Tanay Kothari Wants To Kill The Keyboard

At 11, Tanay Kothari built a voice assistant that went viral. At 14, he co-founded his first company. At Stanford, he won hackathons, sold a startup, and became obsessed with one question: Why does every voice tool still sound like a broken answering machine? The answer became Whisper Flow. An AI voice dictation tool that works across all your apps. You speak, it types. But unlike basic transcription, it learns your tone, cleans your speech, and adapts your style by context. Casual with friends, formal at work. Where Apple and Google hit a 10% zero-edit rate on voice messages, WhisperFlow hits 85%. Naturally, it's become extremely popular, with 90% of its growth coming purely from word of mouth. In this episode, my colleague and tech startups tracker Tanishka Dubey asks Tanay, what makes Whisperflow different from every other voice tool? Why is India a priority and are Indians actually paying? And should founders still build foundational AI models? Listen on. It's Tuesday, the 31st of March. I'm Anirban Chaudhuri. This is The Morning Brief.

Tanay, let's start from the beginning. Why Whisperflow? Why go all in, in voice? And how has the journey been so far?

Voice is a fundamental good in the world. Why? Because one, humans are creative beings, right? We love to express ourselves. If you are typing down your thoughts, a lot of what you say gets just lost. The second thing that happens is it actually feels like grunt work. And so if you could give voice to people, you one, remove a lot of grunt work in their lives, let them be more expressive. But more than anything else, In the future, I want to see a lot more voice-first technologies. Phones are display-first, laptop is display-first. If they were voice-first, then I'd grow up in a world where my kids don't do this all day long and stare down on their phones. And the only way to do that is you have a voice assistant that people trust. And so we started on that journey. We were like, ki sabse important problem kya hai, ki hardware ko hi nahi banara. So then we decided to build that brain-computer interface, which is this device non-invasive. You can wear it and it understands what you're silently thinking about and converts it into text. So I could, for example, sitting here, send a message to my friend. You wouldn't hear a word of what I said. So that was the core thing we built, and we built it for 3 years. And then we realized, ke, 3 saal samya banar hai, nobody else improved voice. And so then we decided to build this little thing called Flow. And when we decided it wasn't supposed to be a new product just yet, it was supposed to be the operating system for our hardware because we were like, "Humain chahiye, tum kuch keliye banalata hai." But then what we realized is people are very resistant to change. If you ask people to change or build a new habit, you can only do that in bite-sized steps. So this is where Humain failed because they were like, number 1, you change your SIM card. Number 2, you change your hardware device. Number 3, you speak to this hardware device. Number 4, you see the display on this hologram that you can't really see on your hand. Our insight was ki we can only create behavior change one step at a time, and we need to just nail every single step. So step 1, ki aiki, you know, use all the apps and everything that you do just as you have. All I'm asking ki keyboard ki jega, you just start using your voice, and I'm going to give you an insanely good voice tool, you galte ni karega. So you'll get your results the same as your keyboard. Once they do that, then I can be like, okay, so you're just using it to dictate things, and log apne bolte ki yaar, I want to tell it to do things. Because they trust it. Most people in this using Whisper don't even read what Whisper writes. They just press send immediately.

So no commas, no punctuation errors, nothing like that.

All of that Whisper takes care of— the punctuation, how you sound. Sometimes when you ramble for a bit, even like fix it for you so you're not wordy and all of that. So it does all of those like editing on top of it to make you sound like the smartest version of yourself. While still sounding like you. So then we can finally have tech where you do not need to see a screen.

Okay. Tanay, how do you see competition in India? I mean, there are a lot of voice AI startups in India. So what is the differentiator here for you guys?

The biggest differentiator is voice is actually so many different things. If you are building a specific agent that does phone calls, completely different thing versus if you are building something that reads things out to you, completely different than if you're building something that transcribes a video or transcribes this recording and makes a summary of it. For example, Granola or like other meeting recorder tools versus a tool where you speak and it writes for you across all your applications, completely different problem space. And so I would say Cartesia and ElevenLabs, I don't even think of them as competitors. And so the key difference is you might open a product, there might be a voice button there, there might be— you call somebody, you'll talk to them on the phone, there's a different voice system that you use. With Whisper, it's going to be the same voice system that works across all your applications, which means it learns so much about you.

What you're talking about is the interface, but what is the voice-voice differentiator here?

Number one, when you are, for example, transcribing a video, You want everything word for word. When you are doing a meeting recording, if a couple of words are missing or wrong, it doesn't make sense because you eventually summarize it and it sometimes figures things out. But for human writing, that is an insanely high bar of precision, and what you write is different than what you speak. So I'll be like, oh, you know what, you want to meet at like 5 o'clock, actually 5:30, kal sham ko. Whisper will write that, let's meet at 5:30, kal sham ko. So it'll fix your mistakes. But here's the thing, maybe Tanay as a person likes to sound like very polished and smart in my emails, but in a text message with a friend, I'm a little more casual. And it'll figure that out. So it'll personalize by application. So in my text messages, it's all lowercase, it's chill, emojis, lesser punctuation. Emails, formal and all of that. And that is where a lot of the magic comes off is because people don't expect it to do that way. People don't expect this to understand ki oh shit, like how did you know I use this word only with my friends and you actually got it and you got the name of this like friend who has a very hard to spell name that nobody gets right. All those little things start to matter a lot when you build for this specific behavior, which as companies like ElevenLabs and Cartesia aren't focused on because it doesn't matter to the problems they're solving. Cool.

So right now you're doing just voice, and that too quite generic. Would you like to expand into different areas in future?

We'll do a lot of different things, and so I stack rank the business opportunities that are available to us, both in terms of how much does that get us closer to our eventual goal. That can be in terms of like, that pushes the product there, it makes a lot of money, that helps us push the product there in different ways, it helps grow the brand And again, pushes us one step closer to getting things there. Building an API-like business, a little lower in the ranks based on our— the way we evaluate, you know, what we should work on. So if I had infinite resources in the world and I could build any and all the businesses I wanted, of course I'd build this business. But I don't.

Got it. So Tanay, tell me a little bit about yourself. Your dad is an investor.

Yes.

Did that help?

So the interesting thing is, so I started coding when I was 9 years old. At that time, my dad did something in finance. I'm glad he became a VC because now I finally understand what he does and we both can talk. But when I was 14, I was at a hackathon at this incubator called Investopad in Gurgaon, and I met Arjun Malhotra, who was one of the founders, and he and I hit it off. And he told me, why don't you come and like work with me? Over the summer. And so a couple of weeks of working together in, he and I decided to start a company together as co-founders. And so that entire summer, my dad would drop me to the Investopads office on his way to work. And so roughly around the time when I was starting my first company was when he started his first venture fund. So we both kind of entered the space together. This is when I was, yeah, 14, 10th grade, he always has extremely high aspirations for me. My dad is like, my son will start a unicorn. And I was like, it's hard to please, but always pushing me harder. Or when I was raising and I was like, oh, I got a term sheet from this fund, dad's like, honestly, you can get a tier 1 investor. I was like, you know what, okay, I'll go get one. And then we got one of the best investors in the world, NEA, uh, one of the oldest VC firms, to lead that round. And then later on he was like, then I think the markets aren't doing too well, I think you should raise right now. I checked and I saw that was true, so then I went and raised, and then the markets collapsed a couple of weeks later. In a lot of ways, I don't think, like, regardless of the scale of the company, his advice has ever stopped being helpful.

Got it. So Tanay, I want to get your take on something. There's a lot of noise around foundational models right now, whether to build one or not. Which side of the debate are you on?

So in a model's journey, there is, uh, step 1, which is pre-training. Pre-training is the most expensive step of them all. Pre-training like a decent-sized model costs you anywhere from like $20 million and 3 months of time. Then after that comes aligning the model with a goal. The goal can be— for code, it's like, I want you to produce code. And then after that, you do fine-tuning. On top of it to get the final things around style, to get the final things around, you know, I want to sound like this, make this more short, and all of that. Like, those all things you can tune in the fine-tuning. Fine-tuning is very cheap. You can fine-tune a model for every single user in your user base if you wanted to. What people are saying when they're saying don't train foundation models, what they really mean is don't do the pre-training step, because there are a number of open-source models available today that have been pre-trained. So Do I agree with you shouldn't pre-train your own models? In most cases, yes. You need to have an extremely, extremely good reason and you need to prove to yourself and a bunch of people around you on why you cannot solve that problem using any other way.

I want to come back to the India question. When I speak to other startups entering India, many say that they are here largely because of the volume.

Most people come here for volume.

Is that the same for you, or do you also see some revenue potential here?

I personally have two main motivations in building for India. I see how transformative Whisper as a technology is, and I can just see people in India craving for something like that. And as a culture, we just default to voice. In US, nobody uses the voice button in Google Maps. In India, half the people I see use that. And so it's inherently more voice-first. And you're giving this country that is voice-first tools that just don't align with them. And so a big part of me was like, ki yaar, I want to get this back home. I know this is going to be incredible. Say the revenue argument wasn't there and Indians actually did not pay us money. I would still choose to do that, but at a later point in time, once we tap more of the revenue-first countries. The second part is the thing I learned is not that Indians don't pay. What we learned is Indians pay if you deliver them good value, but they're very, very picky or like they're very stringent about you actually delivering that value. And that is what we optimized for. And so the very interesting thing I'll tell you about India is the conversion rate is very similar to the US in terms of the percentage of people paying. But more importantly, in US, between the annual and monthly subscription, it's 50-50. India, 80% of the people pay annually.

Got it.

And so once you build their trust, people like it.

Okay, so who are you targeting in India? Like any big names?

All the big names. So in terms of enterprise, I want to get into every single company in India to have voice be the default and Whisper be the default way that they work. So then the question is, how do we do it? Well, we go after the companies that are going to be the earliest adopters and are going to be good aspirational names. So that's the same thing that we did in the US. Now we close basically a new Fortune 500 every couple of weeks, and this is just going to keep accelerating. And this is going to be one of the only tools apart from Anthropic and OpenAI that the FAANGs have adopted internally. And again, we compared to them are an extremely early company, right? It's been 18 months since we existed. With this product and 7 months since we started enterprise sales. Companies like Slack and Notion built their consumer motion for about 5 to 7 years, then they started their enterprise motion, and then a few years after that they were able to close their first set of Fortune 500 companies and all of that.

Hmm, got it. Okay, so can you talk about revenue numbers? Uh, what has been your run rate so far?

So we don't go public with any absolute numbers, so we've been scaling our ARR at about 40% month over month consistently for the last 18 months now. And so in very absolute numbers, it is that we double our revenue every 2 months.

Got it. And finally, Tanay, if I want to have a last word on product acceptance, what does the number on customer churn and retention look like?

So that's the other great thing is our consumer retention. Month 13 is 70% and has been improving. For enterprises, our net dollar retention annually is 350%, which means a company that's paying us $100 now is going to pay us $350 a year from now. And this is what I was saying with one of those FAANG companies starting at 500 seats and expanding to 25,000 seats in 3 months. And so that basically 50x'd the revenue that they were generating, that we were generating with them. And so those were the key things that are extremely important to me as we're building this business. Strong unit economics. It's about 90% gross margin. And so that means that we can just, you know, scale a lot. And the business by default is very, very profitable the way we've built it.

Whisperflow has seen 30x revenue growth in a year. That number is hard to ignore. But Tanay Kothari knows the real test is whether people actually stop reaching for their keyboards, not just in San Francisco, but in Delhi, Mumbai, and everywhere in between. That proof of habit is still being written. That's it for today. You were listening to this episode on The Morning Brief. It was produced by me and Rajas Naik, sound designed by Rajas Naik. A new episode of The Morning Brief drops every Tuesday, Thursday, and Friday wherever you get your podcasts. Keep listening.