Building an AI Operating System with Kay Iversen of HelloTwin.ai

AI is moving beyond answering questions and completing tasks to fundamentally changing how businesses operate. In this episode of the CEO Series, Lenny Murphy sits down with Kay Iversen, CEO of HelloTwin.ai and former CEO and co-founder of Market Logic, to explore the shift toward AI-driven business operations.

Kay shares how HelloTwin is helping small and mid-sized enterprises move from fragmented data and disconnected workflows to more autonomous, outcome-driven operations. They discuss the rise of always-on AI agents, the importance of data infrastructure and ontologies, and why organizational change management may be a bigger challenge than the technology itself.

For insights professionals, the conversation raises an important question about the industry’s future: How can research move beyond delivering insights to become an integral part of business decisions and actions?

Key Discussion Points

• From Insights to Action: Why insights professionals must move beyond delivering information to supporting business decisions and measurable outcomes.
• The Rise of Always-On AI Agents: How autonomous AI systems could transform organizational structures, workflows, and business productivity.
• Building AI Operating Systems: How HelloTwin uses ontologies to connect business data, applications, and AI agents within a unified operating model.
• Making AI Accessible to Mid-Sized Businesses: How simplified implementation, affordable pricing, and improved data quality can reduce barriers to AI adoption.
• Trust, Control, and Organizational Change: Why governance, reliable data, and human oversight are essential for successfully integrating autonomous AI into business operations.

Resources & Links

• HelloTwin.ai

You can reach out to Kay Iversen on LinkedIn.

Many thanks to Kay Iversen for being our guest. Thanks also to our production team and our editor at Big Bad Audio.

Transcript

[Lenny] Hello, everybody, and welcome to this edition of the CEO Series by Greenbook. Thank you for taking time out of your busy day to spend it with myself and my guest. And a little bit of a backstory. We've tried this like four or five times, Kay, I think, so fingers crossed that this time we get it there. But joined by Kay, thank you so much for taking time. We're going to talk about all types of stuff. But why don't we just jump in and you give a quick bio, a background, and then we can get into the fun stuff.

[Kay] Yeah, I'm Kay. Thank you very much for having me. It's a great pleasure. Yeah, I'm Kay. I'm CEO of HelloTwin AI. I'm an engineer, if you like, pure heart, love building companies. And what a better time to really do something than at the moment, you know. And I'm sure we'll be talking about HelloTwin in a moment.

[Lenny] We will be. And my apologies for saying Kay instead of Kay. That's my, forgive my Americanism.

[Kay] That's absolutely fine. People normally, when they hear my name, they expect a woman in America.

[Lenny] Well, yes, it's the Americans. We're weird with that type of stuff. But let's give a little more bio because your background actually is really interesting in the businesses that you've built. And particularly that is you went from Market Logic, one of the founders there, into HelloTwin. But when last we spoke, we tried to do this and got rudely interrupted by life, I was surprised to hear about your background in math and engineering. And I thought that was quite contextually interesting. So give a little bit more so that it frames up the rest of the conversation.

[Kay] Yeah, I mean, for myself, I studied electrical engineering and then originally I wanted to become a professor, if you like. And it's not the same. At that time in Germany, startups was not really known, I would say, so much like in America where you learn this stuff on university. Maybe this time it's the same in Germany. So I took part with my team on a business competition for McKinsey in Berlin, you know. And then, yeah, eventually at the end, we were one of the three winning teams. And then, you know, I had to make a decision. Do I go this path or this path? So I decided that, okay, we found that company. The first company was Photonic Design Automation. It was great at that time. It had a lot to do with math, how people were just working in the labs and we could simulate. And it was amazing demand for the entire thing. And after that, I helped a friend in IT planning reposition the company. This company was also sold. The first company later at that time was also sold. And then we started Market Logic. You remember, I was CEO for 14 years in the insights. It was quite interesting because in the IT planning, what I did before, obviously when we went to companies, we went into the basement. The CEO was sitting on the top. And when we went with Market Logic and because our focus was consumer goods companies, and pharma, and marketing and insights was directly coming after this. The chief marketing officer was coming after this CEO, CFO, CMO. So we went all the way up with the elevator. And, you know, that was a super interesting time. Basically, Market Logic was probably the community that knows the company. But it's a type of an insights management system, takes the entire data, consolidating the data and making them reusable because companies investing 200 million whatever per year into market research. Why not be using what you already have? And we added more and more different capabilities. I did that for 14 years and then went into the board. I was a board member at Market Logic. And it was also a super interesting time, I have to say, with personas, with synthesized data, with all this different stuff. Yeah, and we started HelloTwin. In HelloTwin, the basic idea here is... Let me start a little bit different. The basic idea was originally to say... What I loved at the insights... I loved many, many things in the insights industry because it's intellectually challenging. You're on the business, you know, all this different stuff. But what I hated is insights by itself, I think, always had to hard fight in a company to be listened to. So they always by themselves have an existence question. I would say it like that. And now with AI, it even becomes harder because eventually other departments, the IT department is now running and asking why do we need these insights people? So for me, it was always clear. I wanted to build something where we don't stop at the insights, where we immediately go to decision and to action. Because what I learned there was that people are not against brand management, product management was not against insights. It's just, it was too hard. They had their own ideas. It has to be fast, convenient, and things like that. And I think it's even more now than even at that time, that you can't stop at insights. And I know you have published quite a lot, that I think insights people in general have to get out of the loop and on the loop. They have to become insights stewards for the process. So we didn't know that exactly, but it was clear we wanted to build something from the data to the action. And second thing was, I said to you last time, I had always problems to explain my developers, what a researcher at Procter & Gamble or at Unilever does. And I said, okay, if I now have the benefit to again do it, then let's do an industry which my developers understand. So we said, okay, let's B2B software. It's much closer and they build something. Yeah, eat your own dog food. They build something for themselves. So that was basically driving it to do it.

[Lenny] And it's so interesting now, again, for the audience, that this has been long coming trying to make time for this. And when we first connected on HelloTwin, gosh, it was a few months ago, really. The agentic integration component was just beginning, right? We had OpenClaw beginning of the year, but that was still really kind of fringy. Now…

[Kay] See what's happened last night.

[Lenny] Yeah, with Claude, right? I mean, so I've been using Grok Bot for my own. The point is this idea of how rapidly we are shifting, and to your point, from information synthesis, I think kind of the first generation of AI and LLMs and generation into now action. And then moving, just in the last few weeks, the rapid rollout of the decision platforms, Jev, et cetera, et cetera. So the technology is at a broad level to exactly your point from information to decision and everything that goes in between, but not purpose built for specific functions and capabilities. And that's my take on HelloTwin is that you've built a better mousetrap that's designed for specific workflows rather than general.

[Kay] Yes, that's true. I would describe it a little bit different. Maybe I give a little bit bigger frame and then say what HelloTwin does in it, but maybe it's helpful for your audience. So I agree to exactly everything what you said before. I would frame it a little bit different. I would say now getting this generation of always on assistance. Yeah, so we had prompting before. People ask a question. We all love it, you know, but it's personal productivity. Then, you know, this agents were coming up. We give them the task. Yeah, it's company productivity, but it's also around still person and task. So now what's happening with your always on is that this assistant, ChatGPT, Claude, and so on, they become operators very soon. First personal, but very soon in the business. And then you think this to the end, three things will happen and they're fundamental. And when you, and I would call them, the first thing is, you know, when you really think in this people-centric world, you know, how we have built, how we managed and build organization, it's pyramid organizations. We have the boss or top management, then we have middle management, then we have the doers, sorry for that simplification. And then, you know, the boss or the top management gives instruction, you know, we give this to the doers and then we're reporting up. When we now become operators, you know, as AI, the entire middle management is not needed anymore. You know, so we're getting from pyramid to flat organization, yeah? So that's number one. Number two is, you know, when you really think how organizations today are managed, what rhythm they have, then it's a calendar rhythm. That starts with Wall Street, you know, on a quarterly basis. But when you look in your calendar, I'm sure, Lenny, we can see every week particular things that you have set up where you have a meeting with someone. So they're repeatable tasks. This will completely change to a continuously rhythm. So we go from this calendar rhythm, if you like, you know, or fixed thing to this continuous thing. That's number two. And number three is, you know, we as humans like to coordinate tasks. And this is always on. Operators now as agents, you know, it becomes more, you direct them on outcomes. They figure that out. You eventually say, hey, increase my pipeline by 30%. It's a little bit hard now, and then they use the resources, they learn from it and so on. If you take these three things, you know, we're describing a new operating model. So why do we do this? Because we want to improve the cross profit per FTE. I'm saying cross profit. If we take a software company and we forget at the moment, the AI cost, we could also say ARR per FTE. And if you compare that for pure native companies, they're already existing. And you compare that to other startups and normal software, when it's B2B software company, when it's very good, then it's maybe 400,000 per FTE, you know, a startup is maybe starting with 100,000, 120,000 and then up to, but 400,000 is extremely good. Then you look about the AI native companies can be many millions. So I can imagine when I as a company have that profit, you know, then it's very hard for other companies to compete with me. So eventually they all have to change that operating model. So now the question is, when we have to change the operating model, how do we do that? And in the core of this new operating model is what we call an ontology. There are many components to it, but it's an ontology. Ontology see it as the description of what the terms are, what the metrics are, what the relationships are, what the rules are, because when AI should operate, yeah, I'm exactly not say context. Some people say that's context. For me, it's more than context. Context, one components, you know, but it needs this ontology. And now there are basically three ways to build this ontology. There is the first way I call the builder. That's Palantir. So we go in a company, we take that beautiful platform, you know, with all the forward deployed engineer, and then we're building this new operating model. Yeah, that's Palantir. If you have a lot of money, if it's super complex and so on, that's the right solution. The second I call, so first builder, second I call composer. Composer is basically, again, complicated company has already a lot. You go to Databricks, you go to Snowflake, and there are also a lot of startups. You know, they basically now compose a potential ontology out of your data using AI and other stuff. It's a big startup area at this point. But you need a data team for that. And then you can build custom ontologies. You know, the third, and we do that for the operating model. The third is what I call the mapper. So we have the builder, we have the composer, we have the mapper. The mapper is, I have an industry-ready ontology. I map HubSpot against my ontology. Maybe my ontology, I call it account. In HubSpot, it's called company. So it has to be mapped. I can use LLMs for that. I can pre-map it already. Customers don't have to do that. There's a little thing for the customer has to do. You know, I don't need a data team. I don't need consultants. They basically can sign up and can start. So for whom is that the right thing? And then you understand why we're focusing on small and mid-sized companies. So between 20 and 2,000 people, that's exactly for this type of companies. You know, and that is our differentiation. But ultimately what we want to build or what we are building is an operating model, you know… and there are now a lot of, I don't know if we have to go, a lot of technology things. We're using a digital twin in this. We have digital authorities and so on. But at the core, it's the mapper and we just, how we build it and what we do on top to now allow companies to very conveniently, fast sign up and then grow with it and run their company basically in a new way.

[Lenny] I love every single thing that you just said. And the topic, this topic has just been so in, at least in my face, I think in everybody's face, and I'll give you an example of something that I've recognized recently because you and our listeners know, there's a startup fund and there's consulting practice and then there's just the wild hair, experimental crap that I come up with, you know, all types of stuff. And it's all kind of mashed up in there and I spit it out on, you know, content. Point is, working with a few companies in the early stage, early stage AI first companies in the research space, creating solutions for the research space. And some of them are workflow management, right? They're just focused on the research workflow, not the broader business operations that you're doing. I made the assumption, and I'll get to my point here in a second, thinking, okay, well, this is going to be fantastic for small to mid-sized businesses. And then talking to the founders like, well, no, they actually, they're too slow. There are too many barriers of adoption for them because they are so busy working in the business. And although they see the benefit of this shift to a different framework, it's not as simple as just changing the research process workflow. It's the business process that needs to change to support that. And that's a daunting, scary thing. They don't have the time, the money, the internal talent, et cetera, et cetera, for that level of transformation, even though they feel the need. So I totally get where the Four Deployed Engineers with all of the big labs and Palantir, and it's great. To fill the gap that you are filling for the rest of the market to say, look, this doesn't have to be so scary, we can map this, we can implement it, and we can deliver a solution that does this organizational transformation, not just one little piece transformation is hugely powerful. I think there is such an opportunity there. And a lot of companies are struggling right now because they're beginning to get the impact of even as simple as the, we knew this was coming, that it was moving towards MCP-based agentic purchasing. So the procurement departments don't want to mess with sending an email with an RFP and la, la, la, la. Right. And we knew it was coming. It's here now. They're forcing the issue of like, no, and if you can't deliver an agentic response to our agentic request, you're not part of the consideration set. So it's starting to bite. I mean, companies are starting to lose business simply on that piece of things, let alone everything else. So I love what you're doing. I think there's huge opportunity there. I met a company earlier last week that, sorry, and now I'm talking too much, but I think you'll appreciate this. They came to me with the idea of, look, we created a new AI-moderated interview for UX testing. And I was like yeah, and? You know, and I kind of hated that. I was like, I'm sorry. I don't see the differentiation for that. But then what they showed was like, well, but what if we did this as an expert interview to understand a workflow and the output is an automatic map of the agents for the workflows. I went, now that’s cool. Now you're on to something. Now you're on to something. To enable this transformation that you're doing at scale across the board. So I just, I love what you're doing. How's the market responding to your...

[Kay] Yeah, I mean, I would say we started with this idea three years ago. So, funny side story. When we had the first idea, Co-Founder Elizabeth and me, and I said, hey, I want to do it super easy and I want to do mobile first because then when you do that, you build user interface, which are very simple, you know? So then I said, I want to have a conversational interface and that is it. And there was no ChatGPT out. And people said to me, you're crazy. Never ever, we will chat interface, you know, and then all that… So we were very early. And I think we had to learn how to position this. So there's two points to your question. Number one is we had to learn how to position this and we had to learn, we had to position it differently to investors. For investors it’s much more like, hey, you know, we will have maybe a thousand agents in an organization. You as human can manage them. Didn't we need a digital authority? So a manager of this agents were you as a human, give it a mandate, you know, blah, blah to describe this. But reality is when I going out now to the CEOs of software companies, they not have at the moment thousand agents there. You know what I mean? So they need a much easier entry point, you know, where we saying they can first, you know, track their data, then they can basically get a sense and see. So the machine automatically points them, okay, their problems are, then they can delegate, then they can act. So we have a much, much easier way to show them, we give you immediately value and we can talk about that. From a technology point of view, you know, I maybe have this talent when I think about things I can oversee big problems first to see a bigger picture. So I'm not stopping. So for us it took, because we have a, we based on the Snowflake Data Warehouse. Yeah, so we have basically at the end, yes, MCP data can come via MCP live data. We need this because maybe when you have a data warehouse, you have an extract, load, and transform and you periodically upload data, but maybe you ask a question and you need it now, the latest JIRA issue, and yeah. So the other is good for all the analytics we have to do to make the matrix, but it's not good and also for the action because we have to update. But getting, for example, both together is a quite challenging piece. Yeah, without going much into detail. So that's one thing. We have that under control, but it's challenging. Second, it took us quite a while to get, and it's not any more text to SQL, but to get the text to SQL on a level that it makes virtually no mistake. Yeah, and I'm very proud about the solution. So just technology-wise when you have in a data warehouse, and you can imagine we completely work without a data scientist. So normally you would have a data warehouse. You have data engineers. They're basically setting up Snowflakes and building the model, you know, and all these things. And then you have maybe Genie or Snowflake, you know, on top of that because on your data structure, we don't have that possibility. So we haven't, just for the people who really know it, an anchor model, we have a special model and there's no specialist. We have to do everything what normally a human models, we have to do at the runtime to get that really working without any hallucination. Because we can't have that. If I go to a business person and he asks a business question about the bookings and why the bookings taking thing out, should you want to see the drivers and all the different thing and there are mistakes and yeah, you can maybe interpret the facts differently. But if the facts are problems, you know, then you can throw this away. And you get very fast to an 80%, 85% the problem. That's not good enough for what we do. If you want to autonomously let something run, you know, that's not good enough. So that has taken us much longer to answer your question. So position it in the right way and to get the technology to a level that it works in production, this equality we need. It was quite a challenge, I would say. And we had to break many, for us, new ways of doing the stuff, but it's now working. We will go GA by most probably January again in America. Yeah, and I can't really wait to really scale this, you know. There's a couple of things, you know, I have to do something, but in principle, it's fantastic.

[Lenny] 100%. It's funny, I meant to mention. Last night for the first time, my chief of staff, my Grok Bot chief of staff said, you know, if you just tell me what the goal is, I'll figure out the rest. And it just hit me of like that really is such a fundamental change in thinking of like my role now is as an orchestrator, judger, facilitator. It is not in specifying, right? I just need to evaluate the suggested solution on whether I think it is relevant and then test it to ensure its accuracy to your point, which is where the rubber hits the road. So I think that's maybe the human component of this, the trust factor. Actually, I just saw interesting data that I had the privilege of being involved with about two weeks ago that hasn't been released publicly, but on consumer adoption of agents, right, Muse. And, you know, the difference in trust between, oh, reply to an email or schedule something versus perform a transaction. So giving, you know, access to something with financial weight to it. And I think it's the same thing, whether I'm authorizing an agent to go buy something for me or basing it from a business standpoint, do this that impacts my revenue in some form or fashion.

[Kay] And Lenny, no, absolutely. And then think about the situation here. I completely understand what you're saying. And this is fantastic technology. And I don't want to say that's an easy piece. But our main challenge is, you know, we have long horizon goals. You know, we say, for example, this is a goal for this month or even for this year. So when we put, you know, our performance goals for the year in and our digital authorities, which our name is, okay. But anyhow, if that should really work against this long goal over a year, you know, that is what we have to stabilize. And we can't have, certainly now it jumps and goes in a wrong corner and then follows a different way. You would not trust. So we need an infrastructure, basically, which can manage the always on agents, the ChatGPTs and so on via the MCP server, you know, and tracks. But ChatGPT or Claude can ask in our infrastructure about the situation, gets a context, you know, can even do something. It gets a situational context, you know, with the matrix and the goals and everything, does something. And then you as a user can either decide, we call that steer, you know, like you maybe now have on your Claude HubSpot and then it does something. We measure again, but we only steer, you know, it's like a thermostat when you have your heater. You put the temperature up and then goes up. Compared to control, where we basically now, we're saying, write back to us. So ChatGPT writes back to us. We have a gateway. We again look in the ontology. Is this allowed? Is this okay? Does it make sense? And then we write into a HubSpot back. So we can basically track and trace. So you can run our thing either in steer mode. And this is very easy. You just put in your MCP server, you know, our MCP server on your Claude and you can start. And then you can, you know, put stuff from us in and we basically make sure whatever you do in your Claude, okay, works against your goals, you know, and it use your matrix.

[Lenny] That's a hugely important point. I'm so glad that you said that because that I hadn't quite gotten that detail in my own head. So let me pare it back to you and for the audience, right. And again, this is not a pitch per se for our audience. I wanted to have you on specifically because you're on the forefront of the industrial revolution of building systems that help with business transformation across the board in a different way than the broad based, you know, big tech companies are doing. So audience, if we're getting really nerdy and weedy this is why, because I think this is hugely important. Because what you are enabling is a different level of the stack. The MCP server, as you described that, to be the ontology manager effectively from the standpoint of here…You tell me where I'm wrong. Here's what I'm hearing. What I'm hearing is HelloTwin incorporates the systems that allow for the business processes, which are, you know, basically structured as an ontology, a series of business decisions and protocols, right, that, you know, here's what's important and here's how we do that and manages the other pieces that you're using. So you're going to be the hub that whatever your system is that you're using, right? I mean, Salesforce is buying companies left and right. Okay, you plug in the HelloTwin, you know, for whatever, the Salesforce.

[Kay] But then you're only staying in Salesforce, most probably Salesforce is better. But the reality is that companies at the moment even software companies have up to a hundred apps, email, your financial system, Excel sheets, Salesforce, HubSpot, you know, so that's the problem, you know, so you need an ontology basically, because every offset application has its own semantics. You know, you have to bring that in one semantics on what your agents can work with you on, you know, there's one ontology mapped into the ontology. After that, forget, I don't care if it is JIRA or Asana, you know, or whatever way, how you manage your, you know, ticket. A ticket is a ticket. We know what a ticket is and we know the KPIs on top of it and so on and we can with it, you know, that's part of mapping. That's different in the other solution I described, but because we separate it, we can do it. But the other point you described is super, super, super important because you're completely right, you know, ultimately to make a change, processes with a human centric have to change to be agentic. The problem is not the technology anymore, it's the change management. So how can you really get that change management under control? And our answer to that is super easy. You know, we're saying, what's happening at the moment, it's already clear, everyone uses Claude or ChatGPT, every individual user.

[Lenny] It's just a distribution issue now.

[Kay] Exactly, it's a distribution challenge. So they're using this now. So instead of, you know, and we will have no chance to get them from there into our system. Forget that, yeah? So we are, for them, headless. But what we do is that ChatGPT, which helps you with this and this, with this in your function. This asks our system for the situation, for the KPIs, for the, you know, so get situation about this, against goal, blah, blah, blah. So we impact every decision you basically do here. We measure at the end, okay, if this works against your goals, you know, we even send an event, you know, to them and saying, hey, there is the following so that the user can vote. That's what I call steer.

[Lenny] Yeah, got it.

[Kay] Steering it, it's control theory. Sorry, I'm an engineer. You know, it's a heater on your, you steer and then it gets warmer, you know, control means I measure and when it's getting too warm, okay, we put, so to control, to put even more autonomic systems because you don't want to have danger. I'm saying now, okay, you now have your ChatGPT, you have maybe your own MCP server on it. Yeah, as also my MCP server, I give the situation stuff, but maybe to do updates and other things, you have your HubSpot MCP server, whatever. So now I'm saying switch off your HubSpot MCP server in this control phase, if you want, and write back to my gateway. Yeah, so the MCP server writes back into my gateway. I want to update HubSpot, this, this, this, blah, blah, blah. And in my gateway, I check against, again, my ontology, my rules and everything, you know, if this is allowed, and if it's helpful, and we can learn from it and so on, you know, and yes, we do it or no, we don't do it and we tell why. Okay, this is why we call it control. Okay, because now it’s self driving. Now we can give it goals. Now we can control more, you know, and we can have the full power of these always on agents, you know, which are completely by the customer built in their organization, and their processes, and their stuff. We don't have to do anything, you know, so change management is happening basically. The only thing you have to do is in our system, in principle, has to be the person who sets a business role to get the goals, you know, and so on. So only a few people have to be there to manage, you know, the entire rest of the company towards the goals and the impact. Is it understandable?

[Lenny] It is. I'm going to find time to chat with you more because these ideas, this enabling technology to bring it back into the insights function, right? My belief is that we were moving into decision infrastructure, right? And it's decision intelligence and then actual decision and action. And we see companies that have been doing pieces of that. Simile, the digital twin synthetic company, although I don't think that really describes it.

[Kay] They're using the term a little bit different than we.

[Lenny] Yes, they are. They are. And I explain to people like, no, that is a CapEx decision system that is being implemented. So the model they use is taking context data on humans and customers and running through simulations to create the outcome based upon the decision. But ultimately, all of that is almost, it's just part of the ingredients to get to the decision. That's what they are powering. And that process from an insights function and being the connective tissue feeding into that system, I think, is hugely important. But that brings up… and I'm just curious on your take on this, but we won't go too far down. I've been working on this thing I call the evidence admissibility matrix. It's an ontology understanding of the fit for purpose of a data type to drive a decision. So it's really just a protocol to evaluate, does this data support the decision that you're trying to make out of it? And yeah, we'll have a follow up conversation on that. But here's where I thought that was… and what our audience did to appreciate this. The way you've mapped that out and the different functions from control and build, you know, et cetera, et cetera. The world that we are in now is even at a broad level, we have the LLMs, we have the decision systems, we have agents, that technology is here. So the pieces are there. What's been missing is the infrastructure to implement that, particularly for small to mid-sized businesses. Now, I do think it's interesting to see, I think that Meta is going after that with their system because they already host so many small to mid-sized businesses. I would expect to see AWS and those companies to be looking for a similar platform as you. I mean, they, hey, if you're listening Meta or AWS, you know, offer Kay and his team a couple billion dollars. But that is simply where the world is going is for that infrastructure that has to put these pieces together. And it's not magic. The protocol…

[Kay] Yeah. So in principle, I agree. I mean, one of the hottest areas from my point of view is building AI operating system as a big term. And of course, everyone is going there. But for the Anthropics, OpenAIs, Metas and so on, which have a different approach to the world, which we put more intelligence of the company, you know, into the model provider. You know, there is this big fight of Palantir and Microsoft and all. So use model, built, you know, you are, let's say context layer ontology. It's your, you own it as a company.

[Lenny] Who owns the data?

[Kay] Yeah, exactly. Who owns the data? And the model, okay, is just, we can exchange it. We can either use this or we can use this. So we are rather on that end of the business and so I completely agree. I mean, they have enough cash and they can build whatever they want, you know, so. But I think…

[Lenny] Intelligence is a commodity at this point. And I think that people don't recognize that. That’s the commodity.

[Kay] Yes. Yes. A hundred percent. But it's also a question of sovereignty from my point of view.

[Lenny] I agree with you. A hundred percent.

[Kay] Yeah. So, I mean, for Meta, they would need a data warehouse behind. I mean, can they build it? Yeah. I think competition is always good as you win. So, if there is no competition, then there is no market. I think we have, at the moment, a very specific way how we solve this problem. And if other companies are coming up, then, yeah, I'm sure they will come up. At the moment, I don't know someone who does it exactly how we solving it. The overall problem is solving it for many. Anyone wants to now find a good way how the organization can operate. And as I said, when you are a complete SAP shop, then you run in SAP. If you're a complete Salesforce job, everything is in Salesforce. You know, we are interested for the companies who have multiple SaaS systems or multiple systems, you know, that is, and which accept lesser, let's say, more standardization. I know from my time at Market Logic, when I go to Procter & Gamble or when I go to Unilever, okay, they always want to have something specifically how they run the business, okay? So part of when you go our way is you can be quick, you can be fast, you can be easy, you can take from our 500 matrix, you know, you can do your own matrix, but you can learn from that. You can use our ontology. You can extend your own ontology, but you can, you know, work with the baseline. I think that is a big advantage in a way how we learn that, you know?

[Lenny] Yeah, I agree. And my earlier comment was not that, oh, other people are already doing this. It was the great, the momentum, the direction of travel is in this way. I think what you're doing is relatively unique. Now, you know, we first met years ago when I was helping to consult with some of Market Logic customers and evaluating Market Logic as a solution. And you just reminded me of that. I remember thinking like, damn, that's expensive. The implementation of a customized approach for a P&G or a J&J or whatever. That's a lot of money going into building that. That also, I think is, continues to be a barrier for companies in implementing these systems. Even, I mean, just the, you know, the token usage. So for Claude in those systems, right, there are barriers. So give me a sense, two questions, double barrel question here. If you're targeting the SMB market, what do those price points look like?

[Kay] I call them SME. It's very important because I know in America and Europe sometimes there's a little bit different, but I'm not going for the mom and pop shop at this point because they would eventually need a different ontology, you know. So, we call it the small and midsize enterprises. So it's, I would say at the moment because our first industry model is B2B software. So everything, every company between 20 to 2,000 employees is maybe an interesting. In principle, of course, this type of solution gets even more and more and more important as the more complex a business and organization is, you know. But I would say that's where we feel it.

[Lenny] Makes sense. So what are you doing to reduce friction? And my rule of thumb always is, you know, you have to reduce friction and barrier to entry. Give people a reason to say yes and that just means you need to make it cheap and easy, basically. So, to penetrate that market, I agree with 100%, huge opportunity there. I don't see anyone else really playing in that arena. To implement HelloTwin, how are you making it affordable? We won’t say cheap, but affordable for a business that's already kind of stressed, right? And easy to help them deploy. What does that look like?

[Kay] I mean, the pricing is not completely, this is why we in private beta figured out when I look about our unit cost and depends a little bit. Okay. Actually, not the OpenAI cost because we have a compiler. We not actually use so much. We use the top reasoning models for some piece, but a lot of the other stuff, for example, when the pipeline runs and so on, it's completely deterministic because we had to be hallucination free. So, it's rather cheap, I would say, you know, from an OpenAI point of view. So we not have that problem. We have, of course, the Snowflake data warehouse, which has to run and which cost, you know. So I would say at the moment we're thinking around 500 floor price per month. I think something like that plus then at a certain point consumption or outcome based pricing on top, we have to see. So when you start in some areas of your organization, of course, this can then, but I would say on the other side, what we're doing also to get in, that's also learning from the past is and you would have to see our user interface. It looks like a consumer app, you know, so you can just sign up, you can easily select your apps you want to use, you can invite your users and then you basically can start. It is virtually as easy as this but because some people are worried about their data and so on, we also allow people to have a demo data set and you can stay in our app so forever and also you can sign in and you work with the demo data set. And that's the funny thing also is AI now you can create demo. This was for years a problem to create demo data sets and so on. Now with AI, you know, you can just create them, you know, and then you can basically experience the complete entire application and say, okay, yeah, this really makes sense. That would be great if I could have that and then comes the next sentence and is my data is not clean, you know, so that's the next question, you know, and then we have to discuss and we have to show that we are relatively robust. I mean, at the end, of course, if all is garbage in then comes garbage out, you know, that's also clear, but for example, I give you is so nice. You can, for example, just take your MCP server on HubSpot, for example, and you can say, hey, show me the last three deals. It does it perfectly. Then ask, show me the last three deals in Germany and then in our case, I would say from ten cases, five complete rubbish data. Why? Because some of your people of your Salesforce, you know, write Germany and another writes DE for Deutschland, some writes Deutschland wrong, you know, and so on and the filter is because you said give it for me for Germany, it filters Germany, but all the other things because of data quality, they don't do it. When you go in our mapping solution, the problem is not there anymore because everything of this is mapped with LLMs, you know, basically to Germany. It's clean. So we have a lot of these problems already addressed. So from my point of view, you can start and we tell you when there is missing data and it can't be supported and you basically look there, you know. So is every problem solved? No, but I think I have not seen that someone could not start with that. You know what I mean? So to your question, number one is consumer type sync, it will be like you go to the web, you sign up, so very easy, one, two, three, and you can do it. Number two is you can play first with demo data to see if you like it, if you like the experience, if that's what you want to do and basically for all functionality. And number three is we are relatively robust against data quality issues and we actually have a system which helps you to get your data better and better and better because the machine says, hey, I have so many deals but these deals have no currency or what, I don't know, you know, and it's my assumption and you say, oh, shit, or you say, hey, please go back to the user who don't have the owner who has not put any of this data in and then they get prompted, you know, via their ChatGPT or Claude to really do something there. So, I mean, AI needs data, point, you know, and, yeah, but that are the normal things. I mean, the best is just they have to give a demo or something like that so then people can see it.

[Lenny] Well, we'll follow up and see if we can arrange that. I want to be conscious of your time as well as our listeners’ because we can go on and on. I think this is super exciting. I mean, you guys created a category for all intents and purposes with Market Logic, right, as far as the knowledge management for the insights organization, certainly the leader, all that. I think you're creating a new category as well in this. I think it's hugely important. I want to stay engaged and interested, and hopefully our audiences is thinking, well, we want to stay engaged too. With that in mind, where can people find you? How can folks reach out and learn more about HelloTwin since you're still in beta?

[Kay] The best is go via LinkedIn. I'm normally pretty active on LinkedIn so just send me a direct message and I will respond there. Otherwise, you know, my email at HelloTwin, we can say it here, it's [email protected]. You can send me an email if you're interested. Or let us know via our website that you are eventually interested to become a beta customer and then Elizabeth will contact you and we proceed like that, you know.

[Lenny] And by the way, for the audience, Elizabeth is such a joy. I'm still, I love our conversation. I was hoping Elizabeth would join us too because she's so much fun to talk to. Next time, we'll get both of you on. 

[Kay] And much better English than mine. I also have to say.

[Lenny] Audience, seriously, check it out because to be clear, you’re in beta. You're a startup. You're early stage. You're going right through that whole early stage startup model. So, yeah, check them out. Anything else that you want to cover before we sign off?

[Kay] No, it's all fine. It was very fun to talk.

[Lenny] Absolutely. I've really enjoyed it and we'll definitely have you back. With that, we're going to sign off for this edition of the CEO Series. Thank you, Kay. Thank you to our listeners. Thank you to Emma, our producer and wrangler, especially this one. We just kept having to reschedule this thing over and over again. So she definitely earned her chops on this. Thank you to our audio editors, Big Bad Audio, and to our sponsors. And that's it for this edition of the CEO Series. Bye-bye.