Mátyás Varga, Co-founder and CEO of Capila, builds an AI platform for real estate funds and developers. In a PeerView recorded at EXPO REAL 2026 he discussed with Alexander Morari, MRICS, CEO of Shelton Reed, why the data has to be structured before the AI touches it, how a small pilot opens doors, and why IT security held up so many conversations.
Key takeaway: By Varga's account, the AI does not work on raw data: "we are not processing the data via AI" [03:01]. The data is pre-processed into deterministic tables, and the AI explains and suggests on top of them.
What does Capila build for real estate?
Varga described an AI platform that creates a data layer over a company's data, which the AI can read and work with, and then automates processes on top of it [00:23]. He said the team collects all of a company's data, "not just finance but rent roll", and builds deterministic views over it [03:01].
His own background is in real estate, he said. He worked for a development company and then as CFO for co-working spaces, "10 plus co-working spaces across Europe" [08:07]. He named HB Reavis and its co-working platform HubHub, where he was CFO for two and a half years [08:23].
How does he get AI past cautious real estate decision makers?
Morari put it to him that AI implementation in real estate is one of the hardest fortresses to break through [01:16]. Varga agreed it is not easy, and said it was hard to start with bigger real estate funds and development companies [01:48].
His approach is to start small. "Let's start with a small MVP, even for free, I'm okay with that one, and we can showcase you the value" [01:48]. In finance, he argued, "you don't need some BI reports, some big excels, etc. You need answers", and once an AI chatbot gives those answers, the client implements the solution [01:48].
How does he keep AI answers deterministic?
Morari raised the concern that financial work in real estate cannot tolerate non-deterministic answers [02:36]. Varga's answer was the order of work:
- The team builds a database of the client's data and creates deterministic tables over it [03:01].
- The AI receives those tables and provides "the explanation and some suggestions" [03:01].
- Security levels decide which data the AI sees, so sensitive data is not fed to it [03:42].
He added that the team uses many model providers, with controlling mechanisms between the models, so that one model checks another, and that he spends three to four hours a day on research to keep up [04:03].
What has Capila delivered so far?
Varga said Capila has paying customers [05:05]. He did not name them, saying he would have to check the NDA, and described them as "the biggest real estate funds, developers, etc. in Slovakia" [05:16]. These claims are his own and unverified.
The work so far is financial and wider data dashboards, which he called a modern application built from scratch, "not BI tools because BI tools are, I think, totally outdated" [05:16]. An AI chatbot sits in chat platforms such as Microsoft Teams and Slack, so users can talk to their data and request presentations or spreadsheets [05:16].
Why does he say SaaS is dead?
Front ends can be built quickly, Varga said, so buying a big SaaS tool is not the right way to think any more: "I think SaaS is dead, sorry, but that's the future" [06:47]. His reasoning was that the tool should fit the processes, not the other way round, and that a front end takes "a matter of weeks" [06:47].
Building the data layer, not the front end, is the core of the business, he said [06:33].
Which sectors are open to AI?
"Real estate is one of the hardest" sectors to work with, Varga said, and the easiest are IT companies, which "are just thinking the same as we" [07:30]. He placed finance, insurance and banks among the hardest as well [07:33].
Real estate is slow to enter, he explained, because many people still expect "new IT team to implement in three years, budget of 20 million", which he called untrue today: "today it can be weeks and just a fraction of the budget" [08:42, 09:10].
Beyond finance, Capila works on leasing, extracting data from lease agreements and building rent rolls, and on marketing [09:28]. The goal, he said, is to put all of a company's data in one place and connect it, including external market data [09:28].
Why was IT security the biggest blocker?
The biggest worry for clients, Varga said, is that the AI sees their data and someone else can prompt it, with the data sitting in a data center "in the US or god forbid in Asia" [10:18]. His answers are data centers in Europe and different security levels [10:18].
The team spent the last three months on it: "we have been really focusing on the it security because it was a big blocker" [10:18]. A CFO would like the idea, he said, and then hand it to the IT team, where discussions ran for months [10:18].
Morari closed the conversation by saying AI will develop in real estate and that IT and data security remain a priority [11:17].
FAQ
How does Capila keep AI answers reliable?
Varga said the data is pre-processed into deterministic tables and the AI only explains and suggests on top of them, with security levels controlling what it sees [03:01, 03:42].
How does he win funds and developers as clients?
He starts with a small pilot, "even for free", to show the value before a rollout [01:48].
What does he see as the main blocker?
IT security: clients worry about where the data is processed, and IT teams can take months to verify a solution [10:18].
Watch the PeerView.