Transcript
Today I'm very happy to have Andrew Roberts to our podcast. He's a serial entrepreneur. He built and exited Tiny MC, uh which was a $10 million ARR business. And they had like big customers like WordPress, Atlassian, Salesforce. How do you go from building and selling a company and sort of starting again? I would say a vast majority of founders go back to being a founder.
It's not easy. Often you haven't enjoyed the journey at times. You know, people say it's like chewing glass at times or assembling an airplane on the way down. I mean, it's very, very challenging, but you reflect on your own skills and and and what you're able to do and make a difference in the world. Let's talk about your current company, which is called Deliverables AI. And you're basically an AI company that automates the creation of IMs, information memorandums as they're called, and complex PowerPoint decks.
How did you get into this? A friend of mine who was looking to sell his company to an investment banker involved in my last company, and we worked together on looking for an exit for him. And as part of that, I helped build some of the materials, got exposure to the process. Also, at Tiny we acquired two companies. I sold my company. So I had a little bit of exposure to mergers and acquisitions.
It's a huge mysterious process, very laborious. I remember one of my clients, long, long back when I was in consulting, was Rothschild, and they used to create like some amazing IMs and all that. How close are we to like those kind of decks? Based technology, so the foundation models are absolutely there. Orchestrating all of that is pretty massive headache, and you know, every firm has their own way of doing it. From Deliverables AI perspective, some of your customers have reached that stage where they are completely like building all of their decks and IMs on your platform.
Yes, for our bigger customers we'll take their deck and build it into the system. They're getting 80, 90% of the way there. Obviously, there's still some craft in taking that output, that first really good first draft, and then polishing it before you send it to the client. But that might be like used to take them weeks to get the first draft together. Now they can get the first draft together in, you know, 30 minutes. AI doing all the work, they can go off and get a coffee, and then, you know, they can put an hour or two into the polish side of things versus versus the weeks it took previously.
So now it's absolutely there today with our platform. Yeah. Usually every startup or every company has some very unique insights and either about customer behavior or about technology. What are those some of those insights that you have about this business? Well, you can unpack a PowerPoint into into four XML files. And I'm kidding you not, like one PowerPoint slide I saw the other day from a client was like 14,000 lines code of XML.
So I guess our unique side inside is like this is a complicated, difficult problem. And as an entrepreneur, we should get attracted to complicated, difficult problems because if you solve them, you might be bringing something unique to the table. For somebody who is like very young and like a 22-year-old who has joined an investment bank and, you know, is probably listening to this podcast, what's the what's the single thing that you would like to tell them today that they should start doing to sort of, you know, learn about AI or catch up with what's happening with AI adoption in financial services or investment banks? Hello and welcome everyone to yet another episode of Build AI podcast. Today I'm very happy to have Andrew Roberts to our podcast. He's a serial entrepreneur.
He built and exited Tiny MC, which was a $10 million ARR business, and they had like big customers like WordPress, Atlassian, Salesforce. And, you know, as I I really truly appreciate second-time founders, builders in AI space. And so it's lovely to have you, Andrew. Thanks for joining us. Yeah, thanks for having me. I very much appreciate it.
So looking forward to to chatting all things AI. Yeah. I think I want to understand the arc. I think anybody who's listening to this podcast, which is a lot of entrepreneurs, especially early stage entrepreneurs, right? So, one of the things that we do in this podcast is everybody talks to and brings those, you know, heavyweights like Chamath and, you know, Elon Musk and so on. But, what we have seen is that early stage founders really look for advice from early stage founders because the problems are something that they can relate to, challenges, opportunities.
And so, it it might be very interesting to understand from you that how do you go from like, you know, building and selling a company and, you know, sort of starting again and and then, especially in the AI space, you know, at a different stage of your career. Can you explain the journey? Like, how did you take that decision? How did you choose this space? And and so on. So, it is an interesting dilemma.
First of all, probably my guess is is post-exit founders and and what they do after they after they exit. So, you know, particularly when you're, you know, not of retirement age or whatever you like, you know, what do you want to do next? And I I think everybody sort of wonders through the wilderness a little bit there. I'm part of a group called post-exit founders. You know, what do you want to do next? Do you want to do executive coaching?
Do you want to, I don't know, quit working and study physics or whatever it is that you want to go do. I would say a vast majority of founders go back to being a founder. It's not easy. Often, you haven't enjoyed the journey at times. You know, people say it's like chewing glass sometimes or assembling an airplane on the way down. I mean, it's very, very challenging, but you reflect on your own skills and and and what you're able to do and make a difference in the world and you come back again to, you know, I guess I I've got founder DNA and and that's what you want to go do.
You're a hammer looking for a nail and starting businesses is is the nail, I guess. And I can't, you know, over half of founders go back and do it again. That that's anecdotally. I don't know the the true statistics, but just in terms of people I know I know they inevitably reach that conclusion reluctantly sometimes and then they're like, all right, okay, I want to go do a startup and then they start the process of brainstorming, maybe looking for co-founders, thinking about it. I happen to exit at the end of 2022, which was phenomenal timing for, you know, getting getting into the AI revolution. Done some AI over my career, studied a little bit at university, built some AI integrations with our products with IBM Watson, would you believe?
Back in 2010, 2011, you know, come 22, ChatGPT had just taken off. It was it was great timing. I mean, like everybody on this podcast, I'm sure it was just this mind-blowing experience of like, oh my gosh, like this technology is amazing. I want to do something with it. Yeah. So, it was a matter of then thinking through, you know, what what is how do you compete?
How do you how do you craft a unique story when every single other entrepreneur also had that same epiphany. It it's fascinating that people like you and me, by the way, I exited my company in March 2021 and then started again, right? So, this is my third venture. And here we are actually creating multiple companies. It's fascinating that all of us have gone through all the pain and suffering and all the challenges which come, you know, during building a startup and and yet we are here again, right? So, it's almost like soldiers going to war.
Yeah, exactly. I mean, it doesn't make any sense. Really, we're going to screw this up or something, but here we are, we do it again and, you know, obviously there's many parts of being a founder that are really enjoyable and that's why we do it, but but yeah, you're right. If you asked a first-time founder in the thick of it, if they achieved an exit, would they go back again? I'm sure many of them would say never, but inevitably they do. Yeah.
Yeah. So, let's talk about your current company, which is called Deliverables AI and you're basically an AI company that automates the creation of IMs, information memorandums as they are called, and complex PowerPoint decks. That's the work that I was in. I was like I started my early career strategy at a automaker and then management consulting firm where we did some of this work. But how did you get into this? So first fixer again, you know, thinking about what I wanted to do next.
I introduced a friend of mine who was looking to sell his company to an investment banker who was involved in my last company and we worked together on looking for an exit for him. And as part of that I helped build some of the materials, got exposure to the process. I've also, you know, under at Tiny we acquired two companies. I sold my company. So I'd had a little bit of exposure to mergers and acquisitions. It's a huge mysterious process, very laborious.
And I love the content creation space. So as I was just engaging with with an opportunity kind of kind of engagement, I was always like, all right, what are the problems here that AI might be good at good at solving? And that was when I So I actually used AI to help them build a deck. This is back in '23, '22. And you know, the investment banker was like, "Wow, you've really got a way with words." And I was like, "I got to make a confession. I used Word to help me here." And that's how, you know, how the game the deck came together.
You know, obviously there was a lot of manual copy pasting and doing stuff back then, but but in general, yeah, that's how I landed on it and it is a a very time-consuming difficult process. You put it together essentially a extremely detailed pitch deck. Uh for those who aren't familiar, investment memorandums. For whatever reason, this industry loves to cram like four slides on one slide. So, you know, very dense materials and yeah, very difficult for the bankers to put together. So, you know, I you know, honestly for us working in this it's probably taken us 18 months to get to a solution where bankers are like, "Oh, okay, this is this looks like what I would expect and, you know, what I something we would actually use." Right.
So, is it fair to say that most of your customers are investment bankers or are there other customers in, let's say, financial services or other spaces? Our customer split into three buckets. One is mergers and acquisitions advisors. So, you know, firms that have anywhere from like 10 people to a few hundred who want to, you know, want to embrace AI and want to use AI as part of their workflow. And then business brokers, which would tend to be smaller. They're often, you know, solo people.
Uh so, we have quite a few of those. And then finally, some others in financial services. So, private equity, commercial real estate, those sorts of folks. Okay, cool. And I I think last time we were chatting, I picked up that you have about 40 customers and growing at about 9 to 10 customers per month. So, that's really good.
I think for uh you know, some of the people who are listening to this, early-stage founders, it might be very interesting to know that how did you acquire the first customer, second customer, 10th customer? Just a little bit of like what was the process and what kind of channels and GTM strategy you had? Sure. You know, I I I think it's important to note that we kind of bubbled along in terms of revenue for quite some time there. You know, we were acquiring some customers, but quite honestly, we were losing them as quickly as we were getting them. So, it's only been really since October, November of last year that we've seen the product's good enough and and we're retaining customers and bringing them in, convincing them at a at a very fast clip.
So, does take a while to find product-market fit. In the early days, well, you know, our first customer was our first angel investor who was the banker I told you about originally. And that purchase from TechStrata, which is the name of his firm. So, he was our first customer. And then a combination of Google AdWords certainly brought in a few tire kickers who wanted to talk to us and continues to be quite quite good for us. And then cold outreach via email and LinkedIn.
And then some face-to-face networking as well through industry organizations. So, uh they were the really three buckets there. Or, you know, asking for a referral from people you knew, that sort of thing. Yeah, that's really I I I think one of the things that I'm when I'm talking to a lot of B2B AI company founders, that you know, now that AI can create emails and send emails, they're like the email channel is dead. Gazillion emails being sent. And then the same thing is happening with, you know, calls and bunch of other channels and even one of our portfolio companies founder was telling me that conferences and going and meeting people in person is actually becoming a very, very good channel again.
In in in the AI space. Yeah, you know, we pulled back from it. It's hard to scale, right? Like particularly when you're building the product and, you know, in the SaaS first world, you're trying to look for things to scale. So, we'll get back to it, but you know, I think in the early days you're actually hunting for product market fit, you're not necessarily hunting for customers. So, you're really wanting to have good relationships with people that you can get on the phone who will tell you if your kids are ugly, those sorts of things, which is hard to get from, you know, just inbound prospects.
They just want to be sold to, so you're kind of after a little bit more of a relationship so that you can get deeper insights into the problems that that your target customers are facing, whether or not you're able to convert them to revenue in the early days. Yeah. Would it be fair to say that bunch of your initial customers you actually met them in person and and then build the relationship or was there a lot of remote customers you never met? A lot of a mixture, a mixture. You know, actually cold email was really how we got our first 10 customers, right? So, it was, you know, some cold email campaigns we sent out.
Admittedly, you're right, like this was 2024, here we are 2026, you know, or I should say these initial discussions we were having to try and learn more about the market. So, you know, you're right, it's probably a lot more saturated today than it's ever been. You know, we still use it as a channel, but but the yield is can be very, very low and for a very small target segment. Now, there's probably 20,000 names we could pull in the United States for people we want to go after. It's a fairly small pool, pool, so how many times can you hit those folks up anyway? Yeah, yeah, I totally hear you.
Let's talk a little bit about the product and engineering side of things. You know, I know I know you're super technical yourself, so, you know, like this this kind of a product where you're dealing with like very very you know, a lot of research and very lengthy documents. I am thinking there there's a lot of tooling, there's a lot of nodes, there's a lot of stuff around orchestration layers. Can you Can you actually explain like how you went about building the product and how did you think about the architecture because the AI space is changing so fast. Like I I think 1 year back you are not talking about skills and skill files and agents, but now everybody is talking about those. Yeah, yeah.
Well, it's changed a lot. You know, we're on kind of a version three in production now and already thinking about, you know, version four of of you know, how we might build things. So, today well, it's a LangGraph agent underneath the hood and, you utilizes I guess a combination of like deterministic steps within that. Like produces JSON, go to the step, go to that step. But with React agents, which is, you know, probably more where you where you think things are going, you know, just a a long long reasoning agent, orchestrator agent that that kind of does the whole workflow for you. So, so yeah, we're we look at that like particularly coding agent like getting powered by you know, these coding agents that are now powering the agent SDKs from from Anthropic and from OpenAI, quite good harnesses.
So, you know, you could even potentially, you know, as you said, have a skill to to run the workflow instead of like a like a graph from from LangGraph. We We certainly look at that, but yeah, today the thing in production is is a little bit more deterministic than that if that makes sense. Okay. Hey, so when you say deterministic, I believe you are You know, there's there's a lot of customers that we talk to and they say that, "Okay, you know, maybe for demos, you know, building something with you know, in in Claude with skills and agents is fine, but if you have to run like 10 million transactions a month, not sure if that's the great idea." And so, they want like something more deterministic, sometimes just like create something in AI and then convert it into code. Like as simple as that. So, I I don't know when you say deterministic, what exactly are you referring to?
Yeah, I guess it's a balance between letting the LLM do stuff and writing code that, you know, runs the same every time, right? You know, so for example, you could totally tell the LLM, I want you to plan and then I want you to, you know, based upon that plan, fan out and do, you know, parallel slide generation and then I want you to review step. Or in code, you can say, build a plan, here's a JSON file, now I'm going to spin up agents to go execute the slide generation and then we're going to have a review phase. And that's in code. It's not it's not a skill file, right? It's it's, you know, the Python file that that tells tells it what it's going to do.
So, look, I don't know this terribly. At this point, there's probably not a lot of difference. Like, I think it these long-running agents, if you give them a skill file, will probably follow the workflow 99% of the time. I think it's more about how do you ensure that however you do it, you're getting the quality that you want. And I think, you know, we've certainly lent into more structured evals and so forth. So, so we can get a real sense for how an agent is performing or your multi-agent system to use the use the lingo, right?
I mean, there's often many agents going on at the same time. But however you do it, whether it's like a skill file that prompts it all, one skill file, or if it's like 50 skill files, or if it's like a langgraph graph, either way, you kind of want to assure yourself like, are we doing better than what someone could get from typing in the code? And you that that bar is always getting higher as well. Got it. Yeah, I was going to ask you like every AI founder today is being asked whether you are a small or mid-size or large company like yourself. You know, Perplexity just got bought by XAI or at least the news is like that.
And so, they're being asked like, but Claude can do it. It's almost reminiscent of like, you know, like 5 or 10 years back when when you had a a company and you would go to a VC and they would say, "Oh, but why can't Google do it?" What's What's your answer, Andrew? Like I would love to know what's your answer. Well, it's a good question, right? Because these behemoths we could call it, you know, you fly too close to the sun or something, right? Like you you know, the foundation models are only going to get better from here.
So, you know, what will they be able to tackle? I do think that there is some use cases where there's pretty much unlimited demand for quality, right? Like if you can you know, if the base model can do an 80% job and you can do a 95% job, that's really material. Or if they can do a 90% job and you can do a 99% job. Frankly, even if they could do a 95% job and you can do a 99% job, it's still you're still going to have buyers leaning in going because they care, right? Like quality is important.
So, again, I do think it's really important to baseline that. I think you can potentially help clients. So, that's that's one. Definitely, the performance of your we call it autonomous task performance, right? Like user types in, one in comes out the other. You kind of have to prove that it's at least as good as what they're going to get in quality.
Ideally, you know, 5, 10, 20% better. The second part is like roll out of this is very, very low. I talked to a friend of mine who works for private equity firm. They're probably in the top 0.1% in terms of AI adoption, right? Like a it's a very AI forward company. They invest in AI.
99% of the firms are not doing that what their firm is doing yet. So, there's this huge roll out in the economy that I think, you know, people talk about services as a software. Hey, it's just hand holding or getting on a phone, customizing it for that person's industry, speaking their language, understanding their pain points. There is still a huge role for companies, software companies or services companies, however you want to think about it, bridging that gap of understanding about how to deploy these technologies. Vertical AI is part of it, right? Clearly, you know, you can get a financial services plug-in from Mantropic and suddenly Antropic, you know, codes a little bit more better for financial services.
But I would say you need to go broader and deeper on all workflows in your in your ecosystem in terms of integrations, in terms of the skill files that you develop. Again, hopefully evaluating them against the baseline that's available elsewhere. And then so vertical, bridging the skills gap that's out there in the economy, making sure that your agent does perform better than baseline. And And then I would say firm customization is the other part. And I guess it's part of maybe, you know, getting this into people's hands is they're like, "Yeah, great. I can sit it forward and co-work and ask it to create a PowerPoint for me, but it looks like nothing like these very important templates we've been using for 10 15 years.
It just it's missed, right? So, how do you tailor your agent or anybody's agent, frankly, to a firm, right? So, I think there's there's sort of four things that can come to the table in terms of in terms of a mode. So, it's challenging, right? I mean, at the end of the day, these large models can do a lot in the right hands. And And you know, maybe for the super early adopters who can kind of figure it all out on their own and spin up their open claw and do this and do that.
Hey, cool. Like, you know, maybe you don't need our stuff. But for everybody else, that's where you've got the opportunity to bridge the gap. Right. Yep. And And we see the same thing.
Like I think there's a lot about domain expertise, integrations, like especially as you go into financial services, the integrations itself. And then, you know, there are there's so many things around just like bringing a lot of context and data flywheel within these organizations. There's There's a lot that, you know, like if you take these niche use cases, it will take a while for anybody to sort of come and and sort of automate those. Although, the bigger question is that how do you see like, you know, how will the companies, you know, start like doing a bunch of this stuff in-house? Like I'm I'm not just talking about your use case, but in general, do you feel like given that now it's so much easier to build and ship. And you know, like lot of people who are white coding feel like they can they can just solve every problem in the world when you're doing white boarding.
And then when you actually start building production software, you realize, holy there's so many things that we don't know, right? So, I just wonder like there's a lot of excitement in enterprises also that like I was reading somebody's post that literally there's a KPI right now for the technology team and especially the CCTO and CIO that how much of stuff can you do in house, which shows that how much AI native you are. And and so do you like in your opinion, do you see that happening like lot of organizations might start like doing way more stuff in house as opposed to third-party vendors? So, yeah, I mean, I just just a friend I was talking to yesterday, they do everything in house. But, you know, they've got a team of eight to 10 engineers. They have recruited very good engineers.
He said the total loaded cost of these folks is 750,000 to a million dollars. They've got about like one engineer per product, which is the whole vibe coding, you know, AI-empowered engineer can do a lot, right? So, they're doing amazing things in house, but that's going to cost that costs a lot of money. So, you need to be in a certain bracket of company to be able to be doing that. And for everybody else, like they can't afford that. Same dynamics as always, right?
I guess is, you know, the top end of town might be able to afford to do this themselves, but, you know, down from there I think down from there people can get a lot done. And, you know, even inside this firm, you know, they had financial professionals like creating skills and so forth. So, you know, the AI-empowered employee can bring a lot to the table in terms of automating a workflow or, you know, getting a getting some sort of agent set up. They kind of probably they're probably not like running strict evals, and you know, they're not doing it like next level kind of stuff, but they're you know, hey, it's I don't know, like I build stuff for myself, right? Like I've got a code code skill that will look in HubSpot, look in Amplitude, look in all of our things to try and assess like why a customer churned. You know, it'll run for 10-15 minutes and it'll tell produce a beautiful report about like why the customer churned.
You know, even when I had 80 employees we didn't do that. So, is it is it the perfect churn report? Probably not, but it gets the job done, right? So, I think lots of lots of sort of workflows will be handled that way as well. Yeah, I actually get very excited about that part because we serve a lot of small to medium businesses in in all the vertical AI companies that we are building and I feel like this is a great sort of technology leveler, AI, because now a small to medium business can have almost the same kind of dashboards and applications that a large enterprise is having if you are smart about it. So, there's a lot of democratization of technology which is happening all the way to consumer like individuals as well.
Yeah, and you know, look, even as I said with 80 people or you know, now the parent company that I'm in is the tiny as rolled into is probably a few hundred people. They don't do this level of analysis on churn, I can guarantee it, right? Like they just don't have the manpower to do it. It's just not worth it for them to do it. So, to have AI go off and do it is is hugely advantageous and I think you know, yes, big companies can afford to do some of this stuff, but you'd be surprised how much they they can't either cuz it's just too hard or just the ROI is not there. Either way AI can can help tackle those use cases.
Right. I I also want to know like slightly more granular things from you. Maybe we can, you know, go like very objective and you know, take a short answer source on on this one because you know, I want to cover as much as possible in the remaining time. So, I know that you basically had like 80 people in your last organization which you sold and now I'm guessing like you are like couple of people, right? Like founders and maybe like one or two more people. So, this change in organizational design and how do you think about hiring?
Like any thoughts on that? Well, anybody who's managed a lot of people now it comes with its its share of headaches, right? So, I kind of deferred as long as possible and then also understand like what sort of people we would want to bring in in this new environment. So, I think you know, AI comfort and the ability to to put together workflows on their own and all those sorts of things just super super important. So, yeah, aside from that just trying to defer hiring as much as possible, automate as much as possible and you know, build build little workflows to help you. You know, it's it's it's just fun and interesting to be able to do that.
Eventually though, of course, you know, if you want to scale a business, you need more than yourself to do it. You know, we'll we'll we'll cross that bridge. When we do, it will be a matter of like AI first people who just really get it. So, you know, I know there's a trend for using AI to interview people. Yeah. And you know, I don't know how people feel about that, but one really interesting point I was talking about the other day was that that helps screen people who are uncomfortable with AI, right?
Like if they're like, "Oh, I don't want to talk to this thing." or "This is stupid." or they get angry during the interview and you see that during the transcript, you're like, "Yeah, all right. You're not You're not You're not going to not going to work out here." Obviously, you need to interview them for human fit as well, but I thought that was an interesting reason to have AI interview some of your folks, yeah. Yeah. There's so many things that early-stage entrepreneurs can actually do with AI. You don't need like a finance guy or a chief of staff or so many other roles, right? You can create invoices as contracts come in and you know, based on the timelines and conditions, so AI can automatically create invoices, send it to customers, follow up.
There's so much that you can do today because of which you can have very lean and mean organizations, which is what like Yeah, I think startups with limited resources want to do and I think today it is possible to do. But let me quickly switch to the next topic. So, usually every startup or every company has some like, you know, sort of very unique insights and either about customer behavior or about technology. I I wanted to ask you what are those some of those insights that you have about this business. Well, I mean I think you know we've we focused on this automation of of decks. I mean I don't think people appreciate until you really get into it how intricate and difficult it is to to to automate that.
And you know, I came from you know, Tiny MC was a in the web content management space for like building websites and so forth. So you could think about a complicated website, how many lines of code it is, those sorts of things. Well, you can unpack a PowerPoint into into its raw XML file. And I'm kidding you not, like one PowerPoint slide I saw the other day from a client was like 14,000 lines of code of XML. So I guess our unique side inside is like this is a complicated difficult problem. And as an entrepreneur, we should get attracted to complicated difficult problems because if you solve them, you might be bringing something unique to the top table.
Yeah. And especially creating these kind of documents, there's a lot of other stuff as well, like citations, right? So how do you do citations tracking and all that? And I'm guessing like there's a lot of moving parts in this which which you have kind of, you know, solved. Yeah, absolutely. Citations, silly little things like logo grids, like that's a really important part of these sorts of decks, you know, just the illustration style and in terms of the images you want to place that fit this sort of document.
All those sorts of things need to be thought through and to to get it to get it right. And at the end of the day, the bar for us isn't like fancy banana banana slides, it's like this looks like something I would have made. As as a seasoned investment professional, this is something I would have made. Not is this beautiful blah blah blah. That's very much secondary to like is this some is AI helping me get my job done versus AI creating something that, you know, looks snazzy. Yeah.
How close like I'm sure your output is really good, that's why you have like 40 customers. How close are we to like, you know, like I remember one of my clients long long back when I was in consulting was Rothschild and they used to create like some amazing, you know, I am still that. How close are we to like those kind of you know, those kind of decks? Base technology, so the foundation models are absolutely there. Orchestrating all of that is pretty massive headache, and you know, every firm has their own way of doing it. So, I think the answer is the technology is capable of it, but implementing it implementing it can take is a lift, very much a lift still.
You know, I guess it's up to vendors like us as well as, you know, progress in, you know, the you know, agents that might be able to take a brief like that and do it from end to end. Are the foundation models if you like that the foundation models aren't there to run the whole process, but in terms of, you know, doing the individual component parts, they are. So, if you're able to stitch it together, you can you can get a good result. But again, quality is often in the eye of the beholder for each firm, they have a different way of doing things. So, have you like I I was asking more from deliverables AI perspective that some of your customers have reached that stage where they are completely like building all of their decks on and I am still on on your platform. Yes, so we actually are able to for for our bigger customers will take their deck and build it into the system.
So, they're literally getting their deck back populated by AI based upon, you know, the source source material they have gathered. So, yes, they're getting 80 90% of the way there. Obviously, there's still some craft in taking that output, that first the really good first draft, and then polishing it before you send it to the client. But that might be like used to take them weeks to get the first draft together. Now, they can get the first draft together in, you know, 30 minutes. AI doing all the work, they can go off and get a coffee, and then, you know, they can put an hour or two into the polish side of things versus versus the weeks it took previously.
So, now, it's absolutely there today with our platform, yeah. Yeah. And and couple of final questions like one, how are you pricing this? And as AI pricing is evolving and changing every week, what's the right model of like sort of, you know, charging for these kind of platforms? Look, we go back and forth on this. You know, the model we have today and I think works quite well is subscription plus credits.
So, we have, you know, a monthly subscription that has comes bundled with some tokens or credits. But, it's use it or lose it, right? So, if they don't use it to the max every month, we kind of get to pocket the difference in terms of that. So, that helps with gross margin. If they're using it every month, that's fine. No problem.
That's going to hurt our gross margin. So, that's a little trade-off. But, as long as you if you've got a sort of flat subscription fee, there's a little buffer built in there. And then your credit pricing could also, you know, have a tiny margin on top of, you know, the tokens that or other costs, frankly. I mean, it's not just tokens, right? That that you have on your cogs.
So, you can manage the gross margin there. Okay is is my feeling, right? So, you can hopefully be shooting for 60% plus gross margins. I think that's very doable for AI SaaS. As long as you're bringing value in that subscription above and beyond the AI, right? Like, people clearly see your harness, your app, whatever isn't purely tokens, they'll be happy paying the subscription.
And then as part of that subscription, you can bundle the AI. And then you obviously they can pay more. So, it's also a great expansion vector. I mean, you always in SaaS pricing want to make sure that you as you get larger clients, your price goes up. And I think, you know, that's where the credit model is great. If they're using your platform really heavily, you know, they should they should pay more and they'll be happy paying more.
If they're not happy with your platform, they'll go somewhere else and you know, they don't have to pay you. So, there's really that that usage base on as a way to scale up for larger clients that works as well. So, I think a hybrid is is the answer right now. So, that you can afford to keep the lights on, hopefully have some sort of gross margin to cover, you know, what what you're doing here. I don't know. I personally So, I bootstrapped my last business.
So, this idea of having, you know, raising hundreds of millions of dollars and negative gross margins and stuff like that isn't particularly in my DNA. So, and I'm sure most people in the AI SAS world are in a similar boat, right? Like they just they can't or won't raise, you know, tons of money to fund negative gross margin. Yeah. Yeah, we are definitely in the same boat. But you you have raised some sort of seed money pre-seed or seed money as well.
Yeah, that's right. So, we've got a a great VC, Firsthand VC, bunch of ex-Salesforce folks, Simon Chan over there. So, yeah, we were very fortunate to bring him on board in our pre-seed. We're still out there and going to be finishing off our pre-seed round in the next next month or two. So, yeah, we're still raising, but a fairly modest amount, yeah. Great.
And and just last one or two questions like one in your industry, which let's say the market, what is the AI adoption right now and and where do you see going? Where do you see it going? I think every team at their offsite is saying, "We want to adopt AI more." So, I think the intent is very high. The actual adoption into workflows is fairly low, very low. So, there's quite a gap between the desire and intent and actual roll out at the moment. You know, I think people obviously have their ChatGPT or Claude subscriptions.
And that's probably where most companies are starting, rolling that out to all their employees, dealing with security and all of the headaches that can come with that. And then they're looking at, you know, specialized vendors on top of that for parts of the workflow. And then of course, there are some vertical agents that maybe promise to do it all. I always think that's often harder than it looks. So, anyway, all of those sorts of three solutions are out there and and certainly getting a lot of interest from buyers, but you know, a lot of a lot of prototypes, a lot of people trying to see does this actually work. I think there's also quite a bit of skepticism in buyers.
So, they really want to see it working before they commit. Right. I want to end this with a question for somebody who is like very young and like a 22-year-old who has joined an investment bank and you know, is probably listening to this podcast. What's the What's the single thing that you would like to tell them today that they should start doing to sort of, you know, learn about AI or catch up with what's happening with AI adoption in financial services or investment banks? I mean, I mean, be as AI forward as you can, right? You know, go back to first principles if you can and you're interested, just absorb content.
If you're, you know, at the gym, listen to be listening to YouTube's about AI. If you've got some spare time, crack open, try and try and use like cursor or Claude Code to do something, you know, cuz I think this generation has the opportunity to be AI native and eventually the older generation will probably they'll be fall into two camps. Some some that adopt it and some that are like a little bit like this is too much, I'm out, I'm going to go play golf and that's where the newer generation will have the opportunity to step in. Right. Great. This has been amazing.
I would like to end this by sort of quoting something you told me last time and that has sort of stayed with me. You said that Claude Code has radically reduced the value of a line of code from $10 to 10 cents. So, go and tackle bigger problems. Like don't don't tackle like $25,000 problems anymore. 25,000 line code problems. Go and tackle like bigger problems in life.
So, that was really influential, you know, as a as a as a thought process. Yeah, and you know, I I 100% agree. You've got you can be more ambitious now. You can you can tackle, you know, way, way, way, way bigger problems than you've ever been able to before. And honestly, if you don't what value you bringing to the table, right? I mean, if somebody can code something up, we all know that, you know, once you get the code base gets beyond, I don't know, 5,000, 10,000 lines of code, it gets more and more complicated.
So, you are bringing bringing something to the table, but, you know, yeah, why don't you have set your sights on on, you know, a really you know, really big problem here because if I I I think it's still intellectual property. I'm still very bullish on you know, software and the future of software, but there has been a reset in terms of, you know, the value that one line of code brings. And I just think it's up the bar. It hasn't changed the game. Yeah. And whenever you meet my CTO, I think he'll completely agree with you.
He says something very similar. But thank you so much, Andrew. This was really amazing and I learned a lot and I'm sure the audience also probably took up, you know, learned a lot from this conversation. Thank you. Fantastic. Thank you, Ammar.
Cheers. Thanks. So, I have just
End of episode · Ep #17