AI in Action: How Stonal is Shaping the Future of Real Estate

In the rapidly evolving world of technology, AI continues to be a game-changer across various industries. In this interview, we explore how advancements in AI, from machine learning to generative models, are impacting the real estate sector. Robin Rivaton, CEO at Stonal, shares valuable insights into the practical applications of these technologies and their potential to transform data management and operational efficiency.

Tell me about your career history and what inspired you to establish Stonal…

Stonal was founded in 2017, by Micheal Tolila and Jean-Maurice Oudot, both of whom have deep roots in the real estate industry, and our platform is designed to support asset managers and owners by transforming how they manage data. I joined in 2022, after working as a VC investor focused on PropTech, smart energy, mobility and logistics. I’ve always been passionate about AI, and alongside my work, I also write for a French Magazine, L’Express, discussing tech topics weekly, and have authored several books on the subject.

What we’ve seen in real estate is that data collection is often still done in old-fashioned ways – sending people on-site to gather information for inspections, certificates, and surveys. It’s costly, and once the data is collected, it’s usually stored in documents that meet compliance requirements but don’t add much value beyond that.

At Stonal, we’re using machine learning to change that. Our technology can categorise and rename documents, extract useful metadata – like names, prices, and equipment – and turn that data into objects for analysis. This allows us to do things like compliance checks, identify missing or invalid documents, and verify signatures.

What sets us apart from general tech providers is our focus on real estate-specific needs. Our AI is tailored to the industry, giving us an edge. When we first started, we were converting blueprints into simple BIM models for building maintenance. But now, we’ve evolved to focus more on machine learning. Our AI is highly accurate in French and English, but when we enter new countries with different languages and legal systems, it takes time to train the model. Initially, accuracy can be low, but with the help of humans, we train the AI until it reaches over 90%.

I like to think of our approach as a collaboration between machines and humans. We don’t aim for 100% accuracy because the cost is too high and not always necessary. Instead, we rely on machine learning for large-scale data extractions, while humans step in for more complex documents. The machine is great at being transparent – it tells us exactly how confident it is about the data, so we know where human input is needed.

How did you find your focus more on the property side?

When I was a VC, I had a deep understanding of the real estate industry. As I mentioned, I’ve written several books on the subject and was involved in the public housing policy. I also serve on the board of some large property firms in France, so I’ve always been well-versed in the property and asset management sector. My passion has always been at the intersection of property and technology, so when I met the founders of Stonal, I was excited to learn more.

Looking back at my portfolio, I’m proud that all my startups survived – it’s a rewarding job where you meet all kinds of people. But I started to feel that something was missing. I wanted to build something of my own again, and when the opportunity at Stonal came up at the end of 2021, it felt like the right time to dive back into entrepreneurship.

The biggest difference between being a VC and an entrepreneur is how you manage your emotions. As a VC, even if you lose a deal or a portfolio company underperforms, it doesn’t really affect your day – its just part of the job. But as an entrepreneur, it’s a constant rollercoaster. You might sign a big contract in the morning, but by evening, you find out the product is facing issues or the roadmap is delayed. You have to learn to balance those highs and lows.

It’s quite common to classify “AI” as a buzzword in today’s tech climate. How would you define AI and its specific applications in the world of real estate?

I categorise AI into three main types. The first is machine learning, which is a powerful technology but also the most difficult to implement. In machine learning, the model itself isn’t the main challenge – it’s the training dataset that matters most. Getting access to enough documents to properly train the model is tough, especially when large asset managers are hesitant to share their data. This creates a catch-22: low accuracy at first because of limited data, and limited data because the model isn’t yet accurate enough to gain trust. Finding useful datasets online is also very difficult, which makes machine learning one of the most complicated areas of AI to implement.

The second type is predictive modelling. It’s very popular because it sounds exciting – everyone wants it. But it’s also incredibly complex because there are countless methods to create these models. The issue is that past data doesn’t always accurately predict future outcomes, so you end up making a lot of assumptions, and if those assumptions are wrong, the model becomes unreliable.

The third type is generative AI, which is great for requesting data, creating workflows, or training specific routines with large language models. It’s exciting and has a lot of potential, but it’s also costly in terms of inference and carbon emissions, sometimes for tasks that aren’t very meaningful. While Gen AI is promising, it feels like its more of a tool for tomorrow than today.

What’s unfortunate, in my opinion, is that machine learning hasn’t become more widespread. If we want to create accurate predictive models or build AI systems that can effectively use data, we need to start collecting that data, and machine learning is key to that process. However, many people aren’t ready to invest in or adopt these technologies, so we often focus on future possibilities without addressing present needs.

What are the most pressing challenges in the real estate industry that AI technologies, like Stonal, are uniquely positioned to solve?

One of the most significant challenges in the real estate industry is the lack of high-quality, actionable data to enhance building management and decision-making. Traditional data collection methods—often involving time-consuming, on-site inspections—are inefficient and outdated. AI technologies, such as those developed by Stonal, are uniquely positioned to solve these issues by automating data extraction and document classification.

For instance, asset owners routinely spend significant amounts annually per building to obtain critical reports, such as fire safety inspections, asbestos surveys, and energy performance certificates. Historically, this data has been unstructured and underutilised. Machine learning and generative AI now enable efficient processing and transformation of these documents into structured, actionable insights. Stonal has also introduced an innovative feature that converts Revit BIM models into databases. This approach unlocks the latent potential of BIM models, often underutilised after construction or retrofitting projects, turning them into valuable resources for asset management.

Additionally, Stonal leverages predictive AI to create highly accurate multiyear CAPEX plans, helping clients navigate competing constraints and make informed decisions about resource allocation.

Looking ahead, what do you foresee as the next big advancements in AI for real estate? What are Stonal’s growth plans over the next year for technology improvements?

Generative AI is at the forefront of technological innovation, and their applications in real estate are transformative. At Stonal, the introduction of StonalGPT enables natural language interactions, streamlining how users access and process information. Another exciting development is the use of autonomous agents, which can independently execute tasks based on their environment and objectives. For example, once a CAPEX plan is validated, these agents can source and compare quotes from local tradespeople, challenge them, and establish a comprehensive, actionable plan.

Looking forward, the key to fully harnessing these advanced AI solutions lies in the quality of the data. Without a robust data foundation, even the most sophisticated AI systems cannot deliver optimal results. So, before diving into advance AI solutions, make sure you have the right data in place – don’t put the cart before the horse.

If you would like to discuss any of the topics raised in this piece or if you need support with your leadership resourcing strategy, please get in touch with Lucy Wright on: lucy.wright@beaumontbailey.com.