Murf Unplugged: Winning in GenAI

Sneha:

Whenever we have had like team members who love their job, they create miracles. One of them created an ai film out of only ai product.

Ankur:

Only change they made to launch ChatGPT was the little bit re enforcement learning to make it the chat format rather than predict the next format, right?

Pranay:

Hi everyone, welcome to this Matrix Moments podcast with Murf.ai. Murf is a synthetic speech company which has grown I think 7x in the last 12 months which is incredible, and our customers have used it to generate over 12 million minutes of AI voice overs. We are privileged to be partners with the team for a while now so I'm excited to welcome Ankur and Sneha to the podcast.

Sneha:

Thanks, Pranay. Thanks for having us over and yeah, excited in doing this together.

Pranay:

So I think just for starters, Ankur, I’ll start with you and, Sneha, we’ll come to you after that. But, Ankur, you’ve been sort of in and around this domain for a while so I would love to start with the timing of it all. So maybe if you could spend a couple of minutes on your background, what you see happening and why you thought it was the perfect time to start the company?

Ankur:

Sure, yeah. I think in my experience research always sort of precedes commercialization and more than room for AI as a space and in my last stint at Goldman I was actually pretty close to the space so I was familiar with the research breakthroughs that were happening. I was really pumped up with the 2017 Transformers people that came out. And yeah, that was totally an inspiration to start something in the space for us and so I started looking at all the generative things that were happening especially in the research domain, nothing was coming out in the commercial space yet. And speech sort of struck me as a place where there was maximum ROI so to say to be derived from this sort of research work and it continued to be a very non maneuverable modality in my opinion because once you’ve recorded something getting it to a complete different shape or form whether it be accent transfer or anything you want to do with that speech is a really hard problem, something that just cannot be done without AI and it was really clear at that point to me that this is going to be disrupted in the next few years so we’re sort of like early to this generative AI boom at that point.

Pranay:

Yeah. And maybe for the rest of the founders who are interested in this domain since you mentioned modalities and let’s say that your team was working on things like document passing and other such projects so where are things hitting maturity, where is there still opportunity to go further and just if you could share timing of it all that would be brilliant.

Ankur:

Sure. I think everything started from 2012 paper from Goodfellow on GAN and 2017 on attention is all you need was the next sort of seminal work in this space. And then very recently the fusion work that has sort of made advancements on top of GAN so every five years there’s really a seminal paper coming out which is transforming the research world at least and then couple of years down the line everything changed in the commercial world as well and to your question on what modality would be impacted I think speech synthesis definitely is being impacted by some of these. Image and video again lot of people are very excited about it but it’s a space that's still evolving especially from a commercial front. We’re seeing a lot of promise from a lot of generative companies in speech and video domain but again the complexity increases when the possible outcomes are a lot. And especially in case of video it’s just almost infinite possible outcomes if you're trying to go from a text modality to a video modality the number of outcomes that you can imagine is a lot and that's really proportionate to how complex that is and that’s sort of inversely proportional to how mature that tech has become so far. So the next wave is going to be in some of these complex modalities, that's for sure.

Pranay:

Got it, thank you. And I think it’s actually the combination for me that makes it far more interesting so, Sneha, once again welcome to the show. I’d love to maybe spend a few minutes on your background as well as again when Ankur talks about speech what were your first reactions?

Sneha:

So I have always been curious about building new things, I have been in startups for over a decade now and was certainly looking to do something new. From my time when I was doing my marketing stint at Urban Ladder I was working very closely with creators, we used to build ads of all kinds. And I had seen sort of the challenges with voice over and audio maneuverability pretty much like firsthand. And I remember the first time that Ankur created something, and he showed me the output and I said this is really going to disrupt, we should do this because I just felt confident from having seen that space that yeah, we will be able to sell this, it was that moment for me. Yeah, I think that was sort of like the start but of course like we discovered several other use cases along the way but that's what helped me build real conviction to the idea.

Pranay:

So you obviously relate as the marketer and I’ve been a marketer myself so you create a voice over, it doesn’t add up with the video, maybe the pacing is wrong, maybe you want to delete a word here and there, lot of back and forth involved with the voice over artist to get to the final-final output and this really compresses that into minutes versus weeks. So I think that's brilliant, seen the pain myself so you found an idea, obviously perfect timing, you said let’s go and sell this. So what happened next, how did you go about finding your first customers?

Sneha:

I think it’s interesting even before like the beta product of Murf was out we wanted to get validation from customers who sort of actually need that product, actually need the output forget the product for a bit. So at that point it was really about establishing the quality of the voiceover that we’re able to create and whether that really like fits the bill in the minds of customers. And having been in the startup space I did run it by a few people I knew in the marketing domain in the creative domain but I think a little bit from a conviction standpoint like for the founding team we said that hey, we want to really sell to customers who have not heard about us, don’t know us, probably will be super harsh with us about the feedback but we want to do that. Yeah, so that was sort of like the thought starter behind selling very early on and seeing if there is somebody in the world who finds value and has no bias for us. So me and my cofounder Divyanshu we sort of started reaching out to people on voice over platforms, platforms where customers would come looking for voice actors so that was the bar that we kind of set for ourselves. And we would reach out to customers looking for this and say hey, we have this new AI based product that we’re building and this is the quality that we’re able to achieve, would you want to give this a try. And a lot of customers were like no, no, I'm not looking for AI voice over.

Pranay:

So I'm pausing on this one, right, so when you said marketplaces where people come what would these be, because I’ve usually seen voices and all where people are listing themselves.

Sneha:

Yeah, so Upwork was one such platform where we sort of created an account and started reaching out. We were a new vendor even on Upwork so that was never the GTM it was just an idea to test if the quality of the product really like stands for the customer. And, yeah, we also created an account on Fiverr and interestingly started getting orders to do like more inbound. And I remember that time that -- we did it like on a very small scale and eventually like sort of very quickly grew out of it but it was very, very important for us to establish that at a global scale, with geographies we’re not familiar with, with customer demographics that we’re not familiar with can we still make that call and say that hey, this is still like adding value, this is something that they want to buy. And lot of those customers interestingly ended up buying the subscription when we actually launched the product and that was sort of like the start, very early days. But I think another thing that I’ll add to what Ankur was sharing about the market being super nascent at that time for at least realistic synthetic speech. But we had seen certain like pockets of the market which were small at that time but they were growing like 10x over six months and while that absolute scale was not enough it would like give me confidence that this still continues to grow and with the advancements that we were making on the R&D side when this does become a mature market it will be very, very big. So I think that was sort of the thought behind investing on content very, very early on and being super consistent. Content is like one of those games where you see no ROI for like a really, really long time, we have to keep at it. I remember like the Murf blog and the website and there was pretty much just me, Ankur and DP for like a little short of a year and I used to write everything like myself and that's how we started with little things and scaled that eventually.

Pranay:

So phenomenal tech obviously, again instruction for all founders, do things that don’t scale, find customers who give you brutal feedback, hopefully not from the networks that say good things. But when you had to think about again building a motion and a product that could scale how did you make those choices around the go to market because you’ve created a voice now there's a lot of different places where this can be applied. So what were some of the choices over there?

Sneha:

Yeah, that's a very interesting question. So for us there were two things that led us to go PLG overall, the first thing was that we wanted to be a global company right from the get go. We knew that the market for synthetic speech was – well, for whatever reason, it was the largest in US Canada, it was where all the incumbents already had presence. So we wanted to compete at that scale, we didn’t want to fall short of that, so that was one thought. The second thing was that we very quickly realized that the way the initial like shape and form that our product did take it wasn't very hard to understand value in the product if you have the use case. If you don’t have the use case you probably would like play around and say hey, maybe someday. But if you are looking for a voice over and it works then it just works. So I think these were sort of like the two considerations that led us to go fairly with a like very simple self-serve motion and just scale. We realized that once we were able to get ourselves, our product, in front of the relevant audiences they would try out the product and lot of them would buy. And we said if this is working could we just work on getting ourselves in front of a larger audience which is has a similar use case. I think that criteria still didn’t apply and that's where I think bunch of our overall content strategy in ten based marketing all in that was focused because I think in the early at least year, year and a half, we did not at all venture to sort of like a brand marketing or that kind of zone. We really said that if customers have a use case can we serve them well, are they finding value. And we sort of like really scaled I think that overall effort did pay off a lot last like 6-8 months when this became really big. Yeah, so in hindsight I think it was a good choice of strategy that we did.

Pranay:

So your observations were, a, fast time to value, and b, probably that people are already looking for solutions and you don’t have to evangelize and create demand and both of them added together. Maybe one curious byline but given that ChatGPT was probably the marketing moment for all things Gen AI what have you observed in your domain?

Sneha:

We have nothing – in our business honestly like our business does not relate to ChatGPT directly in any way that I can imagine but I think just the general curiosity around AI and what it can do really exploded the awareness that creators had or started having about what’s possible and even if they were not aware they wanted to get aware. And what that led to that even for audio space there are like segments where we’ve seen like 10x growth over just a span of like 4-5 months. That was I think even in my experience it was quite a moment to experience as a marketer as well and it’s certainly very, very interesting times to be in the space.

Pranay:

I know I’ve spoken to you about this before but building a US first B2B GTM is hard, right, because the toolkit is finite, you can buy a billboard, do an event, run a podcast, write content online, run some ads, hire some outbound sales reps. So I think finite set of options and then it becomes for me at least a game of execution that what can you do really well and where can you stand out. And if we’ve not understood this already you’re really crafty in how you get to customers so when you thought about that okay, there is demand online I need to get in front of this audience what were the 2-3 things that you did that stood out in terms of driving demand for the product organically and maybe again the devil is in all the details so the more you can share the better.

Sneha:

Sure, sure. I think the first thing that I’ll say about generating any kind of organic traffic irrespective of what channel you’re after is about consistency and quality of content. What you say does matter, if you say things that are just irrelevant or is just like really low quality so I remember that until like a year ago I used to review every single piece of content and after that I just started realizing that hey, I'm being probably a control freak about this and stopped doing it, I don’t do it anymore. But for a really long time I was very worried about the kind of content that we’re putting out and yeah, at least I kept sort of our conscience clean on that that we will not put out something that we don’t like and lot of it was written inhouse really with hard work, quality. The second thing about it is actually consistency, and I think like I was mentioning earlier for a lot of channels you just have to keep doing it and for a really long time you see like very, very incremental movement but then when that step shift happens it’s like bigger than any channel that can create for you. So we stayed consistent and I think somehow our early team members did carry that philosophy forward really, really well. They sort of stayed at it, kept creating content on relevant topics. We did not spread ourselves thin and the reason for that was that we knew that there was deep intent in certain categories versus some other businesses where you would take a very wide approach.

Pranay:

I’ll actually double click on that, right, so text to speech very broad category, is that what you went after?

Sneha:

That is one of the categories that we did go after but there were lot of like new market creation sort of possibilities coming up as well but then again we stayed within the domain of voice, we did not just want to generally – we could have created content of all kinds of generative media. Some of it we’re venturing into now because of our product offering increasing but I always feel like any decision that I take for GTM is always customer backed and I feel that if I'm putting in that effort for the team to create content and then for that customer to discover us and land on the site and we don’t have something to offer is just a waste for everybody. So we don’t create like content about like everything in the world but on the topics that we do create we go very, very deep. We sort of explore all use cases and really stay consistent about it. I think it’s just consistency and hard work I would say, quality, that was like one whole piece in that sense. I think apart from that one large piece that we continue to work on is improving the product experience because it’s only half the job done to get people in. If our motion is completely self-serve for a very large part of our audiences and if they’re not able to discover what we create and just are not successful at creating whatever they’re here to create they’re not going to buy us. So I think that just the pace of like product feature launches I think Ankur has this philosophy of just ship fast, this express shipping psychology in the org that I think his entire team has. I think you can touch upon more on that but we just kept hearing from customers and building and building and building and building. And a lot of the features that we built in the early years honestly I think we had not visualized that the product will have these, they only came up when we spoke to customers and they said hey, you're doing great at this but can't you just simplify my work flow by doing this or can't you add this feature, I want to maneuver the voice in a certain way, I want my brand to sound a certain way, things like that. Use cases like the voice changer I think it was completely visualized out of customer use case that I have already created so many videos and I just need to change my voice like why are you asking me to create a script again. So we said, hey, just bring in your video or your audio, we’re going to do the rest.

Ankur:

Just shows us to build something that people really want philosophy, everything we built from our initial launch days to now so far has been all customer demands, just tracking them, because we have a very high volume inbound motion, we have enough statistics to know who’s going to be willing to pay for what and what kind of audience needs to what them. As far as that aligns with our general direction in terms of who we’re trying to cater to it’s really easy to build around that.

Pranay:

And I know you mentioned that you stopped looking at content, team is taking the mandate forward. I will come to the team in a minute but I want to talk about the listicles because when I was doing the research you were on top or at sort of close to the top of every listicle that I went to and Googled sort of prioritizes them now. So how did you make that happen?

Sneha:

Yeah, I think a lot of that credit does not go to me, it goes to my co-founder Divyanshu again. I think very early on what we started doing is that he would like reach out to influencers in the community who were sort of like thought leaders in innovative tech products and just asked them to check out our product and share a review if possible. When we were very, very small nobody knew about us, a lot of them wouldn’t put in the effort to check us out but some did thankfully and we kept consistent about that as well just like with anything organic and over time one influencer posted and another one heard from them and that sort of like started snowballing after a point. But there was this interesting story that I remember about DP reaching out to one of – I wouldn’t name them but they’re a very, very large tech blog and they only review products that they think are innovative. They wouldn’t even open our product to just consider us for a very long time but then he kept at it for two long years until the guy decided to sort of --

Pranay:

So an anonymous e-mail ID so they think – what was that e-mail that he sent?

Sneha:

He just -- I think we just said that the blog is outdated which it was like ---

Pranay:

So were you like a customer saying this blog is outdated I saw this cool product you don’t have?

Sneha:

I wouldn’t say we were a customer because even for a blogger it’s a reader and they care about their readers. And, yeah, I think this was one of them where like the blog was not updated actually for a very long time. Somehow the person had not considered to see what new is coming up in this space and thankfully he did after two years and we did manage to sort of rang bells with.

Pranay:

Awesome. And I know we’ve spoken a little bit about customers already but you're giving them a voice and they can do anything with it. So I'm curious to know like what are some interesting stories or whacky ideas that customers came up with and also sort of surprised you. I know it’s an important factor because they drive a lot of the route map so curious to hear what you saw?

Sneha:

I mean we’ve seen a diverse set of use cases. You know, business customers like there’s someone whose been using us for over two years, they have their own entire LMS system and yeah, they found us early on they were using one of the clouds before us and we’re way more expensive than them but they found the realism in the voices really valuable for their customers in turn and they have been using us for a very long time for creating all sorts of content. Well, on the whacky side like we’ve had very interesting customers, one of them created an AI film out of only AI products. It’s a pretty famous person, I wouldn’t name them but they used our voices for it, they used another tool for generating the script, they used another tool for generating the video. We’ve seen podcasters create entire podcasts out of murf voices.

Pranay:

So I don’t need to..

Sneha:

No, no. I think it’s lovely to see I think some of the changes that are created. I had actually read this somewhere a few years ago that with the creator economy as soon as the power is handed in the hands of the creator to do something new they do things that surprise the builders all the time. And we’ve seen that stuff happening all around with ads, with just like I said podcasters.

Ankur:

People creating poems, we’ve had people do Jazz, it’s not meant for it.

Sneha:

Yeah, even raps somebody sent us like a whole rap that they created, we didn’t have like all these use cases in mind when we created it but yeah, it’s just the power of that economy, I guess.

Pranay:

Awesome. So I wanted to talk about – you mentioned shipping philosophy, you mentioned team and I think one thing that struck me when I walked into your office was that it was just quiet like everyone was heads down working. Our office is much more noisier than that so what’s the philosophy around and I think also meant to be as a compliment but I think the number of people you have and the velocity at which you ship and the scale you’ve built is actually phenomenal. So what’s the philosophy around shipping product as well as around hiring people.

Sneha:

The shipping philosophy maybe you can cover better, it mostly comes from Ankur.

Ankur:

Yeah, I mean it’s a known thing, we want in general what I tell all my devs is we don’t know if it’s going to succeed, right, so what would you rather do. Spend a quarter on it and then say it failed or build in a week and then let’s see how it behaves and then iterate further to see whether we’re going to pursue it or we’re going to dump it or make it really awesome. So it’s really like fail fast in my opinion which makes or breaks startups and that's really what defines a startup, might as well otherwise behave like corporates. So we iterated a lot and because also some of the times we may be wrong with our prioritization, we may be wrong about what we pick directionally or maybe sometimes even in understanding what the customer wanted. Lot of times customers are also not very -- either they’re not very crisp about what they really want, right, there might be a mismatch in our understanding and then their expectations and all of this end of the day can only be sorted out if you give something in their hands, get their perspective and then build on it. All those sprints it’s a mandate right now for us to like keep it shorter than two weeks, we don’t want go beyond that. Exceptionally we might sometimes if it’s really complicated but definitely from research of course where the projects are really long but as you read we build small things, deliver every week, two week’s timescale has helped us a lot in picking the right directions and building on things that are working for us. That's really our mantra to figure out what should we be doing because as I said in the inbound motions helps this kind of a technique. If you were building for a few corporates then you have someone to talk to but this is really a statistical things that we’re doing so you get to experiment a lot more.

Pranay:

I think that’s interesting because at least for us we’ve seen companies at the other end of the spectrum who are building for 12 months, 15 months before they even come to market but it’s probably more nuanced even for the category because in a domain where people also don’t know what to expect, what are the possibilities what they really want you really need to show them something and get reactions versus build in stealth for longer.

Ankur:

We were selling two months from the day we started building this. So we didn’t want to give out things for free for a long time just because it doesn’t prove anything. So we kept the price point of view but we still wanted to charge to see if they really wanted, they really needed or it’s just not being serious about using it. So that was the philosophy throughout.

Pranay:

Let’s talk about the hiring philosophy.

Sneha:

Hiring, well, I think we sort of keep the bar high, we almost never over hire. We look for passion for the job when we recruit for any role it doesn’t actually just apply to research or dev it applies to every single member of the team. If they don’t love what they do at least we believe that you're probably better off doing something else because it’s just a drag to sort of – and we’ve seen this over and over again whenever we have had like team members who love their job they create miracles every single time, that one person is always, always valuable to the team and we try and look at that. In general I think as a team we have almost no hierarchy, everybody has like ownership of something that they’re doing and we trust them. After we recruit them we don’t doubt them on their like skills or ownership, we believe that we hire very good people. We have been fortunate to have some really great team members in the team carrying the org. But, yeah, it’s just that I think like not doing unnecessary alignment meetings wherever it’s not required and everybody knows what they need to know in general as a culture. It’s not expected that somebody will come and tell me something and that is when I get to know about developments happening. So, yeah, I think generally we keep it like super flat, few people owning big pieces, having fun while they whatever they’re doing I think that's the large part of, yeah.

I think something about the office that you mentioned also goes back to like the kind of work a large part of the team is doing whether it is the research team or even the business teams in our domain like a lot of them are working on hard problems, we cannot deny that. And it needs that focused deep work to happen for any sort of monumental outcome to come out. And I guess like we try our best to create that sort of environment that's conducive to that kind of deep work. There is like no clear mandate, you can scream in the office as well, it’s alright. It just happens to be that they’re probably on to something and that's what probably keeps the office quiet.

Pranay:

Awesome. So I think, Ankur, coming back to you I think and going back to technology, so I'm curious to know in this domain of synthetic speech what other tech breakthroughs that you expect in the coming years or the things that you're excited about?

Ankur:

Yeah, I think synthetic speech has been fairly advanced the last few years, have already been very exciting. People are largely convinced now that communication as a sub aspect of it is fairly solved. But art is something that still needs some sort of breakthroughs for this to get used. It’s almost like how a voice coach would coach a voice actor, that kind of maneuverability is what I expect to be there in our offerings in the near future. In fact the idea is you being able to completely direct the outcomes of AI, so far all our tooling has been browser you're trying to deal with some sort of UI or some sort of an interface but I think the next generation of tools that we’ll be building on synthesis are going to be where you're interacting much more naturally. You would be very explicit in terms of how something should be done. You’ll probably not depend on a UI to direct that because probably that’s not a very natural sort of thing to do because there are lot of nuances. So you may want to have very specific requirements on your pronunciation changes, you may have specific nuances on accents that you want, within a sentence you might want to change the emotion. So there is just so much you can do when you get into the art side of things and I'm very confident that in the near future all of that is actually going to be catered to. Other fields I feel which is going to be disrupted is the low resource language spaces. So far English obviously there’s abundance of so resources around building stuff on top of those languages but a key aspect of low resource work is the ability to transfer that knowledge into that language. Or similar things could be said for accents as well, you might load in a certain accent certain language but be able to transfer that knowledge that emotion that accent that maybe pace. So any of those aspects being able to decouple and then put it back into a different packaging because ultimately you will not have access to every single resource in every single language in every accent, every possible variability out there for your model to learn from. So that ability to do this kind of transfers, yeah, we’re already seeing good sort of breakthroughs in this space and it should be close to commercialization fairly soon so we’ll be able to deal with any language eventually in the sort of similar agility that we can do today with English and beyond as well obviously.

Pranay:

That's interesting. So I think on a spectrum we’re going from communication to acting and sort of more emotions, more tuning, more sort of creativity.

Ankur:

More natural ways to tune it, yes.

Pranay:

And I know favorite VC question and maybe it’s not a very good question but all this brings me to data, right, so what role does data play in this, is it a head start in the short term, is it a MOT in the longer term and how do you think about it and the category?

Ankur:

That actually varies depending on what kind of AI you're trying to – within speech as well the answer in my opinion varies. If you're trying to ace our speech to text that is or synthesis the answer varies because if you were the learning domain of AI which is speech to text typically you have one right answer that's true for anything you're doing. An image as well if you're doing image segmentation or object detection any of that so there’s always one right answer, is there a cat, yes, there is a cat. But that’s not true for synthetic media, that's not true for generative AI and that's where the same text can be read in infinite number of ways and what that means is data is going to define how the outcomes are and that means there is huge possibilities in terms of what should we collect and the conversion is no where near. That's one aspect, the other is the expectation from synthesis is that you're going to produce pristine outputs, we’re looking for something that is going to compete with what a voice actor can do today. So from that perspective again synthesis is a space where clean data is important so again becomes a MOT to some degree. And the third aspect I feel is because we’re synthesizing something it sounds like someone, there is that aspect of likeness and we as a company have sort of very strict policies around ethics, around voice cloning and how we sort of prevent that abuse. And that makes the data that we collect and we build in fact ourselves a big part of our strategy because it’s not like you could rather just go out there and pick any voice and just start using that, so that's not an option for us. And from that perspective again I think in this space it remains and will remain to be a MOT at least for the near future. Very long term still hard to predict whether a lot of companies would have enough interest to build something that broad. Yeah, definitely a MOT in the next few years at least.

Pranay:

And I know you mentioned sort of the ethical and legal aspects to it, right, and you also mentioned voice cloning. So I'm sure customers ask you for this all the time so what was your stance on it?

Ankur:

Yeah, we actually don’t do any of self-service voice cloning today. We are vetting the person who’s trying to use it for what purpose, we have voice fingerprint to identify whether it’s the same person who’s given that recording. We have all the measures in place to know that it’s being used in an ethical way. If we have enough consent from that person and as you said demand for it is there in terms of – and the laws aren’t great yet but we’re already sort of ahead of that curve in terms of how to sort of be completely on the right side of this and..

Sneha:

No, I think we sort of have very, very clear transparent contracts with the actors we work with we protect their data with all the might that we can. And we don’t let anybody clone something that doesn’t belong to them or they don’t have consent or like Ankur said we’ve been very clear on that from the start.

Pranay:

I actually had an interesting experience on this one, I was trying one of these AI music generators and I just created 4 tracks, played them out, open Shazam on my phone and Shazam could sort of link them back to at least two out of four it could detect what the music was. Right, so you play music like that sampling it or map to another song so I think the legal implications are very real.

Ankur:

Quite higher

Pranay:

Yeah. So data was obviously one hot topic, the other one I think which everyone loves is defensibility, whether you're a founder or an investor it’s top of mind for everyone because the pace of change is honestly anxiety inducing. I tell people that I switch off Twitter just to save my mental bandwidth from this and to sleep well at night. But in fact a couple of founders even told me that might be a late mover advantage that let’s just sit out for two years, see where all the dust settles, where is the technology at and we’ll start building after that. So when you think about this how do you build comfort like at least in the short to midterm that you’ll be alright and you don’t get disrupted overnight?

Ankur:

Broadly speaking that's true for any field, if you don’t want to be disrupted you need to be that person who disrupts the market, you need to be the thought leader and just be ahead of others and that's true for our space as well. We want to be the people who bring about that disruption so the pace of it continues to be important. To the point of whether there is the advantage in being late to the game from a research perspective I agree to that, you don’t have to be the first one to probably go ahead and commercialize something that was written yesterday by someone but from a GTM perspective once you're trying to commercialize it I still feel being early has an advantage. There’s lot of discovery in the journey both for the company and for the customers as to what they can use this for and what you can build for them. That’s going to take a lot of time and there’s going to have lot of nuances to it that you only discover if you're early. And we’ve been doing on that journey for two years now. We know what will work for whom, what is it that people really care about. There are certain things about when we change something we now exactly know what the customer is going to say about the change and we could actually just see a lot of their end of things where we’re very intuitive, we’re able to say that this is going to play in the customers. So over a long term the defense is what it comes from understanding the customer needs, what and how they’re going to use the product, that's a long term definitely true. But in the short term because the space is evolving so fast innovation in this space and being on top of research is going to be really key. Next couple of years is going to be key, again ultimately once dust settles and it’ll take longer to settle in this case especially in case of generative because as I said there’s lot of outcomes possible compared to the learning sort of space of AI. This is going to take much longer to settle down but once it does the MOT is going to come actually out of knowing what someone is using you for and building for their workflow. So that SaaS space is going to become important eventually.

Pranay:

Just thought of in the domain of engineering and research so we get a lot of slack saying okay ChatGPT happened and everyone woke up to this idea but in the domain of research everyone could see it coming 3-4 years out.

Ankur:

Yeah. I mean the pace like just between GPT3 and the initial version and 3.5 you could just see that pace of improvements and people like to move the goal post all the time of what AI can do and cannot do and then literally the only change they made to launch ChatGPT was the little bit re enforcement learning to make it the chat format rather than predict the next format. That's really just a change in how it’s interacting but from an understanding of the world perspective there’s not a lot of change. So 3.5 was around for one year before I think ChatGPT was launched and transfer was 2017 papers so it’s like five years in the making and the only thing truly that they claim being transformational about it is the size of the model. You just pick a transformer, put some reinforcement on it and then sort of grow with the size and the MOT becomes maybe the training cost. But of course researchers yeah, we’re seeing this happen for a long time, at least the technique isn’t new or nothing that was created the last two years.

Pranay:

And I know you touched upon some of this already but maybe if I go back to the world of say natural language understanding versus anything generative I always use the example of Gong. For me it was a case of Gong started but today that technology is available as an EPI, it has Deepgram and Assembly AI and a bunch of others but they’ve built obviously an enterprise work flow targeting sales leaders so they’re doing all intelligence and forecasting and bunch of other things and they’ve built a massive brand. So in this sector like what do you see, what patterns will repeat, what will be different and is it just this traditional SaaS MOT where it’s depth of workflow and some right to distribution or is it something different?

Ankur:

Long term as I said I think I agree, yes, it’ll become important. But that long term whether that is going to be much longer in generative space.

Pranay:

So that's the difference.

Ankur:

That's the main difference between – so like getting a speech to text good for your specific domain you could do that like ten years back and it was achieved fairly fast. But that was because there’s this one true answer. This is the space where you want that kind of like brand managers will say this is the kind of way that it will fit my product, this is what will sell my product. They’re that specific about when they pick out a voice android they would want that kind of choices with their AI as well so there is a vast sort of area of things that will play out and then people ultimately settle on the best, they get together around it, there’ll be very popular things and less popular things in the area of it. Just the diversity of outcomes is going to delay that process of settling down and then just your GTM being your strategy of MOT that pace will come, I can anticipate that, but --

Pranay:

-- will take longer. Got it. I think we spent enough time so with that we’ll call it a wrap. Once again congratulations on building a fantastic, fantastic business and thanks so much for making time.

Sneha:

Thank you so much for having us and sharing our story, Pranay. Lovely talking to you.

Ankur:

Pleasure is ours.

Sneha:

Pleasure is ours, absolutely.

Pranay:

Thank you.

Sneha:

Thank you.

‍