Music. Your podcast and YouTube blog covering the German startup scene with news, interviews, and live events. Hello and welcome, everybody. This is Joe from StartupRate.io, your startup podcast and YouTube blog from Germany. Today, I welcome another guest in our interview series sponsored by Hassan Trade & Invest. I would like to welcome Patrick here from Frankfurt. Hey, how are you doing? Hey, Joel. Thanks. I'm feeling great, and I'm very happy to be here.
It's good that you feel great, because we had to move a little bit our recording due to me being sick, you being sick, and now finally we made it, I think, after eight weeks? Something like that, yes. Something like that. Yes, great. Let's talk a little bit about our Enabler HTAI and the Enterprise Europe Network Hessen. This recording was made possible by HTII and the Enterprise Europe Network Hessen.
These organizations have made tremendous contributions to helping startup businesses succeed and thrive, providing a range of services from helping to find grants to ongoing partnerships. By taking advantage of these resources, startup companies can network and develop innovative strategies for success on the international stage. The dedicated support of HTIA and the EEN Hessen is paramount in providing startup
businesses with the tools for lasting success. You can learn more in the links down here in the show notes. So, Patrick, we met in the past, we have realized. And if we're talking about the past, I mean like ancient times in terms of the startup world, right? Absolutely. If I recall correctly, we met at the Startup Week in Frankfurt that must have been over 10 years ago, something like that. I think it was a Startup Weekend like nine years ago, but no big difference at all.
Fun fact that I found my first co-founder from a very first startup at this very startup week. That was actually like really my kickoff into the startup world and being a founder. A big mental high five to Mario Hachma, one of our co-hosts who actually organized this startup weekend. Thank you very much. Big shout out to Mario in any case. He's awesome. Yes. Today we are here to talk a little bit about who you are. What you did, and OmniFact, an AI startup.
But first, I was going a little bit through your CV, and I see you are actually a developer at heart. Is that true? That is true. That is true. I studied computer science and business, but I only finished computer science. You've done a little bit of financial analyst, working student, student worker. But then, from what I can see, you've been an entrepreneur all your professional life, going back to 2012. Is that true?
That is true. I studied law for like one semester. Then I saw that that wasn't for me. Then I did what I loved. I became a programmer. Then I thought I had to study something. Then I started business, which kind of bored me. So I went to computer science. And during my studies, I started a small boutique for pen testing, actually, and social engineering. It actually came from the hacker scene and the security scene in the very beginning.
What did you do there, and are you currently social engineering us? If I would tell you, where would be the fun? What we did there was basically we tried to help mid-sized German companies to secure their network. And we especially helped executives who traveled to countries like Israel, China, Japan, Middle East, where you have actual risks of being targeted by state actors or large organizations to be more secure in their travels and their daily lives.
And we were really lucky. We started, I think, like four weeks in or eight weeks in, Snowden came out. And that basically was the best push for any starting business you could imagine. I see. And then at one point, you started building AI and software solutions for other people. Yes. So from there, after the startup weekend, I made Khan, which still is a great friend of mine, with whom I started SecDash, which was like a little bit between today and the past.
It was a startup to monitor the chatter of IT security issues in the dark web and IRC channels and so on to basically get an early warning system for new attacks on websites, on servers and so on. And that was actually the point where I for the first time worked with natural language processing. So that basically was the first time was working with the technologies that later became what we currently call AI.
And that startup failed miserably on the business side, but it was a great learning experience. And during that time, I met Florian and Alex. Florian, I'm still working with both. Florian is my co-founder at OmniFact. Alex is also part of the founding team. And from meeting these two, we got into building Rocketloop.
Which was like a boutique for building digital products for well-funded startups, and larger organizations and I think the past seven or eight years we mostly built, first of all it was called data science, then it was called machine learning now it's called AI but basically state-of-the-art. Machine learning AI implementations but on a project basis we didn't have a product at this time, How? So Rocketloop is still around the company.
You spoke about it in the past, but Rocketloop is still around and you're still working there as the CEO. What was kind of the kickoff, the spark that made you start OmniFact? Good question. There were two sparks, actually. First of all, Florian and me, we always wanted to transit from being a project company to a product company.
And when GPT-3 came out, it was a little bit before ChatGPT Florian showed it to me and I was like okay, we can basically forget that we're going to do individual natural language processing, projects in the next few years so we really have to get thinking and have to find a way around that and at this point it was more like being sure that Rockloop is okay rather than actively searching for a product, And then we went to our customers and, as you know, we are located in Frankfurt.
So most of our customers are in highly regulated interest industries. We work with banks, insurance companies, mad tech companies, pharma companies. All of these have like very high standards when it comes to where they can put their data, how they can use their data in some context, how they have to be able to get be conscious about data privacy and so on. And we went to them, like, a little bit provocative and said, like, look at this. We have a cool demo with GPT-3 at this time.
When are we going to automate your customer service based on that? And then, first of all, they were all very excited. We did some prototypes. But then we talked to the compliance and information security people, and they were, like, very not so happy. Having a single supplier risk with OpenAI, sending everything to the US, was just already for prime time.
And that was, first of all, a little bit of that was basically the start of Flora and me thinking, oh, maybe there is a chance to build a product that enables organizations to use these kind of technologies to automate their daily business processes. I see. So let's talk a little bit about OmniFact. We already know it's... What natural language processing model is behind it? So OmniFact more or less. We see OmniFact as a platform that enables AI adoption across companies.
And allow me to circle a little bit back. So I think we all have heard so much about AI. And there's like McKinsey had a study in May where he said we can automate up to 30% of all business processes in Europe with AI. But as you know, or as everyone sees, there's very little to show for it right now. There's Klana, who has one good example how they automated their customer service. But beyond that, there are not many cases where we actually have great AI solutions in production right now.
I actually have to admit, I'm currently in a little bit of discussion with Google Ads, and it appears they only have chatbots there anymore. And they reply always with the same stuff. But I'm a special case because I'm a freelancer working on the brand of Startup Radio, which completely throws out the rules, even though it's very legal here in Germany. And therefore, they are doing nothing else than repeatedly sending me always the same message.
Yeah, do it with an exact A has to match exactly B. I can't do it. Yes, do it again. A has to match exactly B. I can't do that. So you always get the same message back, and it feels like you're bumping your hand on the wall. So there's good customer service with AI, and there's very bad customer service with AI. Well, I have to admit, I expected more from Google.
I think, to be fair, Google is a large organization, and I think they were prohibited to use Gemini for a long time in their processes. So I think there's a good chance that this will change soon. But to circle back to your question, we try to solve the issues that prevent companies from adopting AI and building solutions like good customer services, really individual, has access to the CRM, to the product information, and so on.
And you can bundle that to give you a very personalized experience. But to do so, we actually think it's a bad idea to focus on only one LM provider. So what we do is basically we have the possibilities to work with all cloud providers, so Entropic, Google, OpenAI, and all Azure-based models. But we also support self-hosted models. And that actually gives us quite a bit of flexibility to basically choose the right model for a job.
Because, of course, if you want to write marketing content for startup radio, you probably want to have access to the most powerful model. And you don't have to think about data privacy in any way. Yes, because apparently the podcast is meant to be published, right? Absolutely. Absolutely. So if you work in this space, you could probably also work with Claw directly or JetGPT. But if you want to review job applications, for example, this is not the case.
There you have to be very sensitive when it comes to data privacy and how you process that. And if you're a large organization, you probably want to use a private model that you host yourself, maybe, just for these use cases. And what OmniFact allows you is basically to set up, we call it spaces, like AI assistants that can automate certain processes and basically tie them to certain models. So you can basically set the data privacy restrictions based on the use case you have.
And additionally, you can then connect your internal data and services also to these use cases. So you can make sure, okay, we have our block database and that is used to create a new block entry, but that can be used with the best public model available. But our HR space is restricted to our local LM that we run ourselves or we have in our virtual private cloud. But that can also interact with our personio and our database of applicants, for example. So that is the main idea.
Of our platform to basically cater to the needs of organizations, in a way that we allow for every use case to have a conscious decision about the data privacy requirements, where you want to run information, and which of your systems and data sources you may connect to this use case to make it more efficient to allow for more automation. I see. When we talked before, you said that is three highly regulated products.
Industries. And they have the usual headaches, control over the data, including data secrets and GDPR, connecting with proprietary systems and fast development in the new models and overtake actually the projects. Because OpenAI can much better, has more financial means to really work on an AI model than any given company or most companies out there who are willing to do that. Absolutely. Absolutely. You basically went down the list that there's actually now a pitch deck.
So it's data control and security. Looks like I've seen it. If you have that established, you can basically have your data and services integrated into some AI solution. And that allows you to actually build meaningful automation steps. But you are absolutely right that if you would build our own models, it would be a David against Goliath case, but not one you could win because that's a play of power and money right now.
But the interesting thing is you have a few certain developments that are super interesting. First of all, you have companies like Mistral, which do an extremely great job to build models here in Europe that are also available, hosted in Europe by a European entity. So you have the GDPR thing covered. And they also allow large organizations to license their models and run the large models within the infrastructure.
And in addition you of course have Meta which released the Lama 3.1 models which are exceptionally well and also run within the company's infrastructure.
Looking forward I wouldn't be surprised if in the future a lot of like meat sized models that can run on hardware that costs I don't know like between 15 and 20k to buy could solve very meaningful use cases And once you have that established, I think the question will always be, do I need the largest, most expansive, but also most closed and single provider restricted model for the use case I have? Or isn't the...
Aren't most use cases solvable by much smaller, cheaper models that I can control myself. And our guess is the future will be hybrid between like models you run within your infrastructure and public models and maybe some things in between. I think it's still an open question how much fine tuning will play a role so that you have like a base model, like for example, Lama 3.1 and you want to be basically adapted to your use case or your language or your knowledge from within your organization.
If that plays a large role, then again, you're going to have self-hausted models, which would be very important for us. Another way to include companies or organizations' information are ways like retrieval augmented generation, where you basically. Use the language capabilities of a model, but basically force it to base their answers based on information you provide them based on the query or the question user asks.
And from what we have seen so far, it looks like that these solutions without fine-tuning, without individual training are probably good enough for 85, 90% of the use cases. And it will be very interesting to see how this will progress as the models improve. Because I think right now, no one has a good grasp of how great the GPT-5 grade models will be. That is an interesting question. We have now doing a little bit philosophy. Can we get really down to business?
What could a potential client that is listening right now actually do with their tools? Where can you apply it? where do you have use cases where you can actually use the tools. Absolutely. Absolutely. So the very first use case that works off the shelf and is very helpful for most organizations is basically knowledge management.
Since you can use OmniFact to make the information in your document management, in your Confluence, in your Notion, and so on, systems available through an iAssist system that is centralized at one point, you can enable much better access to companies', information. Where we have seen this to be the most successful so far is within banks because they have something like, and it's hard to translate, schriftlich fixierte Ordnung.
So they have to basically have a written version of every process that is relevant for the bank's organization. And they are normally organized by the department responsible for the process step, which leads to a document that has thousands, sometimes 10,000 of pages that are very hard to comprehend as a person going through. But with an AI system that can basically access the information and summarize it based on your question. We can make this much more useful as a source of information.
And that's actually the first high-impact use case we have seen to actually make this process documentation that many organizations must have because of regulatory requirements to make them actually a powerful tool for their employees to access the information they need much faster. For everybody with no financial background, Schriftlich fixierte Ordnung literally translates to written fixed order, which is documentation implying a structured or formalized system.
And this formalized system basically tells you how you have to behave your processes as an employee in a bank in order to be compliant with the regulations. Plus, it also can be audited by the supervisor. Plus, you can also be checked against this SFO if you're really doing your stuff. And believe me, if the auditor does such an audit, commonly known as a 54 audit, Paragraph 54 KBG Prüfung, it's completely a fun-free zone here. Believe me, that is the moment when every bank CEO goes, uh-oh.
I want to share some context on that. Why this is so interesting and so hard. Because if you're not within the financial industry, having a documentation of process can't be so hard. But the interesting thing about this Schriftlich Fixierte Ordnung is you have many departments within a bank, and every department is responsible for their process step.
So basically, if you want to describe the process of granting a loan to a customer, this will involve six, seven, eight, nine departments throughout the bank, one bank. And basically, you don't have one document that's grabbed the whole process, but you have one document for each process step. And that actually makes this documentation as it exists completely incomprehensible in many cases.
But with the AI, you actually can now ask the questions like, how do we organize our loan granting process? Who's responsible for that? Who handles the risk assignment? How do we handle ESG risks in this process? And it will basically puzzle together the information from hundreds of documents to give you this explicit answer, which in the past would be very, very annoying to basically get in a very compact way. And you probably would have to speak to three or four people who give you the
right hints to basically find that out for yourself. I see. So, for example, written fixed order. That's one use case. Can you tell us, like a few others, what you can also do? Because this one is, I would say, pretty much limited to European financial institutions. Also, we've worked with one of Germany's largest landlords. And they have an interesting challenge.
They grew relatively fast and not only organic. So basically, they bought together a lot of buildings and joined or merged certain companies. And this led to the fact that for every building they have, they have a different kind of structure sometimes for the documentation about that building. So if you now want to ask, when was the elevator last audited?
And was something found in the audit? Has it been fixed? you really have to dig into your document management system to find this kind of information. Sometimes through PDFs that are not, weren't run through OCR, so you can't really search for it. And we help them to basically build AI systems for these buildings that you now can ask these questions. And this basically reduces the time that they have to get this information from minutes to seconds.
That is not a use case. I'm currently focusing a lot when I speak about the use cases on what we like, what is actually shipped right now, but to, to extend it a tiny little bit. So currently we focus a lot on like how we can make the information that the organization has and that's persistent. So something that is in confluence in the document benefit system to make that accessible through an AI system.
But in parallel, we are currently working very hard on making, also giving our AI systems access to live systems. So if you ask the OmniFactor systems a question, the first thing it asks itself is, what kind of tools do I have to answer your question?
So, and right now it has two tools. It either can give you an answer directly if it's trivial, or it can go and ask a knowledge base, which basically is a retrieval augmented generation implementation that basically gets you the information based on the data source of where we're at it. But what we currently do is we're going to provide a number of additional tools.
One that I showed you a little bit before the call is we're going to have web search and web browsing, which most people know from JGBT. But what is more exciting is we're currently building something like a plug-in system that allows you to create a tool that, for example, would do API calls to your customer relation management system. Or to your product information system.
So if you combine these tools, then you can automate very interesting things because then you can say, oh, I have this customer email and the customer complains about, I don't know, a broken piece of hardware that they bought from you. You can basically then take this information, look up in your CRM, see if the customer actually bought it, when they bought it.
Then you can go and check your knowledge base for the warranty rules for this certain product, can check that against the time it was bought, and can basically check if the customer would be eligible for exchange, for example. And that you can basically now automate end-to-end because our AI system can basically plan the steps. It will ask you, hey, Joe, this is the steps I would like to go. This is what I want to do to provide you with an answer. Then you can say, no, I don't know.
I know already it's beyond warranty, but there's a certain reason that we may exchange it anyways if it's from that product line, for example, because we had issues with that. Then it would change the plan, then it would run the plan and give you the complete answer. And these kind of like end-to-end automations that become possible once you have access to not only data, but also like systems within an organization.
This is what excites us a lot, because I think at this point, you can make people much more efficient. For example, what would be great if at one point in the future, your sales employee meets somebody on a fair, talks to his phone, says, okay, offer this and that to this client, get it ready for me in the drafts. And then the next day he's in office, he just has to review it and click send. Absolutely. I have another example that comes from my daily work.
So currently with all larger customers, I still have like one-to-one calls to basically understand their needs, understand their use cases. And really still learn, also doing sales, but also to learn for us what are interesting use cases. And after this call, I normally, I write down a lot. So after this call, I go to my CRM, basically put in what is interesting, if it's an interesting lead, how large the organization is, and all this information.
I go to our Slack channel and report back a little bit because it's sometimes very exciting and interesting for our team to see what we're working on. Then I go to Notion and put in if we have any feature requests that are new to us that we didn't have, or if they mentioned a feature request or feature that we don't have yet and add that. And then I send them an email, depending on how the call was, with our product presentation and so on.
And ideally, I would like to just put all the things I wrote down in OmniFact, and OmniFact would understand, oh, I see. This customer is interesting to us. It has this potential of seeds. They want to test with that. I'm going to put this into the CRM. They also mentioned this in these feature requests. We have that in Notion already. Let me update that for you. And also, here's a summary I would like to set to Slack, and is it okay for
me to send this email to the customer? so and if we get this point like my life and the life of. Million of salespeople will be much, much easier. And you can basically think the same thing about marketing, about your work with setting up the interview with me, for example. There's so much potential for automation. If you have these tiny steps and this planning ahead that you still can interact and change with and nothing happens with you approving what the AI plans.
But this is basically the midterm goal that we want to achieve. Allow for this to be the glue between all the systems and use AI to actually have like end-to-end automations that take away like the burdensome, annoying little steps we have to do by changing tools, changing platforms, duplicating data and so on. It also sounds to me like you would make the SaaS process more trackable as well as more efficient.
Absolutely. Absolutely. Do you also see, because you have sales on one side, do you also see something like this you could offer to customers in terms of product development, product ideas that they may like to have in the future? That's a very interesting question.
If you think from a product manager perspective, I think having better access to user feedback by being able to search through it more efficiently and also by being able to search the web, compare with the competitors, and to probably combine information from your CRM, from your support and so on. To get a more holistic view of your customer's actual needs can be very helpful.
In addition, what's also important, if you have tools that have analytics within your app, you want to combine this information, like how often was feature X used versus feature Y? Can you group that into cohorts? These are all things that currently are done manually that maybe not now, but in six months or so, AI could do for you. And again, we see ourselves as a platform enabling that. We are not a CRM. We are not an analytics tool, but we want to integrate.
We are integrating with these tools to basically tie this information together, to give you like an assistant, like a student intern that can do meaningful work. It probably needs some oversight and some correction from time to time, but it helps you to do complex steps with a few instructions and less amount of time spent than if you would have done it yourself. Sounds pretty good. So we already know about your future plans.
We usually close out those interviews with two. Now we do three questions. I would assume you are open to talk to new investors. Absolutely. So when we transitioned from the project-based Rocket Loop work to our product company, we did a very small friends and family round. And we are currently planning for a pre-seed round, seed round. The pre-seed round probably going to close by the end of the year and the seed round somewhere next year.
So we are very much interested into discussions with investors. There's still some room for business angels, but at the end of the year, we will probably talk more to institutional investors. Sounds pretty good. Would you also like to hire more talented people? Do you offer jobs? Yes, mostly in communications, marketing, sales, and engineering.
Mm-hmm. For everybody who would like to learn more, for the investors, we link down your LinkedIn profile, and we also link down here your career website. My last question, since this is sponsored by Hassan Trade & Invest and the European Enterprise Network Hassan, is there something you would like to address to the decision makers here in the state, what they are already doing good and or what they could improve in order to make this a more sustainable startup hub here?
The thing is, there's Hassan AI and they do a pretty good job. I think we have a very strong hop in Frankfurt or in Hessen when it comes to finance for the finance industry. We also have one with the pharma industry, but I have the impression that they are not so much involved with all the startup AI and data meetups and events and so on.
So maybe that could be an improvement. what is interesting though we have been in talks with public entities but not from Hesia yet so, from what I know so far Hesia does a great job, with enabling startups, hopefully the next time we speak I can say more about the AI aspect of it because for now, we have not taken too much steps into finding that out. Sounds pretty good.
After long weeks of trying to organize that and more than 35 minutes of recording time, Patrick, I wish you all the best, best of luck, and thank you very much. Thank you, Joe. It was a pleasure, and thank you for your time. Bye-bye. Bye-bye. Music.