agent? How do you look at that?

Simrat Singh:

Sure. So the way we look at it is that we would like to solve problems. 80% of the use cases end to end. And the long tail of the really complex, hard to solve 20% is what gets filtered down to the human agents. And to be able to solve that 80% we do need to take complete ownership of the issue. And that could mean entering data in a CRM. It could mean raising tickets for the issue that the customer is coming for. It could mean raising a return request on Shopify at the back end.

Simrat Singh:

So all of that is handled by our voice agents and only when it is something that is outside of the capabilities of the agent because the data does not exist in the backend. Or it is something that the agent is not authorized to handle is when it is handed off to a human agent. And from multiple conversations with folks who are looking to use this kind of technology, we realized that 80% is a very achievable number because that's the number that covers most of your standard issues which are still being handled by humans today and can easily be automated.

Demetrios:

Yeah. So one thing that we mentioned before we hit record and I want to continue the conversation on is how it's been for you primarily addressing the Indian market and going after the Indian market. I've been to India a few times and I know that there is a propensity for folks to speak in Hindi, English. And so I can imagine that can be very difficult when you're trying to create a voice agent because a what language is someone speaking in? Is it this mix if you're getting a little bit of Hindi and then there's a few words in English, does that trigger the English model? Does that trigger the Hindi model? How do you look at that? And then how are folks receptive to speaking with these AI voice agents?

Simrat Singh:

So I'd split our observations or experience into two categories. One is the macro dynamics of the market and how that responds to something like a voice agent. And second is how do people who are actually interacting with the voice agent respond to it. So the first bit is actually quite interesting because unlike the west where labor cost is really high in India, the arbitrage between using voice agents and having human agents do the same thing is not as high as it is in the West. So the business case that you build for customers cannot be just about cost savings, whereas this is most likely the case in the west where people are selling it as a cost saving too. So what it comes down to, and something that I mentioned previously, is that you are building in efficiency in the second order, which is you don't have to deal with the vagaries of having a human workforce. You manage attrition through having an automated system. You don't have to keep training it again and again when somebody leaves. So all of those things are what make life easy for either a floor manager or a customer support head of a company.

Simrat Singh:

So that's something that we've learned and understood. And it's actually not the best idea to approach it by saying that we'll help you save cost, it's that we'll save you tons of headache. So that's on the first bit. On the second bit, we were actually quite surprised that people are extremely comfortable talking to AI agents. We thought that it would have a very negative reaction from customers. They'll bang the call on it, they won't interact with it. But we've been pleasantly surprised. And I think part of the reason is because they are used to waiting on IVR lines put on hold.

Simrat Singh:

So being able to speak immediately and share their issues is a huge improvement over the existing systems. Even when there are humans who are handling their issues. As far as language is concerned, I think it is a challenge. I wouldn't say that it's as easy as handling only English and Spanish, for example, but we've managed to solve it to a fair degree for the more common languages. So English, Hindi, we have deployed agents in both of these languages and a mix of the two. So that's quite doable. And also when people interact with current systems, they're used to sharing their language preference at the beginning of the call. So if you can do that makes it easy from a technical standpoint also.

Simrat Singh:

And the experience of the user is also much better because then they can interact in the language of their choice and the model does not have to keep switching during the call.

Demetrios:

Yeah, I was also thinking, as you were saying, that India is so rich in different native languages too, so I can imagine that could be another headache that you have to support 32 different languages. But do you find that the majority of folks will either choose either Hindi or English?

Simrat Singh:

That's what's been our observation. And I think it's also got to do with the fact that technology typically trickles down from the metros and the tier one, from the larger cities to the smaller ones. So we are still operating in the, what people commonly call India one, which is most of the urbanized India, and their English and Hindi covers a significant chunk of users. So we haven't run into a wall, so to say, because of the lack of vernacular capabilities of current technologies. And I think there's a lot of work being done there. So we will soon be in a state where we can actually address those users as well. And I don't think that's too far into the future.

Demetrios:

Perfect, dude. Any final words, Simrat?

Simrat Singh:

Yeah, I think we can talk about how people often think that building something in AI requires a lot of technical knowledge, but often it's about knowing how you can use it to solve user problems. And I'll just use our journey over the last nine months to illustrate that, because when we started, we thought that a lot of our time would be spent in doing AI related work. Fine tuning models or building our own models, small ones, to address specific issues. But we've actually realized that there's a lot of value unlock that can happen by just using basic techniques like prompt engineering, the right kind of tool calls API integrations, because even the most basic tasks that can be automated are not being automated today. And there's a whole universe of such tasks that you can do without getting into really complicated AI and machine learning techniques. So the challenge today, especially in a market like India, which is a couple of steps behind in terms of adoption of AI, is to actually show the value of this technology to businesses, to end consumers, rather than building something that's really technically capable, which can do and order problem solving. So that's been an interesting journey and we are still figuring out how to do it because adoption is nowhere near where it should be, even in India.

Simrat Singh:

So I think that's the interesting bit when you're building an AI today.