In this episode of the Connector Podcast, host Koen Vanderhoydonk interviews Geert van Kerkhove, co‑founder and CEO of Oper, a Belgian B2B SaaS fintech focused on digitising the mortgage journey for banks. Geert outlines his professional background, the two fundamental challenges that hinder full digital mortgage processes – borrower anxiety (the need for human trust) and the heavy reliance on paper and compliance checks. He explains how Oper harnesses large language models (LLMs) and generative AI to create a data‑and‑document verification engine that cross‑checks income, bonuses, employer details, and other loan‑related data, flagging inconsistencies for loan advisors and back‑office staff. Pilot deployments with three banks have shown a 4‑6 hour reduction in verification time, though convincing compliance teams about data privacy and model hallucination was a major hurdle. Geert discusses broader market trends: a post‑pandemic slump in mortgage volumes has pushed European lenders toward cost‑efficiency, making AI‑driven automation a hot topic. Oper addresses regulatory concerns by building a proprietary control layer that ensures traceability, prevents hallucinations, and keeps the AI behind a safe UI, rather than a public chatbot. Looking ahead, Oper plans to expand product features – guided sales, renovation advice, and AI‑generated underwriting decisions – and to scale from 15 institutions in six countries to a pan‑European presence in 27 countries. Contact details (sales@operacreatives.com, LinkedIn, website) are provided for interested banks.
Chapters
Welcome to the Connector Podcast, an ongoing conversation connecting fintechs, banks, and regulators worldwide. Join CEO and founder Cohen van der Huydonk as you learn more about the latest available trends and solutions in the markets.
Welcome to another podcast from the Connector, and today I've got with me Geert. from one of the most biggest and most promising fintechs out of Belgium. It's a real honor to have you. Oper, can you tell me a little bit more about you and your company?
Thank you, Koen. Super amazing to be here. So I'll start with my company. So I'm personally co-founder and CEO of Oper. We are a Belgian boring B2B SaaS company. And we help banks on the digitization of their mortgage journey, but we'll probably talk a bit more about that today. And about myself, Geert van Kerkhove, as I said, co-founder, CEO now, being an entrepreneur for roughly 6, 7 years. And before that, have always worked on the beautiful borderline of banking and technology as a management consultant at another startup on the product side, and also as a consultant independent consultant to banks. So that's a little bit on my background and a little bit on Oper.
Oh, thank you very much. And I still remember the day that you switched because that's where we got to know each other.
Yep, indeed.
Since you guys are in the mortgages business, it's often a business that is not well understood by many. So could you tell us a bit more about what are the general challenges in the mortgages industry?
Yeah, challenges per definition are not well understood, but on the other hand, I think a lot of people in their life have bought a house or have bought an apartment. Obviously, I don't think many of— indeed, I don't think many of us just could put the money on the table that just say, here is €300,000 or €500,000 for a house. Um, so everybody had to do with a mortgage process. Um, and I mean, I've been now on the digitization— I've now been on the digitization side for, uh, roughly 12 years. I've been working on this, on this thematic, and I think there are 2 things that happen in this journey, especially in this contact to contract from you as a borrower thinking about, I want to make an offer on a property, can I afford it? Up until you're going to a notary or a conveyance lawyer signing all the documents. And there are 2 main issues if you want to digitize this. The first thing is what we say borrower anxiety. And actually, and that's also the reason why 100% fully digital application processes don't really work because a borrower wants to talk to somebody about trust. Not every mortgage project is the same. So, I mean, if you want to bring that to a digital Uber-like experience, that's really tough. That's the first problem, and a lot of banks are struggling with that.
Okay.
And the second one is the involvement of paper and compliance and, you know, making sure you can cover the risk. Um, and that's what we always say, you know, as a mission statement, we want to burn paper with our software. We want to reduce the amount of paper reduce the amount of verifications, but that's also really tough. But we feel that if you as a bank wanna say, look, we really wanna bring the mortgage process to an experience, these are the 2 things you need to tackle. You should make sure that a client has a fully digital experience but still has the trust to talk to somebody.
Mm-hmm.
And at the same time, it's really alleviating your back office by reducing the amount of data points and paper that needs to be verified, but you still wanna make sure you And then you guys go even a step further because you're bringing AI in this process.
And I think today I'd love to go a little bit deeper on the new product that you brought out. So could you tell us a little bit more about this and what has been the inspiration of this?
Yeah, so the inspiration is, well, of course, you know, when ChatGPT got released by OpenAI, we were, of course, curious. Everybody was jumping on it. So we started going a bit of a level deeper because, I mean, my team and I have a bit of experience really in machine learning as well. So we kind of knew what could we do with machine learning, what can't we do. And what we quite quickly found out is if you look at LLMs and if you look at GenAI, it is really good in predicting in context what is the next word going to be. Like it can, and I mean, there's also the visualization part, but especially around predicting text in context. It is extremely powerful and it is plug and play because the foundation models basically provide you with a pre-trained set. And then if we looked at the problems we were facing, helping loan advisors give better advice for a mortgage is just a textual problem, basically. I will upload 20 documents, 5 salary slips, a tax. I mean, processing the data out of that and basically saying, hey, this is the loan application of Koen. Actually, Koen's profile looks really well. It benchmarks very well against our risk policy. Again, is a textual problem. So we got very excited about the fact that a lot of problems we had could be solved by LLMs, but of course need to be solved in a compliant blah, blah, blah way. I can go a bit deeper into, you know, implementing it for real, but that got us really excited. And we just started, you know, brainstorming. We developed 20 use cases, start— made our clients discussing them, got like a top 3, and then just start doing the work.
And can you shed a little bit on those examples that eventually made it?
Yeah, sure. The first one that's now productized is our data and document verificator. It's a very simple use case, but it's an incredibly powerful one. So it basically takes all the documents and data points that you will submit as a borrower and basically starts cross-checking All the data. Example, you have in you said I make three thousand euros a month, and at the end of the year I have a bonus. Well, what we would do is we would take all the documents that you have provided to the bank. We would process all of them, and we would say, well, actually Koen makes two thousand five hundred, and his bonus hasn't arrived on his account. So we would push a warning sign to the loan advisor, but also to the mid office to say, hey, something wrong is here, or. Employer address of the employer isn't correct. So this is a very basic use case, but you wouldn't know how much lenders still do this verification. Like bank advisors would do this verification and they would do it in the back office. And so we are now like in a full cockpit, fully automating that. And the results are pretty amazing, actually.
What intrigued me here is that you said that you're contacting clients to find use cases. Have you done the other way around? Have you already got feedback from clients?
Yeah, so we have now tried 3 clients that are in the pilot program with us. Um, there, 2 have given their go to put it into production. And the largest hurdle actually was to convince compliance about, you know, how are we using LLMs, is data leaking, et cetera. So that we got covered.
Mm-hmm.
Um, but they really see a, like the processing time for especially a loan advisor to prepare for a loan conversation or to submit the loan application, we actually see that we can reduce 4 to 6 hours of the verification time right now. And this is really a pilot case. I think when we get this into production, we will be able to share much better metrics.
And can you share a little bit about— because I think it's a very valid point that a big topic was data privacy compliancy. Can you share a little bit about that? And maybe general also, what is sort of the trend within the European mortgages market and how does that fit?
Yeah, so I think in the mortgage business, what happened in the last 2 years ago was the worst. I mean, in terms of volumes, mortgage markets went back to 15, 16 years ago. So what you see is that the narrative has very much changed at, let's say, 80% of the lenders are more talking about efficiency and, you know, how can we, even in an increased interest rate environment, how can we, you know, do more with less as opposed to we wanted to go out and sell 5,000 different mortgages, et cetera. So it became very much a cost case. So that of course means that GenAI and using that as a lever for efficiency is a very hot topic right now.
Mm-hmm.
But I mean, everybody's experimenting now. I think we're still in the cowboy days. You have a lot of consultants, you have a lot of internal experiments. But what I think is a key challenge now is like, how are you going to translate an experiment that yields results into something that you can implement into your business processes to really yield that efficiency case? And when I look at GenAI, okay, building the models, it's contextualizing the models, I think. If you have smart people around, that's, I'm not gonna say easy, but it's doable. But then there are 2 other big challenges and one is compliance. We kind of went, I mean, the cloud providers there are quite helpful, but still it is a lot of navigating about, you know, how do you host models locally? Do you do it in a separate instance, et cetera? So there is some infrastructure work that needs to be done.
Mm-hmm.
And also that's why we build up our own control layer. To make sure that also the model doesn't hallucinate. It sits behind a safe user interface. So it's not some chatbot that can go wild. So there's a lot of safeguards we build in actually from when you use the software in the user interface up until where the data and the models are being stored in the infrastructure layer. So we really did that fully vertically. I think that is one part. And the second thing is we provide it in our software. So we're a B2B SaaS. Meaning that it doesn't have to be integrated into a user interface or an old mainframe application. But I think that's a bit our vision on it. But I can tell you the compliance part and the, let's say, the real getting it to real implementation part is challenging.
I think that the biggest challenge you often hear is traceability. How has that then been tackled within the application?
Yeah, that's why we simplified the process. So I think we That's why we build our own control layer. We really simplify the use cases and we bring the human in the loop. So it's a very good point. What you bring there is that that's also why I personally, that's maybe a provocative statement, I don't really believe in the chatbot use cases. Like sometimes we have clients that come to us and they say, ah, Oper, can't you do a chatbot for a borrower to be advised on a loan? We'll get there, I'm sure. But today, getting it unhallucinating, talking to a client about your mortgage portfolio, I wish you good luck getting that past the regulator with the EU AI Act.
No, no, I agree. And there's also a misconception about a lot of people when you say LLM, they immediately make the link with ChatGPT. But what I hear from you, it's a dedicated system. It's been trained for one single purpose. Which also eliminates a lot of the hallucination per definition already.
Exactly. The metaphor that I like to use, and I don't know if it's accurate, but I mean, I know MS-DOS back in the days. In MS-DOS, you would also type in scripts, etc. And, you know, that's indeed, so that's the first implementation. You basically chat or you basically talk commands to get something back. And I feel prompt engineering in ChatGPT is very similar in that regard. But at a certain point in time, Windows came and Windows basically built a UI layer on top of MS-DOS. So it was actually easier to use, it was more correct, you didn't have to know any script language, et cetera, et cetera. And my prediction is that with vertically integrated AI, we're going to see the same. We're going to see user interface being built on top of the foundation models to make it much more controllable, much more traceable, and basically make it much more error-prone.
Well, talking about how the future looks, could you give us a glimpse a little bit about what's next? In Oper?
After AI?
Yeah.
So, I mean, apart from AI, I think that's a very important pillar for the product development. So I feel around the— I mean, our software is really built around mortgage origination. So we help the whole sales trajectory. So you'll see much more features on guided sales, renovation advice, much more complex loan cases. I think that's Something on the product, you'll see quite some very neat features popping up. And then also on the underwriting part, I think, I mean, now we have the whole document completion and document analysis. We now already have also a prototype that basically prepares an underwriting decision. So it basically writes like a human, hey, I would approve, I wouldn't approve Koen's loan for this reason.
Oh God, there goes my house.
Yeah. And then today, I mean, we have a European focus. So we're currently working with 15 financial institutions in 6 European countries. And we're actually preparing now to get into 27 countries as of next year. So we are fully ramping up for that to be, let's say, make the solution fully EU agnostic and that we can serve any bank that wants to do digital mortgages better.
It's an amazing story. And I'm glad that this is sort of the pinnacle of the conversation because I started with one of the biggest fintechs in Belgium. The fact that you came up with those figures, I think it was just a demonstration of my my my definition in the beginning. I'm sure you there's a lot of people that would like to hear more. Geert, how does that work? How how do they get best in contact with you guys?
Oh, you can always send an email to sales@operacreatives.com. It sounds like a very generic mailbox, but it's very effective to get to us because if you go directly to LinkedIn, you can follow me, but getting a response is going to be slow because I get a bit too much. spamming them. So sales@operacreds.com, I think, is the most direct channel. But also feel free to follow us on LinkedIn, what we're all cooking up, or just go to our website because again, that also gets— I mean, their contact form is also getting followed up quite nicely.
Geert, thank you very much for sharing your story today here in our podcast. And yeah, look out for more news in the future and thank you for joining us.
Thank you, Koen. It was great.
Have a nice day. Thank you also to the listeners. Thank you and stay tuned for more fintech news on this channel. Thank you.
Thanks for listening to another episode of the Connector Podcast. To connect and keep up to date with all the latest, head over to www.jointheconnector.com or hit subscribe via your podcast streaming platform.
