#103 AI document processing, 2024 US elections and more w/ Petr Baudis at Rossum

Hello and welcome back to another episode of the Process and Automation Podcast with the Automation Guys. Yeah, we have now done a bit more than 100 episodes. I don't know, it's quite a lot on the, on the podcast here. You are covering all sorts of topics on low-code, no-code, workflow, Copilot, AI. Yeah. So pretty much everything, all things process and automation. And yeah, a few weeks ago we had great feedback from our listeners and yeah, and they seem, we always repeat that, I guess, because it is very exciting, but they seem to love our interview episodes because yeah, in our interview episodes, we talk about use cases and I think that's what most people then can really relate to and then can see how that works in their businesses. And maybe it's also, yeah, it's something different to just me and Arno talking about here about all things process automation and yeah, and Because of that, yeah, we have planned another interview session today. And yeah, you all know Arno and myself are very passionate about all things process and automation, AI and automation. And yeah, we can't stop talking about these topics with others who share the same passion as we do. And our guest today is 100% that kind of person, I think. So without further ado, it is with great pleasure to have Peter Baudis, founder, CTO, and chief architect from Rossum on the show. Hello, Peter. Very nice to meet you.

Nice to meet you too. Thanks for having me.

Yeah. How are you today?

I'm great. How are you?

Yeah, good here. We have finally a bit of sun here in London, and hopefully you are as lucky as we are.

Same here in Prague, actually. Very sunny.

Great. That's great. So yeah, Peter, yeah, we are very excited to hear your insights today and learn more about Rossum. And of course everything what Rossum does, and I guess one big part is document AI, document processing. So yeah, to kick off, can you please share with our listeners a bit about your background and what you do at Rossum and maybe also how it all started?

Absolutely.

Absolutely.

So yeah, my name is Peter Boudis and I'm kind of computer programmer by trade. Historically, I got involved in many, many open source projects. Actually, I was developing some of the core Linux system libraries, the Linux system kernel, et cetera. Maybe most notably, I was one of the early developers of the Git version control tracking system. So I was working with Linus Torvalds and the team on that, adding many of the kind of user-facing features in that area. But over time, I got most interested in artificial intelligence and machine learning, et cetera. So another project I did is I was very involved in Computer Go. There is this board game of Go, which was traditionally very very hard for computers, but eventually DeepMind came and they released AlphaGo. And actually what they used as a reference player to show how much better they are was my program, which was called Patchy, which was at the time most advanced open source computer Go program. So I got a couple of references in their scientific papers. So I got involved in a variety of projects like this, but during the time I was doing my PhD, I met Tomasz Bogar and Tomasz Tunes, who are the other Rossum co-founders. We met in a single lab and working on various— we were working on various machine learning and AI problems, but over time we realized that rather than working on our individual dissertations, we work much better together and we enjoy that so much more. So eventually we left the university and we set up a startup together tackling a variety of problems and we are kind of on a hunt for the hardest AI problem possible that is still— that we are still able to solve and meaningfully contribute and build a practical solution and eventually we we realized that processing business documents and automatically extracting data from those documents is that kind of problem that's very hard and doesn't have an obvious solution, needs a lot of development, a lot of advanced neural network architectures, et cetera. But at the same time, it's very, very practical. So we started working on that. We set up Rossum and we started building the team. We brought the first product to the market. And over time we realized that actually it's not just about the AI extracting data automatically. What real businesses need is so much more. They need a great user interface for all this. They need a platform to set up all the business rules they need. Obviously then they need integrations to get the data where the data needs to get. But eventually we realized the problem we are solving is actually a communication problem because document there, a business document is actually a communication medium. And what actually is the eventual real need of those companies is they work with other companies and they need to execute business transactions with them. So for that, they need to exchange a lot of documents back and forth, but that, that becomes a huge administrative overhead and they end up with open spaces full of hundreds of people, 8 hours a day, 5 days a week, just typing over pieces of information from some PDFs or pieces of paper in worst case to some computer forms. And that's totally crazy. And it's not just very unproductive work and a lot of overhead for the company. But it actually slows down the business and slows down the whole world by extension because those documents, they get stuck in a queue for 2, 3 weeks before someone gets around to them, but they represent some actual business transaction. So meanwhile, this transaction is stuck. So we are making this faster, we are reducing this overhead by bringing this automation. But it's not just about this typing over information, but for example, what happens if something is wrong on the document? Because right now an accountant is picking up a phone, someone starts typing out an email, your purchase order is not matching, blah, blah, blah. And we are handling and automating that as well. This communication overhead can be easily 10%. Recently we talked to a customer that actually needs to return 30% of their incoming invoices because they are in the travel sector and the hotels, they just invoice whatever and need a lot of handholding. So then this is a significant overhead. So we are solving with our product, this problem end-to-end, receiving the document, kind of processing it, automating all the transactions, the approvals, et cetera, all the way to the integration to your source of truth, to your ERP or whatever.

Okay, far, far more than just intelligent document capture or OCR basically, isn't it? So that's what it was in the past, of course, and then that, oh yeah, that's great, great. Yeah, between Arno, myself, and over the last 20 years, we have worked Yeah, I, I would say with numerous vendors to support companies with, with all things process and automation. Of course, over time we, we also talked about low-code and, and yeah, of course, more about AI. So where do you see the, the competitive edge currently for, for Rossum in the market? So you touched a bit on what it does, but where do you see based on what is out there in the market as well? Where do you see the competitive edge for Rossum?

Sure.

Sure. So there's actually a couple of angles. First, no one is selling quite what I described except Rossum. If you look for end-to-end solution to this whole transaction communication problem, Rossum is that, and we don't really have a competition. Of course, there is a lot of other OCR solutions you can use. You can take some RPA components and build your own solution for this problem in-house, et cetera. So you can solve the problem different ways, but if you want just a prepackaged solution specifically for this, you just go for Rosso. But of course also if your problem is slightly different or you don't want to go all in at the very beginning and solve everything, then there is still plenty of reasons to pick us. First, we are the machine learning leader on the market. So we have the— we regularly hear from our customers that when they do POCs with couple of vendors, we are used to come out at the top in terms of the accuracy rates, in terms of the total automation rates. So that's a big selling point. And recently we had a big launch of a new generation of AI that we call Osome Aurora, which is based on actually a clever trick of combining large language models with discriminative AI so that you get all the smarts, all the new advances in the world of LLMs, but you don't get the risks in terms of prompt injections and hallucinations, etc., because we actually use the LLMs just to identify existing information, not to write out new information, which is a big difference from the chatbot style interaction like GPT-4, et cetera. So that's so machine learning, one other big differentiation point. And another is how do you actually roll out and maintain the solution in the real world? Because obviously that takes traditionally a lot of resources, a lot of high expertise, people who understand all your internal IT systems, who understand the business rules, who can then actually configure the system, et cetera. And what we are bringing is a completely new experience in terms of how you customize the solution for yourself. Of course, one thing we offer is professional services, but we also employ a lot of AI and a lot of novel paradigms, how to get all your computations that you want users to check into RawSum and how you can actually easily maintain it. Because our philosophy eventually is that if ideally when a document comes in, no one ever sees it. In the worst case, people can, people could see it at most once and that's in Rossum. So they don't need to go then back to SAP or another system to revisit it again, et cetera. They do all the work in Rossum and just get it done and move on without the extra overhead.

Yeah, sounds very good. Thank you. Yeah, very cool.

It sounds very good for you. What you do and some of the things you describe, it just sounds like futuristic and scientifically impossible. And, you know, I guess me and Sasha, we, in this Process Automation Podcast, what we've seen in the, I would say in the last sort of year or so, a lot of AI is starting to sneak in. Of course, this podcast is not just about AI. We look at business process automation and anything automation. We had some interesting guests on the show, but I would say your solution that you've described just sounds like incredible. In terms of sort of looking at documents and being able to actually extract things out of it, transactional things, and then, you know, sort of making decisions. And I'm just trying to wrap my head around like a practical use case here. So, you know, if I look, for instance, accounts payable or invoicing, just for our listeners, like what is a practical hello world example of this stuff? Because, you know, in a podcast that people do ask us a lot to just describe what we do in terms of use cases. And that would be really, really good for listeners just to understand in a very, very simplistic way what that looks like and how you get started and, you know, how do you build up from there.

Absolutely. Absolutely. So one thing to note is this might sound futuristic, almost like sci-fi. One thing that's very important for us always is kind of to bring proof, not just words on the table, because you got a lot of companies companies that will promise you that after a lot of consulting and maybe a 3-year implementation project, you will get to 100% automation and life will be amazing. And we try not to do that. So instead, what we always think about is the trade-offs between the length of the implementation, how much you invest into the project versus how soon you can reap the benefits. So our goal is that you get the automation benefits and accuracy benefits as soon as possible. And then over time, the automation further increases. So we invested a lot into how streamlined our user interface, for example, is so that even the user interface itself is a huge value add. And big part of that is also that we try to learn from every single click that you make in the user interface. So the AI can literally learn in front of your eyes and adapt very quickly. We have a technology that we call instant learning where when you correct something in one document, With a high chance, the very next document that looks the same will already be captured correctly. So that's one thing. But another thing is why we don't like just empty promises is we actually have everything available, or most of the things available in our free trial that anyone can just go and sign up. You don't need to set up a sales call to listen to someone show on an artificial example. You can just go set up your account and upload a couple of your documents and see what it does. People are usually, I would say, very positively surprised by what it does in practice, right? So just to know those things. But in terms of practical use cases, what we find is that accounts payable automation and invoice processing is actually by far the most popular one. We focus a lot on sectors which have a large variety of those documents when you are on the receiving side, which would be retail, manufacturing, maybe property management. And sectors like this where you have a lot of suppliers, maybe a lot of also small suppliers. So this overhead is really big and it's not easy to just set up a rule-based solution because every invoice that you receive is looking very differently. But it's not just accounts payable. The solution itself is very universal. So we have also plenty of, for example, sales order customers. When I kind of entered this space, I couldn't even believe that Paper orders are still a thing. Yeah, apparently, apparently, yeah. What I learned over the last many years in the company is yes, they are very much still a thing, especially for larger companies and those traditional deals. And then there is like other sectors we are very active in, for example, logistics, quality assurance. You get those lab reports from, for quality analysis, et cetera. So all those sorts of documents where you get many of those documents, they are variable because you work with different partners and each of them just uses their own format and, and and the information on each document is very scattered around. So it's not about text recognition really, because it's not, it's something very unlike, for example, a contract. It's, this is a document which is very structured. It's a bit of a key-value pairs, but plus some tables, et cetera. And that's where AwesomeAI excels the most. I stood a bit abstract. I don't know if you wanted to dig into something even more concrete.

I was thinking, I mean, on SoftSkill, it sounds like you can throw sort of unstructured documents at it and what will bring back. So let's look at invoices. Invoices are of course different for every customer I sort of ever dealt with. That's what you say, there's no standardization there. And Edge AI sounds like that will be able to sort of process any type of invoice by kind of learning and extracting the invoice amount, line items. And then does it go beyond that and perhaps log into your account system and check purchase orders? Or how does that, how does that work?

Sure. So exactly like when you want to automate this process, you will need to bring all the information that you can into a single place. So obviously, yes, if you are automating accounts payable, so, okay, what you are doing first for each of those things, you can also do it differently if your enterprise is just set up this way. The system is very flexible. So I'll just talk about the sort of ideal setup, right? And it is ideal setup. You set up Rossum Intelligent Inbox where you get a specialized email address for the documents. So the documents go directly into Rossum. This has the advantage that if, so first we have in those integrated spam filters, automatic routing to different queues, to different people, et cetera. Also, if something is wrong, you can reply to the email directly from Rossum. And when you see the document, you also see the email. And sometimes if you are doing the GL coding, for example, or other accounting work, it's actually critical to also see the email or sometimes the order number is written just on the email. So you have this flexibility of seeing this extra information, right? So you get, you set up dedicated email address, you get those documents directly in Rossum platform and those emails directly in Rossum platform. Then your staff, they sort of just have a queue of documents in front of, on their screens and they go document by document. Of course, many of the documents they don't even see because they can be fully automated. But the documents they see, they see kind of side by side what was kept what was on the document. The whole experience is optimized so that they can, if there is one character mistake in the IBAN or whatever crazy field there is, they can spot it very, very easily and they can just sort of jump through things, quickly validate everything makes sense. And what you can do for your staff is you can set up a lot of logic checks very easily where you just write very simple rules, kind of Excel-like experience. We actually fight under the hood, but it's like for you as an admin, it's like Excel-like experience where you say, okay, those two amounts must sum up to the amount total. If not, then throw a warning or throw an error. But what's also super important is of course master data. So you have those orders in your systems, you have even more importantly your vendors all in the system and you want to cross-check with them. So obviously we have a big facility for that. We have in Rossum, we have a master data hub where you configure all the imports from all your systems. You have everything in a single place and every document is cross-checked to this master data. And it's also critical to get using the export, right? So that when you export things, you can be pretty sure that your SAP or your NetSuite or whatever system you are using, Coupa, et cetera, is just accepting those exports and it just appears there. You don't need to revisit again to pair it with the correct supplier records. All that happened in Rosu. And if anything is wrong, you just click the reject button. It automatically can compose the email for you, even describing what's the problem. Because if you see those error messages, it can insert them into the email that gets sent to the, back to the, back to the sender. And that's it.

Yeah, very good. Yeah, of course, the stuff we put in manually in the past, isn't it? Around an OCR or capture solution, isn't it? So, so that was a lot of manual work we had to do in the past.

Yeah, I'm definitely going to sign up for a trial.

Looking good. Peter, I remember you have a very exciting other use case. So accounts payable, of course, that's, and finance, it's also very, very, yeah, it's always very, very clear invoice processing. And I'm always surprised anyway how far we still are to a full end-to-end kind of invoice AP processing. So, but yeah, but that's great that you have that technology there and everyone can use it. But you mentioned there is, there's also an interesting use case you have around reading invoices from advertising, isn't it? And specifically in the electoral election space in the US.

So there is this interesting, interesting case for US elections. It was the case for the primaries, but obviously now we are heading to the big matchup, right? You get a lot of advertising. US is kind of famous as being the, having maybe the most, most intense political campaigns. In media and there was an effort to bring more transparency into what, into how many, how much different candidates and different parties spent on the advertising, right? To give the public more transparency. So actually there is a regulation in the US for the broadcast media. That means the radios, the TV stations, both terrestrial and satellite, et cetera. And this regulation says if it's a political advertisement, You need to report this advertisement to the FCC, which is the regulation body for Federal Communications Commission, right? And what they do is they send the orders, but also the invoices for that with all the detailed spend breakdown, the exact flight times of those adverts, et cetera. They send it to the FCC and FCC makes it open to the public to review and to see, okay, what's, what's, uh, what's the spend, et cetera. Except there is a snag and the snag is what they make available is all those individual PDF invoices, all those crazy low quality scans, et cetera. Because so, so what, what you want is an Excel sheet that gives you this, this detailed breakdown. Okay, this candidate run spent so much, so much on TV in this region, region with this call sign from this date to that date, et cetera. But you don't get that. You'll get get a bunch of gigabytes, gigabytes of, of, of data.

No one is looking through it, isn't it? So, yeah. Yeah.

So, and there is, there are companies who provide insights into this, but of course they charge big bucks for this because they have people pouring over those invoices and constructing the Excel sheets manually themselves. So we are thinking this is a good opportunity to, to do a bit of a pro bono work and democratize this information because of course this This is what our AI engine is perfectly suited for. So we let it crunch all those invoices. Actually, it's still crunching them in real time. Anytime some media uploads the document, it bubbles through the FCC infrastructure and then in the next hour it goes to our system and our system automatically parses it and we are providing this information for free. So we are running a public dashboard where anyone can check out the spends of different candidates, trends over time, geographical trends, etc. And it was a bit interesting around the primaries to see who was spending where, maybe how much actually Nikki Haley, for example, was spending and was focusing on the radios versus was focusing on some like contentious areas and cable network, etc. Now it's a bit of a quieter period, but as things heat up for the main elections in the fall, This will be very interesting to watch and then we will be looking at it with some journalists and some data journalists to see what the trends are and who is spending money where.

Yeah, very interesting. Yeah, it is a brilliant use case for that technology. Yeah, I will look deeper into that as well because I guess it gets really interesting indeed in the next couple of months and then we see how late are these invoices coming through? Is it usually like a month later or two months later?

Actually, it can be the very— there is the mandate that the media must report them immediately. Of course, it's not kind of defined, but it can be the very same day the advert gets into the broadcast.

Okay, that's cool. So we see that data nearly then in real time popping up. Yeah, amazing. Brilliant. Yeah, very, very exciting. So yeah, I will put the link to that. I think there's this, the tracker, isn't that the— tool, the monitoring tool. I will put that in the show notes and then everyone who's listening back can then see and go through the data in real time and slice it and dice it and come up with, with their own conclusions. Yeah. Brilliant. Yeah. Thank you very much. So, yeah, so obviously the topics, intelligent automation, hyperautomation, AI. So when we now look a bit more into the future, where do you see things developing or for companies in these areas and also in intelligent document processing and where does ROSEM fit in going forward?

So I think what everyone wants to do is automate more, right? So, and that can be on two axes. First, like reduce the number of exceptions in your business process, but the other thing is automating more and more end-to-end and there is like many processes you can automate and we specialize specifically on the processes that happen between companies and that's kind of interesting for us and it's a bit different focus than, than for example RPA platforms which focus specifically on processes happening within a single company, kind of automating that end-to-end. For us, what's more interesting is the process which happens in one part in single, in one company company and then ends in another company, which is exactly, for example, this invoicing process. Because that kind of process needs a medium and that medium is a document, kind of universally recognized document that pretty much any human trained in the field can kind of look at and understand, okay, what is going on here, what the Incoterms are like, all those sorts of esoteric things, right? So that's what we are focusing on. That's why we also have this built-in email functionality, for example. In the future, we want to connect those companies more tightly and our Our long-term vision is right now you have a lot of people in the company and you kind of have a number of transactions that every person in the company can process, number of business transactions with your suppliers or with your customers, et cetera, whatever, whenever this process happens. Right now it's a couple of thousands a year that you can handle. With RawSum, it might be low tens of thousands a year. Our long-term ambition is one person should be able to handle 1 million transactions per year. Year in those companies. And so we are aiming pretty high, but we believe with the advances in AI, with advances in those system integrations, etc., this is perfectly possible. And that's, that's what, that's the kind of value that we are building for our customers. Another aspect of this is you don't just want to automate, you are always looking like, okay, what is automation? What does it bring you in terms of, okay, you can process those transactions faster, kind of just in time, sort of. Another thing could be you get more insights from those transactions, more actionable insights that can bring you real business benefits. The simplest one is, for example, payment discounts. You often get early payment discounts for invoices if you are a manufacturing company or a retail company. If you never capture this document, it's sitting in this queue, you cannot take advantage of those discounts. But we can process the document pretty much the moment it comes in and then you can kind of decide decide for yourself, is it worth for me to pay this early or not? Or you can also, okay, even if you want your staff to review every invoice before payment because you are more conservative company, you can still prioritize those documents and let them skip the queue, so to say, and get them processed faster so that you can make this decision, right? So those sorts of insights in the future, it would be insights, for example, around ESG, which is going to be a big topic in the future with those EU reporting requirements, etc. So all sorts of opportunities open there once you have all your incoming business transaction documents in AI-enabled platform. And that's what we look at for the future. And we think some companies might actually become largely automated. There is this actually very interesting case. Just a couple of weeks back, our leadership was sitting down with the leadership of one of our significant customers. I cannot share their brands right now, but they are a travel agency and for me, it was really, for example, in this kind of company, it's a CEO-level project to automate those kinds of things because when you think about it, a travel agency is largely a finance company and largely cynically said, big part of what it does is just re-invoice things from your suppliers, from all the hotels, the travel companies, etc. to your customers, right? And this is perfect example of something that in the future can be a largely automated company and the only thing you want to keep is the front office, both for the hotels to build the trust and reputation for the hotels and for your customers so that they get personalized care. But then inside, ideally from my perspective, you should be able to build this company with just a couple of people inside.

Yeah, of course, because all these matching invoices with orders and suppliers and that stuff, it doesn't really add at that value, isn't it?

So at the front, this is the way you say, which the cost is passed down to the customer. Yeah, I think, yeah, very, very interesting use case. So, uh, anything you want to add, Sasha, before we move on to the non-technical and business?

I have one, one, uh, one thought because that always comes up when we speak about intelligent document processing. And of course structured data and unstructured data, of course, now we can make more sense out of that with AI, but what about handwriting? So I had one conversation last week again, we get documents and there's lots of handwriting still on it, so some doctor's notes and kind of stuff. What is your thought with that?

So of course, this is like something that we support. We are We are not, for example, like to be clear, we are not saying we are the best in everything in this area. And this is an example of something that we support, but we are not the very best vendor here because we just focus on different things. And I think this is also important message for whoever is, for example, running digital transformations in companies is just look for the best, best tool for the job. First, don't just fall for the promises, do like see it in action. And if you are looking for an AI-enabled platform, you should be able just without any prior notice to just send the document in and see something tangible, right? Otherwise it's kind of more smoke and mirrors thing, maybe. But the other thing is like, yeah, what's your use case? Is your use case that you need to process insurance forms or process doctor notes that you have on some shared drive and you just make sense of it? Or it's your use case that you are exchanging a lot of documents with other companies? Because for example, if you look at invoices, handwriting is super are rare. Orders, the same thing. It's just like printed out, sent, and read back again, right? Or ideally not even printed out, just send the PDF, right? So it's a very different case and just choose the best thing for the job. So of course we have customers who sometimes get some orders with some random things scribbled and we can handle that. But if you just have those insurance forms that are fully filled with handwriting, look for the, for when who has handwriting and form processing as their key strength, because they will be better for you.

Okay, good to know. Yeah, perfect. Thank you.

Great. So I think we move on to next part of the program today. And it's tradition on this podcast where we like to sort of, from our listeners' perspective, get to know our interview guests a bit more in a more, like, we've kind of understand the technical bits on the business side of things. And usually what we do is we ask a quick few questions. So I don't know if you're up for that, Peter, and we can fire away with a few. I promise you they're not very uncomfortable ones. So if you're happy, I'll do that.

Yeah, let's do it. Bring it on.

Okay, so the first one is who's your idol and why?

Right. So I don't honestly idolize people much. I don't have that many role models. Who's— when you ask this, I think two people come, come on my mind. First is Jensen Huang, the CEO of NVIDIA. I actually spoke with him personally and I followed him a bit and also his management philosophy and that rings very close to me. And of course I am, I admire his also foresight and the way in 2005, how he actually back then, if you think what was AI and he decided actually to double down on general compute back then, et cetera, and it was like an amazing bet, but also the way he runs the team at the company, I think it's super interesting. And the second person that comes on my mind is Adrian Newey, who is the CTO of Red Bull Racing. He's the best, best engine designer in, in motor racing, Formula 1. And actually, I'm not really into motor racing. I think I've never watched Formula 1 race in my life, but I admire the technology. Behind it. And again, like, it's kind of, he's the CTO, but he's a very inventor type person and he's kind of my, he represents the archetype of the kind of CTO I would like to be.

Yeah, interesting. So yeah, Formula 1 is Arnos and my sort of second or third passion.

Awesome, amazing. Yeah, I got a, now I'm just finishing book by him, How to Build a Car, and now I'm thinking, okay, I should just watch a couple of races. Maybe I will get into it.

Yeah. And you're going live into a location, of course, watching it on TV is one thing and then watching it live. So with the engine sound, I think there is something really exciting. Historically, it was even louder. So that has been taken away a little bit, but it's still very exciting when you see the cars going by.

Yeah.

Right.

Yeah. Yeah. When you're in the UK and up for it, we can invite you to Silverstone or so.

That's a deal then.

Perfect.

Right.

Yeah.

Thank you for that. Um, yeah.

So the next question, ah yeah, we have, yeah. Imagine you could get the contents of a, contents and knowledge of a book instantly. Which book would you choose?

Hmm. Okay. Interesting one. I think the answer differs long-term and like for a lifetime versus short-term, what I'm thinking about the most right now. So I think short, kind of long-term, there is like two bibles of science and computer science. There is The Art of Computer Programming by Donald Knuth and The Physics Lectures by Feynman. And those would be the books I would love to have in my head for the physics lectures. I think it's— physics is a great way and studying physics on the mathematic level as well as great way to learn to think from the first principles and think rigorously. So that would be it. I think short term, right now as a CTO of a hypergrowth startup software as a service company, the books I think about a lot are Inspired and Empowered by Marty Cagan, who is kind of guru of product management, modern product management, and it's actually a discipline that's very surprisingly hard to get right, how to manage building products, and it's not just about, yeah, how do you organize your engineers in a sprint? It's more about how do you map user needs? How do you think about this? How do you think about this in a B2B or B2 enterprise segment because we are usually selling to enterprises and things just move at a different pace and the feedback loops are different than if you built Instagram or a company like this, right? So that's right now, I think that would help me the most, getting those partners on board.

I can imagine when you started building the company, I remember that from building our own startups and yeah, MVP is sort of getting to MVP and pilot proof of concept is brilliant, isn't it?

So then, but to market is a different story, right?

Yeah, it's such an enjoyable journey until then, and then you have live customers and then suddenly completely different product topics come up and support infrastructure topics. Yeah, it's a wild ride.

Exactly, exactly. So yeah, in the first years of Rossum, this was a roller coaster as we are kind of learning this on the way. Of course, now it's a bit different story as we move to like, we already have a pretty strong base of enterprise customers, but to get to this point, yeah, it was kind of adventure at times. Yeah. Yeah. Cool.

Thank you. So we've got a couple of minutes left, so I'm gonna tackle one. So what's the best piece of advice you've ever received from somebody? Best piece of advice.

I don't have a, like, a nice quote or anything that I could probably parrot back, but I always, the piece of advice that has the strongest effect for me is pointing out some obstacle cycle that turns out to be just in my head. Because I think what limits us usually, of course, there's plenty of objective circumstances and I realize I speak a bit from position of privilege in some sense, even though I don't have rich parents or anything, still I have like, I had a lot of possibilities in my life, but I found when I was limited by something, very often it was something just in my head, not thinking big enough. Worrying too much about something. So any piece of advice that when I talk to someone about some obstacle I see, very often, like, what's most powerful is kind of turning it on its head and either point out it doesn't exist or that it doesn't actually block me and I can walk around it.

That's brilliant. I like that. Yeah, it's really, really good advice to anyone. So mindset is a really strong, strong tool.

Yeah.

So yeah, if you could be an Olympic athlete, what sport would you choose?

Hmm. So I'm not that much of a sports person, I admit. So the first thing that comes to mind is chess or Go.

Yes.

I guess it's not an Olympic sport, right? But I would just love for it to be one.

Yeah, it should be, isn't it? Yeah, it's—

Yeah, right?

Yeah.

And then you remember that, or maybe that, I'm not sure if they still do it, but it's this boxing and chess, isn't it, going on?

So I don't know that one.

Yeah, there is, there is, there's competition. Definitely in the past there was something like you box around and then you, you continue playing chess. So just to, I guess that was made as a, as a thing. So because they said boxers are stupid. Yeah. And then they wanted to prove a point. No, it's not, it's not the case. So they're highly intelligent or can be highly intelligent. Of course, not everyone, but, um, so yeah, that was, I think it's maybe some video, YouTube videos out there.

That sounds actually really cool. I gotta look that up. The only variant I know is like drinking games around Chess.com, but that's a bit different beast, I guess.

Awesome. Awesome. Yeah, that's all what we have on the question side. Yeah, Peter, it was a great pleasure to have you on the podcast today. It would be actually great to continue our chat on another episode in the future if you're up for it. And yeah, and I'm sure some of our listeners would like to get in touch after that episode, what will be the best, the best way for them to reach out?

Absolutely. I think the best way is, of course, you can drop me an email, peter.boudis@rawsum.ai, or you can just connect with me on LinkedIn.

Yeah, that's brilliant. Yeah. So I will, I will also connect all these details into the show notes and the books you mentioned. I think that's always great for everyone to just maybe jump into the content you just, yeah, you just mentioned there. Obviously we put the blog post in and link to the company. So all that will be in the show notes. Yeah. Yeah. Thank you very much for listening to the podcast today. Yeah. And as always, we will be back with another episode very soon. Thank you, Peter. It was a pleasure. And yeah, we will be back very soon. And until then, let's automate it.