Here is the full transcript of the conversation between Fund/Build/Scale podcast host Walter Thompson and Dipanwita Das of Sorcero:
Dipanwita Das 00:00
If we didn’t have a product in market in that established, repeatable, scalable way, for a while, because the market was also pulled between, “am I buying AI? Am I buying a product? Am I buying what?” So we also chose a market that wasn’t fully formed. And it’s a market that’s still changing and still evolving.
Walter Thompson 00:28
That’s Dipanwita Das, CEO and co-founder of Sorcero, a life sciences startup that applies advanced analytics to medical and scientific content to improve patient outcomes. Before her team started coding, she interviewed around 300 people, synthesizing those customer discovery conversations into product and technical requirements she shared with her co-founders and early employees. We discussed aspects of her company’s journey from seed to series A, such as their R&D approach, aligning customer personas with product development, and how she learned to hire people who can handle ambiguity while building something from nothing. I’m Walter Thompson, this is Fund/Build/Scale. Fund/Build/Scale is sponsored by Mayfield, the early-stage venture capital firm that takes a people first approach to helping founders build iconic companies. The podcast is also sponsored by Securiti, pioneer of the Data Command Center, a centralized platform that enables the safe use of data and Gen AI. Dipanwita, thanks very much for joining me here today. I appreciate it.
Dipanwita Das 01:42
And thank you, Walter. It’s good to be here.
Walter Thompson 01:44
So tell me about Sorcero, the company launched in 2018, well, before the current AI boom. How did you initially connect with your two co-founders.
Dipanwita Das 01:55
So before I started Sorcero, I had a digital communications agency based out of DC. And we had a number of public health organizations as customers. One of the things that we were tasked to do was to build scientific communication platforms for them, i.e. platforms that they would use to train and, and inform very different stakeholders, from lay people to advocates to legislators on important public health issues, medical ramifications, economic ramifications of those issues. At scale, obviously, in our case, we were doing it one by one, we would go into a new country, we’d set up this platform, think of it as a combination of training tools and a software environment. And I started looking for efficiencies and scalabilities in whether the software was available in the more corporate world, i.e. the life sciences have their software, because it is their job to effectively communicate the use and safety and efficacy of their products every time they take a new one to market. And it’s also marketed to very different people, the patients, the doctors, the regulators, and everyone else. And I found that that did not exist. Obviously, this was a market opportunity, and one that aligns very well with my personal goal of doing things that were generally good for the world. And that also interacted very closely with my ongoing interest in adult learning and development. How do you equip the workforce with new and better emerging technologies so they can do their work better? I already started noodling on the idea where I met when I met Walter Bender, who’s now our Chief Scientific Officer, at a colleague’s house for dinner. And we got started talking about AI, about personalized learning, about personalization of content itself, and saw that there was a real opportunity to take some of those foundational concepts and start to transform how an industry dealt with content, dealt with customers, and actually ended up making sure that their life-saving products could actually get to the people who needed them the most. Our third co-founder, Richard Graves, we’ve known him for a long time. And he was already a very well established entrepreneur. And I was looking for someone who was very good at the zero-to-one kind of going from whiteboard and a couple of ideas, to the first product to the sales to setting things up. And so he was a natural addition. So Walter looked up to technology, I had more of a product vision. And Richard, who is now our Chief Commercial Officer, started to take on what is commercialization of the product itself.
Walter Thompson 04:46
So it seems as though early-stage teams, they’re all kind of self sorting as far as different roles. So I think I might know the answer, but who’s the chief storyteller? Who is the main marketing person at Sorcero?
Dipanwita Das 04:59
I think it changes from audience to audience quite a bit. But I would say to the world of customers and partners, it’s, it’s very much Richard. And it continues to be him and his team actually now. And you need a very different kind of storytelling and you kind of need an evangelist for your product. I think when it came to painting that picture to our team, to our product team, as we hired a CTO and others, that was much more so me, where I took all of the customer discovery I had done and I had spoken to over 300ish people, combined it with my vision for the world, and Walter’s vision of applying AI and technology in certain ways in service of human decision making. And that informed that internal team, as well as the board and investors. So it’s really, depending on the audience is the right answer.
Walter Thompson 05:54
Let’s talk about the initial R&D aspect. How long did that take you? How much time did you spend doing R&D before you actually brought on your first customer?
Dipanwita Das 06:02
I would say both not very long and quite a bit. So we brought on our first former customer in a use case that we still serve today in 2019. So that was pretty soon after, and it was a top-10 pharma. But R&D For us sort of is in two halves. One is on the technology, the infrastructure and how you apply AI? And how do you tune these models? And how have you deployed safely? And how do you do the data? And how do you know there’s that whole world of the platform itself? And then the second part of R&D is how do you get customers to use it? And how does it go from being technology to being a product? And how do you insert it into their workflow? And then our first customers in 2019, and 2018, and 2020. I’m continuously they served to inform and fine tune that part of the R&D. I would say that the tech took us a good long time. You know, we have a lot more technology in the market right now in terms of middleware, and choice of foundation models have all sorts of infrastructure to deploy multiple AI services into a single product that didn’t exist when we started thinking about LLs, and transformers and AI. So we started having to build a lot of it ourselves. And as we have matured, we swapped out our initial builds with production grade from the market, and that R&D continues. And then now our R&D in AI has shifted and more on the AI services. How do you use all of these great models out there? What is good for which task and deployment security and fine-tuned to industry? And then the second part is the user experience? How do you actually embed this kind of software into workflow for an organization in a market that’s never really had software-driven workflows before? And so that’s definitely a goal-created discovery process?
Walter Thompson 07:58
Is it correct to assume that most of your initial hires were people who were deeply technical, they were steeped in this knowledge, they weren’t coming into this new? How do you build a team like that and transition from academics in theory to people who can turn that into an actual software product? Was that an easy process? Was it painful? And if so, how?
Dipanwita Das 08:20
I think it was a mixture of both. And I love your question, because we had to make exactly that transition. So when we made our initial set of hires, I did hire folks who knew the industry into the commercial org. So we knew what industry we’re going to sell in two. So we definitely hired people who had sold into our market into that industry. And that helped immensely in not making us look like absolute idiots when we talked to customers. I think, on the other hand, on the technology side, the culture shift was more from prototyping, and hacking and spinning something up on your laptop, to a production grade environment and cloud deployments and SOC2 and all of that. And I think it took us a while to make that transition. Because at the end of the day, a really, really good technology team is usually supported by an equally good product organization. So you can’t just sort of transform the technology org without doing the same to the product work. And at Sorcero, we did it together. We transformed the product organization and the technology organization at the same time. And over a period of a year, started to both cycle folks out move folks around and and very quickly, I would say we started seeing the impact of that on our product and on our customers within three, four months of starting to make the shift. And I would say today it’s entirely transformed.
Walter Thompson 09:45
How long did it take you to get to your first 20 customers, roughly?
Dipanwita Das
I mean, now?
Walter Thompson
From launch to —
Dipanwita Das 09:55
I would say, like fiveish years. We didn’t have a product and market in that established, repeatable, scalable way for a while, because the market was also pulled between, Am I buying AI? Am I buying a product? Am I buying what?” So we also chose a market that wasn’t fully formed. And it’s a market that’s still changing and still evolving. So as our market and buyers evolve, they’re learning how to buy software. So we had that added challenge of selling software to a community that didn’t traditionally buy software. So we had to also get them on that journey of, how do you think about software? When you’ve had I don’t know, maybe reports before this in your email?
Walter Thompson 10:40
What’s your bar, I suppose, for a minimum viable product?
Dipanwita Das 10:44
Someone we know suggested we use the frame, “minimum lovable product.” And that was an interesting frame. I would say that the first product we put into our customers’ hands, really led them to start asking for more. It told us what else they needed, what else the product needed to have on all of the sides in order to be a true enterprise product, not just whatever they were interacting with. And what’s interesting about this roadmap is you can make a bunch of assumptions, but you have to be really careful in not going too far ahead. It has to be substantial enough for them to see value, but not so substantial that if something is off center, it’s too hard to correct. So I would say that our minimum, the first product we put into a customer’s hands today is a feature in our larger product. And it taught us what we needed to know about our customers, our users and their expectations. And once they get hooked on to that piece, they very quickly want the next seventeen.
Walter Thompson 12:04
You’re working in a new frontier of technology that has massively high expectations. How did you approach the TAM issue while you were in fundraising mode at the idea stage, seed stage?
Dipanwita Das 12:16
I would probably say My TAM calculation was a bit amateur in the early years. And you know, we were also looking at multiple markets at the same time, we had a more horizontal play. And I think this is pretty common to AI products, where you sort of take a problem, let’s say you talk about, I don’t know sales enablement, and you assume that sales enablement for industry, 1,2,3, and 4, and that’s your TAM. And that wasn’t our use case. But that was a similar approach. What we very quickly found out was, of course, that these industries operate in entirely different ways. And to fully take on an industry you need it to be vertical SaaS. So the first thing is that I will probably say your time is going to change and maybe become more refined. And so I look at Sorcero’s journey very much as a journey of refinement. And our TAM has changed to it’s still very big, but it’s much more realistic. There are just certain TAM numbers, which might seem exciting to an entrepreneur, but as I have learned means nothing to the investor, because it’s just frankly too big and too amorphous to make any sense and to plan around.
Walter Thompson 13:31
Who did you imagine you’re competing with when you first envision this?
Dipanwita Das 13:38
Well, initially, I thought about knowledge management, and software companies and consultancies, who would support a knowledge management as our key competitors. And it’s interesting, because knowledge management is also a huge umbrella. A lot of well — let me dial back, what I would really want to say is that knowledge management decisions support these philosophical terms, almost. And in some cases, they translate into very specific workflows and business problems with quantifiable ROIs. And in some cases, they describe and undo itself. And I think that when Sorcero started, we were very much in that world with knowledge management, and, you know, maybe asset or document management, which was very big and very old and staid and established. But that wasn’t really the business problem that sort of settled ended up or has ended up solving for. So that’s how I look at it is a change again, for a really truly de novo product, which is coming into a market that’s not bought software before. What you’re competing against is a very interesting set of things. And it took us probably three and a half years, like it wasn’t until 2021, I want to say, that we knew what we were competing against.
Walter Thompson 15:02
But that process of, I guess I want to say, decanting the idea and putting it on the table so we can all look at it and study it and turn it into something. What were your goals for the pre-seed round? As far as that idea, you’ve got trapped in a vacuum bottle, let’s say like, what does that process look like? What are your goals for taking that idea that was just floating around and turning that into, you know, R&D or papers or prototypes? What was the plan? And how did you build your way there?
Dipanwita Das 15:31
Sure. So let me maybe break it out into let’s start with the R&D and the product side of it. The first thing we knew is we needed to figure out the tech, right, so if we were going to apply transformer approaches to models, and then fine-tune them to work very accurately in an industry against a particular task, we needed to be able to do that. So we needed to have the mechanisms or the infrastructure to do that. So we needed to figure that, that was one. Number two, the other key part of our thesis was unifying, and automatically enriching unstructured data from any source. So we also had to make sure that that was doable. And we knew how to do it in a method that was repeatable and scaled. The next part was when it came to putting AI-powered features into a customer’s hands. And we’ve had customers using AI-powered features off of Sorcero’s many generations over the last four, four years now, maybe five, we have to make sure that we could build trust, and that the customers would have a way of interacting with the system and feedback loops. So there was a lot of work and just making sure that that infrastructure, the tech existed, all. And then after that is the user experience, the front end of it. So that was one big bucket of AI. Number two was discovering what GTM was going to be like for us, because the temptation is to say we can solve this problem for all industries, right. And we obviously said that most people do and that’s perfect. It’s not a pejorative, it’s a great way of discovery. So we had to take that big spread and turn it into very quick, very cheap little experiments, to see where the market actually lay. And very quickly found out that medical affairs and sort of this new commercialization model and pharma is where the real market was. But by then, we had indeed cycled through a number of markets that have been in like three years. So we had already spent that time and had valid use cases in different markets. But now the markets were deep enough. So that GTM piece was important associated with ROI and articulation of value of the product itself. And that I want to say continuously evolves, we have some customers who really want to buy the tech, right, they want to know what the AI does. And then other customers who just like, I have a business ROI I need you to like corresponding to. So we have both sides of the coin. So figuring out the messaging can feel a bit like voodoo and in moments, but when it hits, it really hits. And then of course, the last piece, which I’m sure we’re gonna spend some time on is the team, who are the people and the personalities that do this, but for the pre-seed investor. And I think I got very lucky in that we had some great pre-seed investors, both individuals and some funds, they knew what was going to change and what was going to stay the same. And at the end of the day, I would like to say they were investing in my team and I and the market opportunity very broadly, rather than a very specific problem. And in some cases, it can be very specific, but a lot of companies start fairly broad.
Walter Thompson 18:52
So what was the interval between pre-seed and seed?
Dipanwita Das
A year.
Walter Thompson
And seed to Series A?
Dipanwita Das 19:02
Another year.
Walter Thompson 19:04
And so can you share, what was the inciting incident, what was the validator for the Series A? Was it a technical achievement? Was it revenue? Was it customers?
Dipanwita Das 19:11
I think it was a combination of technical achievement and customers, it was about who the customers were, rather than just a revenue number. The revenue number, of course, was a signifier. But the fact that we were at that point signing on top-tier pharma was this definite signifier, because that shows a certain grit and maturity in a group of people. Because you can’t just say you’re going to take on top 30 pharma, but not actually have what you need in place to do it. So I would say it was a combination of that. We also have had a customer who’s been our customer going on, I think now for your five so also the ability to retain and expand and keep working with the same group of people whether you keep betting on you was enough. But one, so it was a combination of factors in it. And I think also the venture community and investors were seeing that life sciences was an extremely an easy and extremely rich market one of the richest, I would say in one of the highest margins, where software was hard to come up for their day to day operations, and there was gonna be a lot of opportunity to transform and truly digitally transform.
Walter Thompson 20:26
Are you thinking about launching an AI startup? Mayfield’s $250 million AI Start Seed Fund is actively searching for idea-stage entrepreneurs, who are working on the cognitive plumbing layer. That’s models, middleware and tools, data, infrastructure and semiconductors and systems. Mayfield has a long track record. Since its founding, the firm has been an early investor in more than 550 companies, which has led to 120 IPOs and over 225 mergers and acquisitions. If you have a fundable idea for an AI first startup, email, aistart@mayfield.com. Every business that incorporates AI has to account for data privacy, compliance and governance. That’s why Securiti developed its Data Command Center. Instead of setting up different tools that can eat up time, money and add complexity. Securiti’s award-winning technology automates compliance with global privacy regulations, includes a library with more than 1000 integrations across data systems, and identifies data risk to enable protection and control. To learn more, visit security.ai. So I think you said you talked to more than 300 people before you committed to writing any code for Sorcero? How did you know that, you know, 300 people was enough?
Dipanwita Das 21:43
I zeroed in on a number of roles. So, you know, obviously, any of our customers now have different tiers and roles and geographies, and geographies and other diversities that are represented in their organization. And since our vision is at some point in time, they are all going to touch our software, I wanted to make sure that those roles and people are represented. So I spoke with folks who were doing AI. And then I spoke with folks who are on the business side, and I spoke with folks who are writing and I spoke with folks on discovery. And so I wanted to make sure that all the aspects of the industry were represented, as well as similar adjacent aspects of payers and the providers are also represented. So it was more a question of making sure there were at least twentyish folks representing different roles and industries in the total pool for me to have enough. I shy away from the use of the word “data,” since it was quite qualitative, and a little bit of quant, but a lot of qualitative to say, “okay, I have a decent idea of what, you know, a first piece of software that they could use or would use would look like.” Also, it gave me a really good sense of the market’s maturity in terms of them being able to evaluate a piece of software and determine that it was right or not right for them.
Walter Thompson 23:08
So approximately how much time do you think you spent talking to each person, a range, a ballpark?
Dipanwita Das 23:16
30 to 45 minutes a pop? So it was a while?
Walter Thompson 23:21
And how did you collate all this data? What did you do? I mean, were you making your own spreadsheet?
Dipanwita Das 23:28
I wish I had Sorcero at that time. It was a little more manual, right? I’d read the notes, crunched the notes, I was really just disciplined about it. It’s a little bit like if you’re going to cook a big meal, and you don’t clean the kitchen right after, it’s probably going to sit like that for the rest of the week. So I tried to collate as we were going, and that was really helpful. And what I did was I’d take the recording, I’d use a transcriber on it. And then I’d sort of pick out the key points that I was looking for. And just if nothing else, dump it in an Excel sheet and organize. So I did it in the way that maybe our our customers do insights now, could do it better. But that’s what I had.
Walter Thompson 24:13
I mean, it got done. But how did you synthesize your takeaways and share them with your two co-founders? What was that process?
Dipanwita Das 24:20
I categorized it in sort of what is the one problem that everybody said they had if it was right in my mind, and I say this all the time that if I had built what they had asked for, it would be an enterprise search company because everybody starts by saying, “I can’t find anything.” And there’s truth to that. But then you have to kind of dig into the why. And the why was very interesting, because that’s really where I started to zero in on the roles and the functions within pharma that were most struggling with access to information because it really mattered to them, and there was a business ROI. So one was like the governing problem. The second part was what was the business problem? Then the next part was, you know, do they buy software? Do they have budget? What kind of like, you know, things like what is a number that they couldn’t pay or sign off on? And at what level without going through a full procurement process, which is really important when you’re trying to price a piece of software. Next was, you know, “how are they doing this today?” So the status quo also came out of there. And that’s where we learned so much about how our market operates today. And obviously, the problems with that. And so those were, I would say, roughly four categories of information. I tried to dig out a little bit more also on their approaches to AI. So for folks who I was talking to who are in the AI or data science team, I will try and understand how their companies are looking at AI, bringing it in, are they planning to build, buy, concerns, regulations.
Walter Thompson 25:58
I’m not sure if this is a useful question. And your response will tell me, I’ll just ask it. How did you determine whether Sorcero would be a horizontal or vertical product because that’s something that customers revealed, or was that a planned decision as far as the recipe you were making all along?
Dipanwita Das 26:14
It was a mix. So I believed that it would end up being vertical, but I wanted to test out which vertical it was going to play into, nonetheless, it’s just that these industries, particularly the regulated ones, are very particular and how they do things. And it’s very difficult for a company to be all things to all people. Otherwise, you’re going to have 24 use cases and 15 teams. And absolute pandemonium when I would say it was, necessity is the mother of all inventions, I couldn’t possibly have sort of 100 people doing little bits of things for 10 different verticals, it became very clear, also, which is the vertical that was going to see the maximum ROI. And therefore buy more and more of, some of the other verticals be tested, they would be very, they pay money for it, but it’d be very sort of adjacent to their core business. Life sciences is where we have an opportunity to actually become core to our customers. So that’s obviously the place we gravitated towards.
Walter Thompson 27:22
So I was talking to Navin Chaddha at Mayfield and he said that they’re specifically looking for half-baked ideas with this funny because you used the same phrase in our pre-interview, you said a half-baked idea in a familiar space can attract investment and interest. And I’m kind of wondering, where do you as an entrepreneur working in a very hot sector, where do you set the bar as far as a half-baked idea? Is just that something that’s easy to grasp and the team can back it up? Or how far out on a limb, can you go as far as “I think I can do this, and I want money to prove it.”
Dipanwita Das 27:57
I think that’s changing, right every day, I think a half-baked idea about anything with AI in it. A much lower bar would suffice, two years ago, 18 months ago, today, not so much. Because there have been a lot of I think wrapper companies, etc. And the real danger, of course, is that so many of these foundation AI companies are building the tooling around these models, right. So if you’re trying to look into that space, the real problem goes away. So I think that a half-baked idea and a problem that isn’t going to go away over the next two to three years because it is already in someone’s business interest to build it is worth investing in and making sure that the team has honestly enough customer obsession. Like if you’re really really, really obsessed with solving the business problem and don’t care at all, if you need to change how you solve it over the years. That’s been a good mark for us personally. But that could be worth it. It’s in nobody’s existing business interest to go solve that and thus, more likely than not, they’re going to in the next two to three years,is a good place to start.
Walter Thompson 29:22
So would it be fair to say, “I have a cohesive cognitive framework for solving this problem. Would you like to invest in it?”
Dipanwita Das 29:31
With some market proof to back it up that this is actually a problem, right? And you know, whether it’s from reports or people’s 10Ks or something that says that this is an actual problem, or even that I’ve gone and spoken to 70 people and here are some of those people and what they said could be a good place to start. But some of those people that I interviewed are now customers or will be, so that was actually a really nice validator, right? And a lot of them will tell me that, “oh, you built the thing you said you would,” which is always delightful to hear. But I think that if you can combine a clear cognitive framework with evidence of market or buyer interest, because someone’s gonna have to write you a check, that means that they have to think it’s a problem, not just that you have to think it’s a problem. And so if there’s enough people who think it’s a problem that is not you, and there’s some analysis, and you have a good framework, then yeah.
Walter Thompson 30:32
Was it tricky initially, trying to effectively communicate the value of Sorcero?
Dipanwita Das
Oh, my god.
Walter Thompson
Because you have customers who probably are not tech savvy, none of them. I mean, how could they be, they haven’t studied —
Dipanwita Das 30:48
Oh, it’s not a pejorative, they know things I will never know in my entire life. We all have different areas of expertise. I would say, oh, my god, and more, right, because I think that the trick is, AI is not quite software, yet. It’s still a thing unto itself until it becomes just accepted as software’s accepted, and no one really asks about what code base you’re using. If you’re just building a software product, we’ve kind of messaging two sides, you’re proving AI and you’re proving the business problem, you can do both together, and then the business problem has to be attractive. And then the AI also has to be attractive and has something has to have some special sauce. So no, it’s absolute agony. And you know what my customers now say? “Could you just talk to the business problem? Let’s not talk about AI,” unless they ask the question, but most of them do. They want to know they’re buying something that is also a bit special. But you start with the business problem. You know what it is, that’s half the battle won. I think with the investor community, it’s a mix. It’s a mix, there’s some people who were putting the AI first, and there’s some putting people who are putting the market in the business problem. So you have to navigate that.
Walter Thompson 32:00
Unlike other hype cycles, I’ve observed it seems though, in this one, academics and researchers really do seem to have a first-mover advantage. So what are some of the just candid, just the mercenary things that aspiring founders with academic backgrounds, what are the things they can do to leverage their credentials with investors today — without making their eyes glaze over?
Dipanwita Das 32:20
Hmm, find a practical lay person way of explaining how your thesis is going to show up in a product and what it’ll be able to do. So I’ll give you an example. Early on, we were very proud. And by early on, I mean, 2018. So you’re gonna have to cut me some slack on this, we were very proud of how fast we could process a certain number of words, that was pretty impressive. It made no difference to a customer. So the fact that we had a really impressive processor didn’t mean anything in terms of buying behavior, so making sure that you know, the thesis is linked to us being able to deliver something to the customer that they cannot and do not have and might never have without you. I think that’s the link to make. And we’ve gotten much better all around it over the years. But it’s a very rare personality I have found at the intersection of groundbreaking research, and what I would call a customer-facing production mindset. It’s a beautiful thing when it happens, but it’s rare.
Walter Thompson 33:32
Let’s turn to just dealing with investors. Everyone’s gonna have to do that. Unless you’re bootstrapping, actually, aside, is anyone bootstrapping an AI startup in 2024?
Dipanwita Das 33:42
I mean, if you’ve got a previous exit, perhaps potentially, I don’t know. But it will be hard, right? It just takes a longer cycle to get things into market and get them scalability, I assume that there are some that are, and there’ll be pretty great companies as well. Or if you’ve been able to use what is out there in the world, and not have to do too much R&D around it. I hope so, that kind of just shows that the AI itself is maturing, right, that you have enough tooling and info there that you don’t need to invest in considerable R&D.
Walter Thompson 34:16
That’s true. I just it’s just, it’s the old, you know, this old Silicon Valley trope of like, you know, two people in a garage, just sweating it out. I’m kind of like, I don’t know — compute costs money!
Dipanwita Das 34:25
It costs money! It costs money, even if you make a mistake. That’s even worse. So you’ve made a mistake and you spent a bunch you want to pay for it? Yes, absolutely. That costs money.
Walter Thompson 34:37
Would you have reservations about working with an investor who didn’t seem to have a strong technical grasp of what you were offering?
Dipanwita Das 34:45
Not so much anymore. I think in the earlier years, and we definitely have investors with a very strong AI background involved in Sorcero and they used their background to do some checking on our veracity, on our claims as well. And that was useful. But now I’m more interested in “do you know my market?” And can you insist, especially if you know, some kind of investor that would take a board seat? “Can you help me strategize around pricing, around GTM, around other things that will accelerate my growth?” So I would say it’s a changing shape, it would be lovely if that, you know, there was someone at the intersection of both somebody who understood how to position and speak about frontier tech or beat deck or you know, new tech, whatever you want to call AI, and also knows life sciences or that market very well, that’s sort of ideal state. But I think it changes from time to time, because at the end of the day, we’re not selling technology to engineers, we are selling a product.
Walter Thompson 35:50
So how did you work to a basically structure a team from the start, that could support the research and the business aspects?
Dipanwita Das 36:03
Painfully. The kind of business I had before and this one are extremely different. They’re both businesses, they both have a product or products, and they definitely sell to customers, but very different kinds of products, very different kinds of customers, very different kind of financing mechanism and very different kinds of revenue mechanism, right. All of it was different. So I would say that we did an excellent job of the researcher, hacker, prototyper, I think it took us several more cycles to figure out the right complexion on the product and commercial side. And this really is not a reflection of anyone on the team. This was also about articulating exactly what we were looking for. And I think that I have learned a lot, like I said painfully on what, you know what that persona is, that’s right for us and which role at what time, and also not assuming that most people are everyone has an entrepreneurial streak, which means they’re going to just go out and figure it out. That is not most people. And that’s maybe a vanishing rarity. In fact, when you’re looking at just the world of people broadly. So that’s maybe the biggest thing is that just the assumption that they will learn on the job, they’ll just figure it out. That was maybe where I hit the tallest wall or the thickest one.
Walter Thompson 37:34
I’ve heard people describe personas for early-stage teams in different ways, but it seems like the builder persona is like the most — you need a bunch of worker bees. Otherwise, you don’t have an operation.
Dipanwita Das 37:46
You need a builder persona with an incredible drive. And sort of a personal investment in mastery, where each of them wants to be really good at what they do, almost irrespective of where they work. So those same people will be excellent wherever you put them, because they make it their business to be excellent. Because you can’t. Like that’s the thing, because you’re going to want to figure it out. And you’re going to really have to want to figure it out.
Walter Thompson 38:17
So those people, they don’t raise their hands and say, “here I am.” How do you sniff them out in the room? How do you identify them? What do they look like?
Dipanwita Das 38:23
I think people who’ve been entrepreneurs often fall in that category. Like if you founded something and you’ve gotten it to a certain stage, you have to have had that because there are loads of things, you’re going to find out that were not quite right in the way that you thought about it. So I think both Walter and Richard represented very different sides of that entrepreneurial coin. And that was hugely helpful, because they will go figure [things] out, even if they’re encountering something for the first time, I think that’s also true of my CPO. On my CTO, now, they’ve all been founders and founding team and taking something from zero to one. And I think the other part is, how long is zero to one going to last? Zero to one often lasts a lot longer than one takes. And the temptation is to say it’s now a system and it’s often as a modular system as one thing. So I think that zero-to-one people also know other zero-to-one people. And as we matured our hiring process, we would change the questions. We asked, “have you built something, how big was your team?” Were their processes defined, or did you define them? “Did you have to deal with a blank-page problem?” And a lot of people have come from a lot of incredible brands and never faced a blank page. So when they have a blank page, they really don’t know what to do with it. And it’s not a pejorative,
Walter Thompson 39:45
I was going to ask, no pressure, but what are some of the attachments you had about hiring and recruitment that you don’t have anymore, that you got rid of after going through the process several times?
Dipanwita Das 39:55
Brands. Because I found that unless you’re recruiting someone from like the first 50 of that brand or first 100 of that brand, which is a rare commodity for some of the better-known brands, you’re not getting the builders. So even though they might know a lot about how good looks, they don’t know how to get to good or how to get to great. So that was one. I think, number two for me was people who were coming in with a playbook that they were very sure of. And they just wanted to work that, you know, kind of impose the playbook. But we’re not very good at sort of teasing out what the right is, you know, temperature for this group goals. I think number three is really asking questions around learning and curiosity and what and how many times in their career? Have they been the first? Like, how many times they’ve been thrown in as the first like a first-time manager? Or when were you the first-time manager, did you set up the process? Or did you follow the process? And more of those questions? I think the last one is not not not making that assumption that everybody is going to be super, super driven to figure it out. That piece is a little esoteric, but is really important. I asked this question about “what is anathema to you?” Whenever I ask in an interview, I find that very revealing.
Walter Thompson 41:22
Have you ever had a response that made you just not want to hire a person, a response where you were just like, “this is not a fit?”
Dipanwita Das 41:29
Absolutely. And I think it’s great, because there’s also mutual, because at the end of the day, I think most interviews, hopefully are mutual. And it’s not just about hoping to get a job, it’s also about investing and making sure that you can make that person successful. If I keep hiring people that are not successful, that means there’s something off with my hiring process. It’s not just, you know, they’re bad. So I think that yes: what’s anathema to me or them or to other people in the company, I think it can be very revealing, and it checks out mutual fit.
Walter Thompson 42:08
You mentioned earlier that you discovered that the journey between an idea and building a sustainable business was a lot longer than you expected. So I guess my question is, do you have any specific advice for idea-stage AI founders who are trying to raise funds right now in early 2024?
Dipanwita Das 42:25
I think one is maybe don’t over-index on your first five customers, they could all change and you might want to fire them. Because that space is evolving really, really fast. And there’s a lot of what I would call component companies emerging who do very well at a task, like, “I do summarization, I do writing, I do, I do.” But a lot of this is getting eaten up by workflow and Office and other tooling, and kind of getting baked in. So it’s going to be tougher as it matures. So like the early spike, it might not last. So I think that might be an interesting area to look at. Also monetizing AI sustainably, scalably, repeatedly, does take time, if you’re selling into verticals that are moneyed, and you’ve gotten in, you could get a very high revenue off of like two or three customers that is irreplicable into a larger market, which is both like really delightful when you get it first, but then also really misleading. In terms of whether there’s going to be other people, I think about that. And then like number three, making sure that you have some IP, it’s less about the patent, that something that you do to make the thing work. Because unless you’re building a foundational AI company, there’s something that you got to do that makes it work, that is you and you and you could be a data play, but might not be. But either way, there’s something you do to it to make it work that sets you apart. And workflow if you can, that seems to be the sticky bit.
Walter Thompson 43:58
And last question for you: If you were up for a job with an early-stage AI startup, what’s one question you’d be sure to ask the founders during your interview?
Dipanwita Das 44:08
I would actually ask them a lot about their bio, their business problem, the size of the problem, how are they doing it? Now I’d really unpack that. And then either unpack that in the context of the solution that they have created or are creating, like that will really and then I’m going to probably check how open they are to input. If there is appropriate input.
Walter Thompson 44:36
Would you ask about runway? Or is that a secondary consideration for you?
Dipanwita Das 44:40
It’s almost a secondary consideration for me, because it doesn’t matter in some ways, right? You could realize very fast you could realize very slowly and I think with the kind of employment movements we’re seeing in the market right now runway is obviously not the problem with many of the companies that are conducting mass layoffs. But to me, I think it’s most important to understand that there’s a real need for the role that they’re trying to fill. And that is it’s going to be something that is deep and well tasked and really needed, and they know where it fits. So I think those are maybe the two buckets, right? I want to make sure that the role is the right one, at the right time, properly described, quantified, like they’ve thought through that. And the other part is they’ve thought through how they’re going to make money off of that software in very tangible ways. Even if they’re in their first sale, like I don’t care if they’re there for a sale. But those are the two areas maybe the
Walter Thompson 45:41
Dipanwita, thanks very much for a fantastic conversation. I’ve really enjoyed the time today. I appreciate it.
Dipanwita Das 45:45
Thank you, Walter. I hope your listeners enjoy it as well.
Walter Thompson 45:52
I’ll be right back with some show notes after a word from our sponsors. Fund/Build/Scale is sponsored by Mayfield. If you have a fundable idea for an AI-first startup at the cognitive plumbing layer, email aistart@mayfield.com. The podcast is also sponsored by Securiti, pioneer of the Data Command Center, a centralized platform that enables the safe use of data and Gen AI. To learn more, visit securiti.ai. Thanks again to Dipanwita Das of Sorcero. Coming up next, I’m talking to Poshmark CEO Manish Chandra and DevRevCEO Dheeraj Pandey to uncover the frameworks they’ve used to build sustainable startups. We’ll get into a lot of topics, including how they built and scaled early teams, developing your value proposition, early marketing and branding, acquiring initial customers, managing fundraising and balancing investor relationships, pivoting, and also personal development. One of my favorite words in the English language is “penultimate.” This is episode nine, which means the next episode of Fund/Build/Scale will conclude season one. I want to say thank you personally to everyone who’s supported this project since I was at the idea stage. I really appreciate it. If you’ve listened this far, I hope you got something out of the conversation. Subscribe to Fund/Build/Scale so you’ll automatically get future episodes and consider leaving a review. Follow Fund/Build/Scale on LinkedIn and YouTube. For now you can find the FBS newsletter on substack. The show’s theme was written and performed by Michael Tritter and Carlos Chairez. Michael also edited the podcast and provided additional music. Thanks for listening!
