How a weekend hack became a multimillion-dollar AI startup | Adit Abraham (Co-founder & CEO at Reducto)
Episode 146

How a weekend hack became a multimillion-dollar AI startup | Adit Abraham (Co-founder & CEO at Reducto)

Adit Abraham is the co-founder and CEO of Reducto, which helps leading AI teams extract and structure data from complex documents and spreadsheets in their pipeline. Within 6 months of launching, Reducto went from 0→7 figures in ARR. Reducto has grown to process tens of millions of pages monthly

Play Episode

Adit Abraham is the co-founder and CEO of Reducto, which helps leading AI teams extract and structure data from complex documents and spreadsheets in their pipeline. Within 6 months of launching, Reducto went from 0→7 figures in ARR. Reducto has grown to process tens of millions of pages monthly for companies ranging from startups to Fortune 10 enterprises. They just announced a $24M Series A. Before Reducto, Adit was a Product Manager at Google, working on Ads and Search, and conducted machine learning research at MIT's Media Lab.

---

In today’s episode, we discuss:

---

Referenced:

---

Where to find Adit:

---

Where to find Brett:

---

Where to find First Round Capital:

---

Timestamps:

(00:00) Hackathons, YC, and an unexpected pivot

(05:23) The weekend project that became Reducto's breakthrough

(09:11) How customer signal led to PDF processing

(14:46) Landing a Fortune 10 customer

(22:42) Building “transferable features”

(25:58) How caring beats sales skills in startup growth

(30:28) The strategy behind Reducto's horizontal expansion

(36:18) Hire slow, go-to-market fast

(41:45) A technical founder's guide to sales

(43:45) “You’re going to fail”

(46:27) Why startups win

(48:30) Key insights from Reducto's fundraising journey

(51:43) Less structure, more impact

(55:00) How frustrations shaped Reducto’s culture

(57:35) The question you should always ask in meetings

Brett: So what was the first glimmer of what then would become Reducto. Not when you decided you were going to go build the company, but like if you trace all the way back to where it first began, what is that? 

Adit: We didn't know this at the time. but when. Raunak and I decided to, you know, work on something, and to apply to YC, we actually applied with something fundamentally different. So we were building long term memory for language models. that was a little too early. This is before people cared about long term memory. but, Raunak and I have known each other for quite a while. My earliest memory of him is, I was a junior at MIT. I had signed up for my first [00:02:00] graduate ML course, and I had a lot of imposter syndrome about this, like, It was just something that I hadn't done. And I remember on the first day, the professor introduced to the class like, Hey everyone, meet Raunak.

He's going to walk you through your first PSET. And to be clear, like, Raunak was a freshman at this point, so this was his freshman fall. so it was kind of obscene to have me sit there, and see this person who had been doing ML research since he was 12. I ended up talking to him as a result of that.

He joined the same sort of living group as me at MIT. we were a big little pair there. So it was this sort of mentorship program as part of that, but when we really became closest, we both had this sort of middle transitionary period where he had wrapped up his previous company. He knew he was graduating soon. I was very bored at my job at Google and was vocal about that. And so we started applying to hackathons together. We did Anthropic's hackathon when they first released Claude. Won that hackathon together and just even putting aside the project just really enjoyed working with each other. When he was graduating, we just had this conversation [00:03:00] around like, hey, like I I think I want to start a company, like, would you be interested in starting a company together? And it was, at least for me, immediately a yes. There was no question in my mind that he would be an exceptional person to work with. He's a person that I admired, a person that I enjoyed working with. And so we weren't too focused on, like, the idea that we were working on. We weren't too focused on, like, whether we had funding in hand.

It really was just like, I am excited to do this, and I'm excited to do it with this person. we had a almost like hard and fast rule of if we get into YC by day one of YC, we will choose something and stick with it at least for that period. Because we were really worried about sort of pivoting around aimlessly. We have friends and mentors that we knew had had great experiences with YC.

And so we sort of just, defaulted to that. Honestly, it wasn't like a sophisticated conversation of this versus raising a standard pre seed, it really was just like a, we know really smart people go to YC, we know that they sort of go through a similar stage and process that we are in today. so we just did it, got in, really enjoyed our conversation with [00:04:00] Diana and moved forward from there. 

Brett: And what was the process by figuring out what you were going to actually apply with?

Adit: So remember all the long term memory solution that I mentioned, even though I still think it was too early then, did go kind of viral on Twitter. Like it was immediately an attractive demo. It's cool to see a language model, bring up something that you mentioned in the past. and so it was just a reasonable thing to apply with, but even a day after we got into YC, we were very upfront, like, Hey. Every conversation we've had is sort of with a enthusiast, not somebody that's willing to pay for this. So it might be an extension of this idea. It might be sort of like a platform around this or something completely different. But like, we don't think that this is necessarily the end state of what we're going to build. 

Brett: How did you even think of focusing on that first specific idea? 

Adit: It just felt like a limitation of language models of sorts. The very early form for Remembrall was actually a hackathon project that Raunak worked on. so it was just meant to sort of expand the context window of language models. effectively that was the sort of goal there. When [00:05:00] that sort of had initial traction on Twitter, there were hundreds of people reaching out asking to be onboarded. It seemed at least worth looking into it from there. 

Brett: Did you think a lot about complimentary skill sets or how we would actually build the company together? Or was it more that like, we just feel very aligned. We enjoy spending time together. There's mutual admiration.

Adit: we know that we Compliments each other. Well, I'm like I've always enjoyed sort of more of the go to markets and product side Raunak is a deeply deeply technical person But in terms of the actual decision there really was this sort of question of like, can I imagine going through some of the best and worst periods of building a company with this person? And something that I mentioned a lot is Both of us have been in long term relationships for a while, and I really see co founding a company as very similar to that, honestly. and the things that you learn in terms of settling disputes and like talking through things, not as a me versus you, but as like a we need to get through this together, translate well to being a good co founder, which I've always [00:06:00] seen in Raunak, and I hope he sees in me as well. 

Brett: What else did you write down or align on?

Adit: I honestly think that for the most part everything sort of boils down to we don't want to be dishonest with ourselves about what we are doing. I think if you ask Veronica as well, if we sort of look back on our prior attempts at building a company. For Raunak the answer is that it was really more of like a consulting business than it was like a product company. For me it was this idea of going back to like, it only matters if people are using your products, like it wasn't focused on that sort of core pursuit. we just didn't want to spend a year, two years of our lives sort of going around like a parade of pretending to build a company.

We just wanted to focus on this one thing. there's details about that, um, in terms of what we discussed.

Brett: Fast forwarding back to where we were a second ago.

what was the process by which you started to iterate away from the original memory IDN? One of the things that you mentioned was sort of willingness to pay. you sort of put this [00:07:00] out there, it was resonating, but then the sorting criteria was, do we think there's a business that can be attached to it?

Or what was like day one, two, three, four, after you sort of had some resonance with the original memory idea? 

Adit: Yeah, so we set up demo calls with everyone. We got a bunch of them onboarded to the platform. and as part of that, we would ask them about their use case. Like, what does this actually solve for you? And we would rarely ever get a real answer. it was more of this like, this seems cool, like maybe I'll try adding it to my AI application.

Like when users chat with it, maybe this will be part of it. but it was never the case that people were saying like, Hey, like I've noticed that customers get angry at my product because X, Y, and Z. that was sort of the first signal, two as we sort of started asking people to actually pay for the product, it was very clear that it was in this like enthusiast category of people are willing to pay 10, 20 a month because it's interesting, but it wasn't something that you know, their product team needed or anything like that.

so. As sort of that, like, exploration of [00:08:00] where does the product go from here, one of the more common feature requests we got was, Hey, you're managing my user's chat history. Can you manage the files that they're uploading as well? And we had a very simple, we uploaded documents, we parse it using an external parsing solution and just chunk it for you.

And interestingly, that had a lot more interest than the core product itself. we put a lot of work into improving that because the loss cases that people were mentioning were because we chunked a document incorrectly or read it incorrectly and so on. And what Reducto is today, really wasn't supposed to be a pivot, it was almost this like marketing stunt where, it's embarrassing if you look back at it now, but we had this very ugly Streamlit app.

with a super simple document segmentation tool, it would do nothing other than split your documents into little boxes of content. And that, I can't exaggerate enough, like this was a weekend project that we just sort of put together. We had the document context feature. This thing was sort of just [00:09:00] showcasing part of what was going on behind that. We posted that as a technical blog within YC's forum of here's how we segment documents. And interestingly, that just immediately clicked with a lot of founders there.

We started getting replies like, Hey, this is better than what I'm getting from Textract. Is this a hosted API? Like, where's the Stripe link? Like, how do I actually use this? and the pull for that was just so clearly stronger than everything we'd worked on to that point that we decided to really double down and that, kicked off everything that we've built to date.

Brett: Was that an easy, obvious, intuitive decision or was it one that you labored over in some way?

Adit: we'd spent enough time exploring things that clearly were not resonating that it almost felt like in comparison, we were getting punched in the face with like, uh, this is obviously something that people care about in order of magnitude more than everything we've tried.

Brett: And were you both sort of getting pull in that I could use this and I will pay for this both at the same time?

Adit: Yeah, and it was also something that we were sort of deeply familiar with by that point, because we'd [00:10:00] spent so much time building on top of that point, Google Cloud, We knew how annoying it was to build a PDF processing pipeline. we knew how much time we were putting into post processing that content.

So when other people were mentioning that they had the exact same experience as us, it was this, like, clear moment of, okay, we weren't trying to build a PDF processing company. These companies are not trying to spend X hours of engineering time on PDF processing. It's this bottleneck that's stopping them from building the things that actually matter to them.

so if we can be that sort of ingestion team for our customers, if we can take that problem off their plate, clearly that is very valuable.

Brett: When you started to think that there might be this Pivot. Did you think a lot about the long term business opportunity here? Could we create a compounding advantage or moat? You know, how competitive is this space? How many end customers are there? Or did you just sort of follow the customer pull and just said we're gonna go all in and didn't think about any of those types of things.

Adit: We did think about it. there was this immediate gut fear. I think especially [00:11:00] younger founders like ourselves are probably more afraid of competition than they should be. PDF processing, as you know, is not a new space. So every day, as we were looking into this, we'd find a new different solution.

that claims to solve PDF processing. the way that we've always thought about it is there is an initial wave that will get you part of the way there. nothing that we tried was at the accuracy level that we would have wanted for ourselves and clearly wasn't at the accuracy level that our customers wanted.

And so our question wasn't, um, is this market big enough? Because it was just obvious that people were spending so much money on this already. The question was more so like, can we be better in a meaningful way for our customers? And as soon as the answer for that was yes, we stopped thinking about it in the sense of what else is out there?

We strictly think of it in terms of how much better can we be for our customers today?

Brett: How did you figure out how much better you had to be to sort of satisfy the criteria of I will go spend the next X number of years of my life on this?

Adit: A lot of it was vibes based, 

it was trying really hard documents and seeing what [00:12:00] our initial attempts would yield versus other solutions. I think we also really believed that we had a interesting technical approach to what we were doing. A big part of why Reducto is able to be the most accurate solution is there was an old era of parsing PDFs using the metadata and just trying to read the file itself. Our insight here was that these documents were made for humans like you and I to read. Every little visual cue, like a gap between two paragraphs that's me telling you, hey, this is a new semantic piece of information.

Or a tab structure in a list is me telling you, hey, this is a sub idea of that parent idea. And so we wanted to almost read these documents the way that a human would. And that is a very deep problem, but thinking about it from that sort of first principles lens, really unlocks like a whole range of different areas that you can improve parsing on.

Chunking was a big thing that we started with, but for every element of a PDF, like parsing a table or a form region and so on, there are a lot of visual cues that matter. And [00:13:00] we've been focused on sort of capturing that entire long tail.

Brett: What was the process to get to that insight from posting this feature and sort of seeing it resonate to sort of there?

Adit: The first thing that we worked on in terms, at this point, we have six different vision models in a VLM, but the first model that we trained was for layouts. Um, so just breaking down a document into sub regions. the like very clear example for us there is as a human, when you look at a multi column layout, you know how to read it.

Like you don't think about the fact that you read the first column and the second column. but that's not immediately obvious if you just have text sitting there. oftentimes the issues that you'll find is you'll read the text from left to right. you might merge the columns together and so on.

And so we tried to break it down into like how do we understand the paragraph structure and understand the position, such that we know that these are two column layouts. and it naturally extended from there into everything else that we've built.

Brett: What was the four weeks after you sort of said, Hey, let's pivot into this [00:14:00] insight.

Adit: it was kind of all over the place. we had gotten this advice of you should launch early. You need to like get this out there. Don't be afraid of failing with your launch. But we knew that we were in a space that if we launched too early and the product wasn't good, it would sort of be a waste of everyone's time, right?

People already had PDF processing solutions. And so we decided that For this little slice of layout parsing, we want it to be decidedly better. And once we did that, we would try to launch as soon as possible. So we ended up launching, I think, two weeks after that initial blog post. and got a lot of interest on that launch because we had a playground that had a pre populated document.

You would see that this is a hard document that, Is working well here. but we also allowed you to upload your own file. and so people will almost always have some mental example of like, I've seen this file fail all the time and seeing that work for them is like an immediate, okay, this is interesting.

So we had inbound come in from all sorts of companies, startups, consultancies, and [00:15:00] large companies as well. and we also were just reaching out to everybody that we knew. We had a, what I think is a kind of interesting way of getting people's attention, which is that I would upload a hard PDF into AI applications, and I would show them that their application is hallucinating.

would be a very clear, like, here's the document that I sent in, and your model is actually making up information that wasn't there. and for any product team that cares about that sort of accuracy, that's enough to get the first call. and so we went from there.

Brett: Why are PDFs stilll a thing in 2025?

Adit: honestly, that's an interesting question. Like PDF as a standard has been around longer than I've been alive. I think it was started in like 1993 and I've met the people that worked on the original PDF to printer driver conversions and everything in between. it's a convenient format. Like ultimately it's almost like a whiteboard for the human mind.

You can. put information any which way and it's just images. part of it also is just the convenience of there's been so much infrastructure built up around PDFs that that's just how enterprises communicate today. on our end, I mentioned that we treat this as a [00:16:00] vision problem. So we actually just see them as images.

We convert them to images. and so it's not even about PDF as a standard. It's about documents and human content overall.

Brett: Did you think much about who you wanted to be your first three or five customers and did you want them to be similarly shaped or very different? Did you want the use cases to be similar or different sort of those types of things?

Adit: We knew that the accuracy delta that we were presenting really mattered and industries where accuracy matters, right? The people that we reached out to tended to be in finance and legal and insurance. simply because we knew that for them, making a mistake on somebody's insurance claim is not just a silly typo.

Like that's a tangible thing that has millions or even hundreds of millions of impacts potentially at scale. so that's the type of company that we reached out to. At the same time, a big part of how we sort of approached the initial push is we recognized that this, like, bimodal splits of startups were immediately pushing for AI features and enterprises were interested, but at the time there weren't a lot of [00:17:00] true production enterprise AI features to use reducto.

so we mainly targeted startups to start because they would integrate us by the end of the week, while still starting those initial conversations with enterprises as well.

Brett: And did you let anybody use the product or did you want to sort of white glove, be involved with how they were leveraging it and deploying it?

Adit: Even today we don't have fully self serve onboarding. So we have a free playground where anybody can go through and see that their documents work. But in terms of actually using the API and deploying it, we manually onboard every single customer.

Brett: How did you make that decision?

Adit: There's a few parts of it. One, we learn a lot from that process of, you know, hearing about their use case, seeing what they're sort of considering in their decision.

and two, there's also the second part of, we want everybody to see Reducto as this best in class product. And we really didn't want to be in a situation where, you know, somebody integrates us into their production environments. And I don't know, let's just say our infra wasn't ready to handle. the next million pages that we were hit [00:18:00] with.

at this point we are processing many hundreds of millions of pages. I learned yesterday that if you printed out every page that Reducto has parsed, it's something like three and a half times the size of Mount Everest. and you can just imagine like if we did that on day one, it would have been this horrible experience that everybody would have been able to point to and we just didn't want that.

Brett: And so once you decided that as you were kind of nurturing more enterprise relationships, you would work with startups. Did you care what type of startups or how mature are the use cases for that sub segment?

Adit: We did and we still do. a common question that I get is like, why is this not just purely usage based? And I think part of it is if we are approaching it from this like white glove experience perspective, we almost need the end customer to have a recurring use case that's as part of a real production workflow.

And so our pricing tiers and everything are Almost assuming that you're processing at a certain volume. So the startups that we work with at minimum tend to be processing above 15, 000 pages a month. in some cases, hundreds of thousands or even millions.

Brett: how did you make [00:19:00] that decision?

Adit: I think it might be partially like a visceral response to initially serving the like enthusiast crowd. there's a lot of fun and excitement with working with that, but we really wanted this to be something that's was sort of tangibly impactful, and was like real infrastructure for companies to build AI products.

And in order to do that, we wanted use cases that mattered to the end customer.

Brett: You said that this sort of got kicked off two weeks post launching the first real version of the product. What was the actual launch?

Adit: So YC has a way for you to announce that you are part of the batch. Our initial launch was a simple playgrounds where you upload the documents and see the outputs. and we just post on social channels saying, hey, we know that PDFs suck. Everybody has experienced that. Reducto's building vision models to address that.

Try it out for yourself. and I still, to this day, when I do sales calls, my framing is not like benchmarks or anecdotal data. It is really like you will see it on your own data, that this works in a way that other [00:20:00] solutions might not have.

Brett: At what point did you have the conviction that you would go spend years of your life working on the problem?

Adit: So when we started a company, we sort of had an initial, like, we will try this for two years, at least. If we stopped before then, we probably might not have tried hard enough. YC also has some interesting stats on, the percent of companies that make it to two years and how they do after. Outside of that, I really think it's almost been this like, learned love for what we're doing.

I think if you asked us two years ago, like, would you be excited to work on PDF processing? The gut reaction would be no. It doesn't sound like a fun problem, but as we've gone deeper into it, I think we've just really enjoyed what we're doing. And there are so many little moments of pride where I've seen our customers be put head to head with their competitors on socials.

And it will be like the PDF upload feature where people will say, like, it is 10 times faster on this AI application than this other one. and of course, like, it doesn't have a Reducto logo slap there, but like, we know that we are helping power that. and that is enough for us to want to work for [00:21:00] this, work on this for who knows how long, 

Brett: When did you think you had product market fit? 

Adit: I don't think that we really thought we had product market fits until our enterprise conversations started getting escalated in a way that we didn't expect.

As you know, there's a very large company, that powers a really broad range of use cases with Reducto. And we saw through that process that they, at one stage in the sales process, they brought 14 of their engineers to meet with us for eight hours of that day. their entire team basically just sat with us to learn more and the depth at which they cared and the way that they clearly thought about details for this, Showed that it mattered to them, and that I think was the clear signal of if it matters for these people and it matters for the other people that we're talking to, then clearly what we are working on matters overall, and from there it's just a question of are we the right people to do it?

Brett: You mentioned that sort of when you put the word out for the very early version of the product, that a bunch of the early customers were startups for all the obvious reasons, and you started to nurture [00:22:00] enterprise relationships. What was the source of those conversations and what did it look like to actually start to engage them?

Adit: It's a broad range, in some cases it was me asking friends at those companies for intros, Scale. ai is now a customer. my roommate in college was on a team there. And so I asked him for an intro that led up to the engineering manager and it took about a year for us to plan them as an actual customer.

 There's a, a fortune 10 customer that we signed on relatively early. we actually really lucked out in that they tried. a document on the playgrounds, and decided to sign up for a demo call, they almost landed in our lap in that sense. So it's been all over the place. A lot of it was just doing whatever we can to find the companies that we knew had this use case.

Brett: For enterprises that took a year, how did you debug If it just wasn't an urgent and important problem versus they're a large enterprise, it's going to take 6 to 12 months to get the deal done irrespective of market puller enthusiasm?

Adit: I think the clearest signal is [00:23:00] how much our champion seems to care. So there are enterprise deals that take a while because their legal team is taking forever. You know, there's just hoops that you need to jump through. And there are enterprise deals that, I don't know, maybe I'm reaching out to the champion and they're just never responding to me and I'm almost like Begging them to hop on a call. You can very clearly tell which bucket you're in. Um, and at first I was in the camp of like, every lead cannot be dropped. it's almost sad if they don't hop on the next call with me. and I really changed my perspective over time. do you know Ryan from Vendor by any chance? 

Brett: I know of, but not personally.

Adit: gave a fantastic sales call. Like overview to our YC batch that just changed the way that I look at this. Framing was just the best answer you can get in sales is a yes The second best answer you can get is no and your job

Brett: And the worst is the long drawn out potential maybe.

Adit: Yeah, exactly and So now we're at this point where if we're in the buckets of the person that we think should really want this clearly doesn't care, it's maybe not the right time and that's okay. but if that [00:24:00] person is trying really hard, then we put a lot of work into to make sure that whatever other logistical steps there are, we're doing what we can to help fix this. 

Brett: In the first six or 12 months of the company's life, how did you decide what you wanted to build yourself from scratch versus what you wanted to leverage other infrastructure, open source, et cetera?

Adit: So we had a sense of what the minimum viable version of a document processing product would look like. it should parse tables, it should parse images. and to start, we cobbled together that pipeline with a combination of open source and our own models. Over time, we've sort of expanded to cover that full pipeline. But actually, , I think for what we do, we are not this sort of platform. There's no like major UI components and all of that. In that restrained scope, a lot of the way that we've been looking at this is not a build versus buy decision. It's a, does this feature even need to exist? Like our customers really, really asking for this, or is this something that we are sort of lying to ourselves about it being important. And so, for a lot of what we've built, the reason why we've been able to do it with a [00:25:00] small team is we've constrained the set of things that we do say that we do. We do those things very, very well, which means we haven't had to sort of take off the shelf solutions meaningfully in our product. 

Brett: So maybe on that point, how did you go about figuring out what to build in what order? And kind of to the point that you were making, what is the role of what you're hearing from customers?

Adit: we have Slack channels with almost every customer. Many of them have mine or Raunak's phone number so they can call us when, they get a new client and that client needs something done. I think at this point we are fortunate to have enough data points where we can see that dozens of customers are asking for this one thing and it's like meaningfully impeding on the value that they expect from us. And those are obviously the types of things that we want to prioritize. In the early days, a lot of it was intuition about what we would want if we were building a product on top of Reducto. and almost like being the type of company that isn't blocking you from getting what you want. 

Brett: Did you think at all about what you were hearing from [00:26:00] customers who didn't buy but would if you had this set of features versus existing customers that wanted feature one, two or three?

Adit: the way that we generally think about that is there are features that fundamentally are very transferable. And there are features that are very, they're almost like point solutions, right? And so when I think about a, I don't know, let's say somebody comes to us in healthcare and they say this type of document needs to work, the limitation there might be our table parsing and improving table parsing there is not a healthcare specific thing.

It's something that will improve our table parsing for customers in finance and insurance and so on. And so those types of things we will focus a lot on. But there are customers that will come to us and maybe they're looking for some niche file extension, or maybe, you know, they have an internal documents type and that is the only thing that they care about getting parsed. given our team size, that's not the type of customer that we've decided to chase. We're very upfront about whether or not we are or are not the right solution for them. 

Brett: How structured or unstructured are interactions with customers over the [00:27:00] past couple years? Like, do you just have informal chats in Slack and use that roadmap or is there more structure to it?

Adit: For the larger enterprise customers, we do. Standard like quarterly meetings with the leadership and all of that. for a lot of our customers it's really just like they will text me or they'll message on Slack and we'll get lunch or dinner and we try to have a regular cadence of just meeting with them in some capacity whenever we can. 

Brett: I thought it could be interesting to talk about, some of the biggest customers that you landed, obviously the ones that you can talk about and maybe sort of share the origin story of a few of them, and sort of how the deal came together. 

Adit: There's a very broad range, I've mentioned the startups, I've mentioned the trillion dollar market enterprise. But there's a lot of companies in between there where Scale AI is a customer, there's a really large set that I unfortunately can't name today. But what we've seen across these companies is they're incredibly technically talented, and they have a more premium products that they sort of have impressions that they want to maintain there. And at [00:28:00] every stage of our conversation, you can tell the difference between how a company like that approaches their vendor evaluations, the features that they're building, and that they aren't looking for like a good enough solution that they can sort of stop thinking about this problem for. They care a lot about the edge cases of our processing. they will ask us a ton of questions on like how we came to a certain decision. And that rigor of building exceptional products I think forces us to also build an exceptional product because if we are not at the bar, they can't be at the bar for the products that they're building on top of Reducto. So yeah, overall we've been very lucky to have companies that I consider to be, you know, some of the best product teams in the world choosing Reducto. we're hopefully going to continue seeing that trend over time. 

Brett: Maybe sort of on a similar thread, one of the ways that you've approached sales, and I think one of the ways that. Folks who are great at sales approach sales is you often get very deep with your customers long before They've signed and then you spend a tremendous amount of time in some cases going to their office [00:29:00] weekly. Are there any interesting sort of things that you've picked up that you just thought have been intriguing to you as you've had this chance to work with and study so many different companies, maybe that is different than when you basically had one true large tech company experience in Google?

Adit: I would say it's the pace of iteration on AI teams in particular is A lot faster than I think people give it credit for, because we are in this phase where Every week Twitter will have this, this is the new best thing and here's the thread on, like, why you need to drop everything else.

And so people have a lens of skepticism, but they also have clearly developed frameworks for rapid experimentation and sort of plugging and playing with different things. So we end up learning a lot about what the industry is doing and where it's headed as a result of what our customers are doing and what they're asking of us as well.

Brett: Maybe you sort of partially answered this in sort of the last answer, but, how did you balance as you were spending time with big enterprises and their needs and wants relative to mid-market or startup needs and wants?

Adit: I think we are [00:30:00] lucky in that the most likely thing that a user will complain about is that the parsing output wasn't what they wanted it to be. When we talk to enterprises, we are very focused on what we say is our scope. It's actually been helpful on that lens of they have the engineering bandwidth to build things around it if we say that this is not what Reducto does. we sort of solve the problem by just having the offering be almost the same to both mid markets and enterprises. There's some details that differ in terms of the scale at which they're operating and so on. Those we will, of course, take care of because we don't want it to be a reason why they can't use the products.

But outside of that, that's my answer. 

Brett: What are the most important things you've figured out in terms of founder led sales thus far?

Adit: One is that it's hard to quantify, but caring really, really matters. My background is, of course, not in sales. Like, I don't think the reason why we are selling at the scale that we are is like a matter of wordsmithing the answer well or anything like that. it really is just like a obsession with getting it right. and I think [00:31:00] customers feel that. Whether it's another startup where you're talking founder to founder or even enterprise, like they can tell that the way that they're talking to you is not the way that's like a scaleup's BDR would talk to them. And we've seen this for the fortune 10 that I mentioned, we actually certainly were not like a full feature parity option compared to some of the other vendors that must've talked to them. But A big part of what they were betting on wasn't just the immediate state of the product. It was me and Raunak. It was them seeing that when they brought up a missing feature or had a complaint about an edge case, we would have it fixed for them an hour later. And that sort of energy I think is almost contagious through the sales process. When people can tell that you care about the products, they start to care about it more too.

Brett: Did you think deliberately about, do you want to verticalized in some way and have even early on, kind of spend the next few months just focused on healthcare companies and shaping the product for that, and then we're going to go after financial services and sort of those types of [00:32:00] things?

Adit: We did briefly on a go to market perspective, obviously it would be easier to sort of be a healthcare specific document processing company and to be able to point to the entire industry for that. But I think a big part of what we are doing, going back to what we were talking about earlier is, it's a very different problem to be able to parse a specific type of document than it is to be able to not know what's about to come in and still be able to accurately parse that. And to that end, even though it's harder for us to do this, it is very valuable for us to not deal exclusively with healthcare customers and not deal exclusively in insurance or finance or any given use case, because the learnings that we have from each type really transfer across them and they've made the product into what it is today. 

Brett: So did you make a deliberate decision that we're not going to verticalize and that kind of being stretched in all these different directions was going to, I guess, maybe avoid the local maximum sort of problem?

Adit: We just saw it as this like forcing function almost of this is what it will take [00:33:00] to build a truly exceptional product. We really value being able to see data points that are outside of our distribution that we've just never had to deal with. And you can only do that if you're a horizontal product.

Brett: You touched on this a little but from a product building perspective, it seems like most effort has gone into making the core document processing engine better and better. But I assume as you spend time with customers that there were all sorts of adjacencies or potentially net new products that you could build alongside it that would be accretive from a revenue generation perspective. How do you sort of think about, do we just want to make core better, do we want to stamp out adjacent use cases, sort of those type of early trade offs?

Adit: So this definitely matters a lot. Part of the initial interest for Reducto is RAG as a paradigm introduced this need for thinking about chunking, for example. So it wasn't just, can you parse the documents? It was, after you parse it, what do I feed into my Vector DB? That was a Initial selling points, like that was a big part of why people would sign up for Reducto. And I almost [00:34:00] never get asked about chunking today, like we've built up this entire pipeline for it, it's incredible, but what we've sort of come to realize is a lot of what we want to do and a lot of what our users want to be able to do is downstream of understanding the document exceptionally well. So today we do have features that go beyond, you know, just reading the document. We have features that will let you almost do like process automation on top of Reducto, where people will classify their documents, they'll split their documents, they'll do structured extraction on their documents, all with just Reducto. But every single one of those things is downstream of our ability to parse the documents. Every single one of those things benefits from our ability to parse them really well. So in terms of what we spend our time thinking about, Even though we have these other features, a majority of it is still that like core, if that makes sense. 

Brett: Do you think intentionally about resource allocation, X percent, insert a core document processing versus not, or is it just more intuitive?

Adit: It really is just intuitive. Um, we're a small enough team that this is just a weekly conversation that we have. There's always going to be sort of fires [00:35:00] or things that we need to put out that take time away from that. But a majority of our time is on that core problem. 

Brett: Did you spend time thinking about, you know, we're getting these different requests from customers. This request might lead to X dollars in revenue and this might lead to Y dollars in revenue. did you think about the difference between pain points and like the value of solving those pain points?

Adit: We do. Equations are one example. For months, we would have people reach out to us and say like, hey, I have equations as part of these research papers. Can Reducto parse this? And our answer would be no, because we knew that before we sort of focus on these one off things that some people were asking for, we needed to fix issues like table parsing because 80 percent of our customers were asking for this and the dollar value of being able to parse any table well is just orders of magnitude higher. So it does factor in, I don't think that it's like a rigorous exercise for us.

Like we don't do a market opportunity analysis for each feature, but there's almost like a gut [00:36:00] instinct of what will actually move the needle for the company. 

Brett: How did you begin to think about where the long term edge in the company is going to come from?

Adit: A big part of sort of like the, the source of my optimism about where we're headed is the, the nature of what people need to do with these PDFs is changing. and what I mean by that is all of the large cloud providers have sort of like the de facto PDF processing solution that people have used for a while. But where we are today is you're getting these models and these agents that are incredible at reasoning. Anybody that you talk to is convinced that over time, more and more business decisions and actions will be made by these agents and models. And if you're moving to a future where humans are not really part of that loop, but you have this massive repository of human data and intelligent systems that need to interface with that human data, it's an infinitely [00:37:00] valuable problem to be sort of the interface that connects it to. The U. S. alone has on the order of 4 trillion paper documents, um, at some point those will be scanned, at some point those will need to be fed into GPT 5 or 6 or whatever else is out there by then, and if we continue to be the most reliable way to parse that data if we continue to be the company that is clearly the way to improve your end reasoning as a result of what you're passing in. I don't know exactly what the number for that opportunity is. I'd be guessing if I said something, but I know that that is incredibly important for the world. 

Brett: In the early phases of building the company, how have you chosen where to go very slowly and where to go very quickly?

Adit: Hiring was the first thing that came to mind for where we went slow. I think we were at four employees when we crossed a million in ARR for the first time. and it's something that I sort of heard both sides of the story for. Um, of like, we [00:38:00] should have been hiring faster or like this was the right thing to do. But we care a lot about the efficiency of each person. And I don't mean that in terms of like, um, performance management perspective, I mean that in terms of having a large unwieldy organization was never attractive for me and Rodak. and so we were very thoughtful about each individual person that joins. our founding ML researcher is this guy that did his entire PhD in document processing. Um, and he had the best open source models in document processing and nailing that one person was way more important to us than like trying to get a 10, 20 person engineering team that didn't care as much. In terms of things that we've tried to move fast for, I think part of this is like the almost reactive nature of being an AI infrastructure company. Especially today, the rate at which models improve is honestly insane. And there are paradigm shifts on what matters constantly. I mentioned that RAG was sort of the, like, go to reason to use [00:39:00] this, and chunking really mattered then.

It doesn't matter so much now, and now people care a lot more about the embedding approach for our chunks and so on. And we need to be at the forefront of what the best ingestion looks like for today's models. 

Brett: Are there any areas where you've changed your mind, where you were going quite slow and now you've decided we have to go very fast or vice versa?

Adit: Two things. One, we are newly starting to ramp up on GTM. at least so far, I'm the only person on GTM at the company. But we're finally at a point where I really think that the product is resonating with people. And that was sort of like the key gate that we went across before we worried about hiring our first few GTM hires. And the second is on the rate at which we work on the core model work. We started with this constraint, we will only do PDF processing because we don't want to sort of, there are others in the space that sort of extend across all file types, but not at the same accuracy, and we really did not want to be that company. but we are at a point where our [00:40:00] customers don't want to be sending only this file type to Reducto and like sourcing their own pipeline for this other file type and we need to be almost like the unified ingestion pipeline for them and that just requires hiring more exceptional engineers to be able to do that. 

Brett: On the GTM hiring, to the point that you're making, you've made it very far with no GTM org other than yourself.

And you said that sort of it felt like at the right time because the value prop was resonating with customers. does it actually mean to resonate? And do you think that three months ago wasn't the right time and nine months ago and twelve months ago or knowing what you know now you could have started to build a GTM org earlier?

Adit: that's a good question. I don't have like a exact month breakdown for when it would have been okay or wouldn't have, but a big part of how me and Raunak see this is, I learn a lot from the sales calls that I hop on. It's not just a matter of learning about what the customers need. It's a matter of learning where the product needs to be for people outside of [00:41:00] the voices we already hear in Slack. I really didn't want to outsource that too early. It felt almost deadly for the company to lose that sort of pulse with customers. But we have reached this volume where I know that there are more companies that are ready to onboard with reducto than we can handle onboarding. we are.

supply constraint, not demand constraint. And that's the point where, it's not that I'm taking a step back from GTM. It's just that there's so much interest in what we're doing that we shouldn't sort of be our own bottleneck. 

Brett: And how did you decide as you're going about building the GTM work for the first time in what order, you know, you could more hire on the success side or sales engineering or post sales and free you up from that. You could obviously have people sort of building pipeline, you could have AEs, you could sort of do anything.

What, what's been your sort of process or thinking about the first couple, roles you're going to hire for?

Adit: Our first GTM hire is likely going to be our founding customer success manager. The thinking there is really just, it is more [00:42:00] okay for us to maybe not prioritize net new ARR. It is not at all acceptable to me for existing people that trust Reducto to not continue to be this excellent solution for them. I never once Reducto to be at a point where we are dropping the ball on our customers. And so post sales and making sure that they're getting value from it's that we are listening to what their needs evolve to, as the first function for us to solve. Once we are confident that every customer that decides to commit their confidence to us, is happy and growing with Reducto, then it's, of course, a lot more important to get more people in that pipeline. We sort of sequence it as we're hiring a BDR first, and then at some point we'll hire a junior AE to focus more on the SMB side. And then we're going to hire enterprise AEs. I mainly see it as a function of what is this sort of step in the pipeline that I'm not even learning from anymore. I don't think it shapes my understanding of the company or the products to do outbound content, for example. And so a BDR is the right person to hire [00:43:00] for that. And we're just sort of progressively stacking from there. 

Brett: You mentioned this a little bit, but for technical founders who have not sold to enterprise specifically,

are there any other things that you figured out that would be useful for them to understand sort of either as global principles, or maybe if you're selling to startups versus selling to enterprise, here's the key things that you need to know.

Adit: I think if you're a technical founder, you are probably somebody that approaches your own buying decisions from a very rational perspective. you're probably, as you're sort of thinking through your startup stack, Choosing whatever you think performs the best, which is in fact how our startup customers treat Reducto. it's very objective. I don't think that that is true at least in my experience for enterprise sales Like it matters a lot what our accuracy numbers are in their benchmarks But it's so clearly Relationship driven in a way that I at least did not know as somebody that doesn't come from an enterprise sales background. Enterprises are these massive behemoths of [00:44:00] organizations where it's not just a matter of is the team that you're working with happy, there are leadership people involved that maybe you didn't know to talk to. They're just like a whole legal process and a security team and a procurement team. And the only way that you will navigate that efficiently is to have a set, or at least one person who's really willing to sort of champion that process for you. and spending time with that champion, like making them excited about not just the products, but about working with you is way more important than I would have known, two years ago. 

Brett: So specifically the depth of relationship in the champion that's going to help you navigate everything?

Adit: Yeah. Our seed round investor, I think is exceptional at this, uh, Liz. and a lot of what I sort of think through is she mentioned that a lot of her early customers were at a point where like she would feel comfortable inviting them to her wedding. They're friends for her, not just like a paycheck. And the more I sort of think of it from that lens of really just getting to know them as people, the easier everything else becomes. Like you just have a natural [00:45:00] relationship. It's not quite as transactional as it would have been otherwise. 

Brett: One of the interesting things I think is when a lot of people talk about. how their company got into really strong product market fit. It's often like nine years after they start the company, right? And they're at some event and it's coming up and They sort of paper over a lot of what happened just because it was so long ago and it kind of tied up into a nice bow.

One of the nice things about talking to you now is that you're sort of recently into this really strong product market fit where you kind of have this dynamic where both you have really strong market pull, a deeply satisfying product, that's solving a real problem and thus translating into an economic model that actually works for the company. And so I'm curious because it's sort of fresh in your mind, are there other things we haven't talked about that are broader lessons that you think other founders might find useful who are six months away from starting a company or three months into starting a company?

Adit: The number one thing that I've learned is actually not a specific skill. It's [00:46:00] almost just this general idea about how malleable you as a person are. And what I mean by that is I think on day one of starting this company. I had reticence about doing things that I wasn't comfortable with because there was almost this like fear of doing it poorly. The first demo call would be daunting. Prompting a user to pay even after they're telling you that they're interested in the product is like weird thing where it almost doesn't feel right. You get so many reps in over the course of building the company of like, Sucking at something, doing it again and again and again until you become kind of good at it and you almost forget that you were bad at it to start. Those sorts of like skill issues kind of become exciting, where now when I hear that like, Oh, maybe I'm doing a horrible job of post sales of like growing contracts over time. That's not a bad thing. It's, it's not like, a negative reflection on me. It's more so just like an opportunity for us to [00:47:00] unlock this new lever for the company. if I was talking to myself and trying to instill something in myself six months before starting the company, it really would be this idea of just do the things that you know you need to do, without a fear of failing at them, because you are going to fail.

And that's, just not important in the broader picture. 

Brett: Do you think that's always been the way that you've seen the world? 

Adit: I don't think so because especially when I was, you know, working my first job as a new grad, I don't think you get the same level of opportunities to fail. I think it's a very important thing to learn how little it matters to fail on a micro level. and so it really is just something that we got as a result of working on the company. It wasn't something that I think is, you know, intrinsic to me or intrinsic to Raunak or probably most of their founders. 

Brett: What is sort of the meta idea about what enables startups to even exist, particularly in this context, which is you're not creating a new market. You're just doing something better?

Adit: This goes back to sort of this idea if we didn't want to scale the team too [00:48:00] aggressively. I generally think when you have a small set of people working on a really important problem. each person there is really forced to go deep into the nuances of what matters. What that leads to is if you're an engineer in like a 40 person organization focused on PDF processing, you probably have a very siloed sense of, you know, what your task is for that quarter for that year.

But for us, like every single decision and feature that we're working on is in this, like very real context of Scale AI needs us to parse this document. We've seen this type of document fail for this other customer. And that, I think really changes the way that we approach our work. It changes it on a technical level because we have, you know, tangible things that we're working towards.

But it also changes it in the sense of how much we care about the problem because we know that there's somebody that's waiting and counting on us to do it. The problem isn't abstracted away into like a dashboard of numbers. and that might sound abstract but I think that's a very real competitive edge and it's probably been true for every company that's [00:49:00] going up against big established players like we are.

Brett: I think that that's a great point. And I would assume that large companies understand this, but there's something about the vessel of four people that really want to make this thing work.

There's some magic in that.

Adit: I don't know how to quantify it. Clearly it translates into something meaningful and we see this where one of our large enterprise customers had an internal document processing team, like they had engineers staffed on the same problem. And yet at the end of the day they ended up choosing Reducto because for them they saw that day over day Reducto was getting better.

It wasn't like a month over month or a year over year They were watching the product improve over the course of their like weeks long trial, and that matters in ways that you can't like quite put a label on.

Brett: Something we haven't talked that much about thus far is kind of how you chose to capitalize the business and maybe In the few rounds of financing that you've raised all with relatively short time horizon, you can sort of share a little bit [00:50:00] about how you approached it and maybe are there any higher order bits that are useful to other founders that are starting to raise capital?

Adit: So, as you know, we went from pre seed to seed to series A over the course of a year. I don't think that that is a testament to our fundraising skill set. It's more of a testament to the business itself. We've been incredibly fortunate and the reason why we raised that funding isn't actually money itself.

We actually still haven't spent our seed funding. It really is because the investors that we ended up raising from, we really admired the way that they thought about the company and thought about the problems that we were dealing with, that we voiced to them. If I were to give advice to other founders, and again, I'll caveat that, I think in any sort of mutual selection relationship, there's always one party that thinks that they are worse off in that transaction. that applies to people dating that applies to like you selling a product and somebody buying the product. and that applies to fundraising as well, where oftentimes when you're selling a product, you sort of come at it from this lens of like, I need this [00:51:00] person to buy and like, there's a desperation that comes with it. And there's a similar power dynamic by default with companies raising from investors because investors meet thousands of companies and the default answer is no.

but If there's one thing that that's you sort of need to keep in the back of your mind as you're fundraising, it's really just this idea of what you are offering, this ownership and what every founder hopes will be a generational company is incredibly rare. It is ideally like a opportunity that people don't come across often.

And so there's details of communicating that that I'm sure people workshop and improve over time, but the fundamental thing to like keep in mind and to try to strive for is to be in a position where ideally you don't need to fundraise. Because when you come at it from that perspective, you're not attached to the outcome of any given investor call and you really start to choose not based on like I need dollars in my bank today, you start to choose based on, will this person be the person that I reach out to when I'm confused or struggling with a core moment of the [00:52:00] company?

Brett: In the case of raising your seed from first round and your A from benchmark, what's sort of the way in which you prosecuted the relationship in this case with Liz, or Chetan, that sort of gave you the confidence that they are the people that you would want to be an owner in your business?

Adit: The number one thing that was most helpful for us was actually talking to companies that raised from them that didn't work out. think it's often the case that you find out what it's truly like to partner with a person and the bad scenarios as opposed to the good ones. For both Liz and Chetan we directly had incredible interactions with them, but everybody that we reached out to trying to like poke holes in their story didn't have a like instance of vacuous interactions that are cases where they like really let them down. Both of them Across their relationships clearly put in so much effort like their heart and soul into the people that they were partnering with and that sort of consistency I think is very rare and it's made it obvious to us that this is somebody that we [00:53:00] would want To work with as well.

Brett: In the sort of 18 months since you've started the company, where do you think the luck has been?

Adit: All over the place. I think we got really lucky in terms of our fundraising process, not in terms of the dollars, but in terms of the people that we get to work with. I think I got exceptionally lucky in terms of who I landed as a co founder. Like, I don't regret that decision whatsoever. We were super lucky to have very understanding early customers that, you know, saw the promise in what we were doing, even if, on day one our uptime wasn't three nines of SLAs.

Like, they knew what we were working towards, and it was almost like they were building Reducto alongside us. I don't think we would be where we are today if you sort of stripped away all of those, those happenstance encounters. 

Brett: What about sort of the inverse of that? What are the most skillful things that you've done thus far?

Adit: I think a lot of it is the product work itself. I really think the work that we are doing, not just in terms of how we think about the models that we train, but also in terms of how we think about what [00:54:00] the product experience is like, what is and isn't in scope, sort of prioritizing that to really make the seamless experience end to end, is not sort of like a happy accident.

We haven't been throwing stuff at the wall to see what sticks. We are very, very thoughtful about the things that we work on. The way that our team works is, we'll of course have small tickets, but every single week each person has only one thing that they are focused on. and when we do our weekly team meeting, that one thing is what we discuss, there's a singular focus for what we are working towards, and that's transpired into what we've been able to build.

Brett: Are there other rituals like that? This idea of a singular thing that someone is accountable for? Are there other things like that in terms of the way that you built a company that you think have had an outsized impact?

Adit: I generally think that we lean on the side of not forcing structure for the sake of structure. This is sort of me like rubberbanding from seeing what I was annoyed by at Google, we don't do daily one on ones. we don't have sort of like extensive write [00:55:00] ups for the work that we're doing for the sake of showing that it's a lot of work.

The way that we approach product building is we have a channel on Slack called Daily Updates, where everybody will post whatever's working well or isn't. If you need to talk to somebody, we are fully in person, so you will just talk to them for the time that you need to. And the only structure that we really have is that week over week checkpoints.

if this is what we said we were going to do, here's the impacts that, we actually landed versus, like, we need more help here. And that's worked out well. We really try to see it as everybody's solely responsible for the thing that they're working on. If you need additional hands, like you need the whole team to stop what they're doing and just label data with you for the next three hours, each person on the team will sort of fight for that and I love the energy that that creates.

Brett: How did you focus on one thing? Don't you have customers asking you morning, noon, and night for things?

Adit: This applies to everyone at the company, not just me, because I mentioned that we have Slack channels with everyone. So we'll get requests and tickets throughout the day. That's something that we're still figuring out to be honest. it's just a, a [00:56:00] natural part of building an early stage company is just, there's so many things pulling at our attention.

To the best that we can, we try to sort of, silo those into like directed periods where we're focusing on customer support and so on, but we do try to block off time for the things that we see as core fundamental efforts for the company.

Brett: In what ways did effectively never having a job help you both in building the company? And obviously you had a brief stint at Google, but it was probably short enough that it, you know, it wasn't like you had a 15 year career building and scaling companies. So in, in what way did sort of not really having work experience been immensely helpful to you building the company?

In what ways do you think it's sort of been made the journey more difficult?

Adit: I think a lot of what you see as like default practices, like Google created the structure of OKRs, for example, are things that people repeat at their startups because they've seen it work that way in the past. And for better or for worse, Raunak and I have a sort of fresh slate that we are working off of.

And so as we [00:57:00] think about how the team works together and all of that, we can try these things because our mentors will mention them and they'll say, this is how we do it. And sometimes those things work well for us and we stick to them, but there's no sort of coercion of every great engineering org does this and therefore Reducto must do this as well. it really is just like a, we try it. If it works, great. If it doesn't, that's fine. Like we will find our own process for it. 

Brett: Is most of the time when you're trying to figure these things out, it's mainly just trial and error to the point that you're making?

Adit: On some level. Yeah. we will have a Friday team meeting and we'll say, Hey, We want to try this structure for the next two weeks, we'll all write a one pager for what we're working on, and at the end of the two weeks, the team will just honestly say, like, I didn't think that this was helpful, I think that it distracted me from the time that I would have spent working on it, and of course that's a hypothetical, but that's fine, there's no attachment to that idea, it's, we will do whatever it takes for the team to do their best work. 

Brett: What have you found is the importance of just working [00:58:00] insanely long hours thus far in building the company?

Adit: I know startups that almost like measure themselves on the number of hours. And I don't think that we are in that bucket. Candidly, of course, by nature of what we're doing, it does sometimes require that, but we really try to focus on the end output and the quality of that output more so than the inputs on that sense. And so I couldn't tell you how many hours each employee works because it's not something that I've logged anywhere. They're doing incredible work. Sometimes that means late hours in the office and weekends at the office. Sometimes they don't need that. And as long as our customers are happy at the end of the day, like, that's what we care about. 

Brett: Maybe just to wrap up, we could end where, we often do, which is, who's someone who's had an outsized impact on how you've approached building the company or maybe it's something smaller and kind of what's the thing that they've imparted on you?

Adit: For me personally, one of my first managers at Google, this guy named Ollie on YouTube ads. changed my [00:59:00] perspective on what organizations need to look like. And what I mean by that is ads is one of the largest software organizations, maybe ever. There's a lot of bureaucracy there. And I remember coming in as a new grad expecting that sort of bureaucracy, but Ollie in meetings would just. interrupt and say, guys, this doesn't make sense. Like, why are we doing this thing that we know is going to have close to no impact? and there'd be this like awkward silence after where everybody sort of knew like, it wasn't the right thing to focus on and somebody just needed to say it. I think after seeing him do that again and again, I sort of have an internal monologue of is this thing that I'm doing just me trying to produce output for the sake of feeling busy?

Or is this like truly the thing that's going to move the needle? And we try to do that on every level of the company. And I think that's been very, very helpful. 

Brett: Nice. Well, thank you so much for the conversation.

Adit: Thanks for having me.