As we are approaching the London PropTech Show 2025 in less than two months, the excitement is palpable. In this episode of "PropTech Power Talks," we delve into the Maersk real estate portfolio and the future of AI and agentic workflows in the industry.
Jacob is an integral part of the Maersk real estate team, focusing predominantly on logistics real estate and overseeing global capital market initiatives. His expertise in integrating AI into real estate processes, particularly on the industrial side, provided us with valuable insights.
Q: To start off, could you share a bit about your background and what you're currently responsible for at Maersk?
A: I have a classic finance background. After a number of years in global private equity, I transitioned to institutional real estate with a focus on international and global real estate. Currently, I am responsible for capital market initiatives at Maersk. At Maersk, we are mostly on the occupier side, but we also have several capital market initiatives in our portfolio, such as sales, leasebacks, divestments, and equity stack replacements. Our portfolio encompasses approximately a thousand sites globally, offering plenty of opportunities to work on.
Q: Let's delve into the current state of generative AI adoption in the real estate sector. How are we seeing it impact assets, investments, and capital markets?
A: I have been following this space for quite some time now, both in commercial real estate and AI. There's still a noticeable lag, though. The current state of generative AI is very fast-paced. While the interest in commercial real estate is certainly there, especially among advisors, investors, and brokers, finding crossovers where technology can prove its worth in a real estate setting is still a challenge. The interest currently surpasses the promises. However, I hope that in the coming years, the use of generative AI will be further integrated into the industry more broadly.
Q: Do you think the adoption of generative AI in real estate is picking up pace or hitting a plateau?
A: It's a bit of both. Many companies have implemented models like Microsoft Copilot and similar tools, but their one-way usage limits their usability within a real estate setting, given it's a very physical asset. Nevertheless, AI is accelerating, and almost every real estate company is beginning to explore its capabilities. AI as a copilot is somewhat limited, but integrating this technology as an assistant or an AI agent in many of today's tasks will likely lead to better outputs. Many SaaS providers are now offering generative AI products catering to the real estate sector. However, they are still primarily SaaS platforms. Companies can integrate this technology within their real estate portfolios to truly benefit from it.
Q: Could you elaborate on the main benefits that AI brings to real estate operations?
A: I see it in terms of usability. At the moment, there's a lot of discussion about predictive maintenance where AI has a significant benefit. There's also a lot of talk about document management, essentially data management.
Additionally, moving forward, AI in marketing might not be as relevant to many real estate functions, but it plays a role as a real estate researcher. Let's take it step by step, starting with the data side. Data is one of the areas within real estate where we typically have a low amount. We have several data points, but the overall dataset is very small compared to many other industries.
The data within this industry is often tucked away in various storage databases. However, with the right use of AI, we can bring this data forward and analyze it in completely new ways. The data in real estate is often unstructured. There are two key components: financial data and legal data. They are often in separate formats, but there's a lot of overlap where AI can interact with and analyze these data sources. AI research is one of the newer inventions and essentially involves an AI entity or agent looking through complex data, finding inconsistencies, and tracking data development.
This is how you can set up an AI researcher or agent. AI is particularly interesting in predictive analytics. Traditional machine learning has done excellent work in predictive forecasting, but in complex scenarios like weather and financial markets, generative AI is catching up with traditional machine learning. This development is very interesting and could provide an additional tool for real estate.
For example, how would rent develop in a certain market? How would the yield potentially develop in a micro-market? And from an occupier's perspective, what is the availability and suitability for my business? For instance, in MAERSK's case as an occupier of logistics, what are the best-suited locations? AI can help answer these complex questions and provide guidance on the direction to take.
Finally, I would mention marketing tools. This might not be where I know the most about generative AI, but it's an area with a lot of excitement, especially in text-to-image and text-to-video applications. This whole area of generative AI seems very promising in the near term and the medium to long term.
Q: We'd really like to know how Maersk is specifically implementing AI in its real estate portfolio.
A: I'd be happy to dig into the details. As an occupier, we have different demands for real estate, particularly in terms of space usage for our operations, which translates into our global portfolio.
It's a massive portfolio, a huge database of contracts. While there are similarities in these contracts, they also vary significantly across the globe. We can use generative technology to better understand the data within these lease contracts. One of the key projects we're currently running at Maersk involves extracting and analyzing this data in a more coherent way. We have a global portfolio, which presents several challenges, including data extraction and the actual extraction of data points. One of the key findings is the issue of languages.
While English and other top global languages are covered by many models, we've been pleasantly surprised by the technology's basic understanding of even smaller languages. Depending on the level of data you want extracted from leases, you can use it as a first step in the extraction process. Even in local languages, it seems to find many of the data points we're looking for. This has transformed the data extraction exercise from development to truly showing its capability. It's not perfect—we're still struggling with issues like hallucinations—but if you set it up correctly and have a good sense of the data you're looking for, it's a very handy tool. This is one of the key focus areas for Maersk at the moment in our real estate portfolio.
Q: What about the types of AI models that Maersk is using? Are they text models or vision models?
A: We use most of the models available in the field today, including traditional machine learning and OCR. We have a lot of SAP and SaaS applications to assist us. Internally, within generative AI, we've built a sandbox environment to access various generative models. New and more promising models are introduced almost weekly. We use the most popular models daily, and we also try some lesser-known ones. However, the most important factor for Maersk is data security. Ensuring no data leakage or jailbreaking of models is still an issue, so we use AI internally for internal purposes to ensure no bad-faith actors are involved.
To answer your question, we use and test most available models. Regarding SaaS providers, I'd say Microsoft Copilot is one of the key tools for most employees today. We also use GitHub Copilot and many other generative AI tools in our toolbox. This will continue to develop over the coming years.
Q: Yeah. That's interesting. As you talked about data, how does Maersk handle the storage and management of its real estate data?
A: It's such an interesting question, particularly in real estate where there has been a lot of focus on data for many years. Given the wide range of financial data, legal data, and operational data, these are typically stored in various places and at different levels. The key focus going forward is really how to integrate them. There are many startups trying to help companies with this, and they're doing an excellent job.
We are trying to aggregate or collect many of these important data points in a data lake environment and get them to interact with each other. The best example, probably a few years into the future, is receiving an invoice for one of our properties, automatically booking it in our accounting, and matching it with the actual lease agreement in place for that property. This way, we can see that this invoice has been agreed upon in the lease agreement.
In terms of storage, we have a long-standing agreement with Microsoft, which has helped us secure our data storage environment. However, as a global company like Maersk, it is still very siloed. We are trying to find common layers where edge AI can assist us in analyzing and solving these data issues.
Q: What would you say are the key takeaways from your experience with AI and what you see as the next steps?
A: It's been a hectic few years following the AI space. Being from a commercial real estate background, many concepts within machine learning and generative AI are still very new to me. Even machine learning professionals and data scientists find it challenging to stay updated with the rapid developments. Being updated is probably the biggest hurdle at the moment, even for those closely following the advancements.
The key takeaway is how we, as a company and as every real estate company, can improve data quality and ensure the data is correct. Sorting data properly is crucial because AI can capture so much value. If your data is sorted, analyzed, and classified correctly, you can use generative AI in a completely new manner.
There are two main takeaways:
- Evaluating language models: New models are usually compared or benchmarked against their peers or recent benchmarks. This is a good starting point, but focusing on benchmarks relevant to real estate, such as legal or economic questions, will reveal a much wider discrepancy between models. Benchmarking output going forward is a very interesting field to follow within real estate.
- Return on investment: Most current ROI cases are indirect, focusing on risk management and better access to data. However, the actual ROI is difficult to calculate one-to-one. I believe this will change soon with the development of reasoning models and language models. We are heading to a place where it won't make sense not to embrace this technology. It will be a requirement to use these tools within the next few years to capture all the value from owning and managing real estate.
Q: With the advancements in AI lately, where do you envision further AI adoption in real estate in 2025?
A: I would say the low-hanging fruit, which I might have mentioned briefly, is data management. We are working on lease extraction and data extraction from key documents in real estate, such as SPA and even font-level documentation. Another really interesting area moving forward is predictive maintenance. This area has developed quite a bit over the last few years. With the increasing capabilities of vision models and growing awareness of how we can potentially predict maintenance, this area is becoming very exciting. Predictive maintenance outside of real estate is almost a given in industries like aviation because it makes so much sense. However, it hasn't been shown as effectively in real estate yet.
You hear stories of asset managers dealing with roofs close to collapsing, where earlier intervention could have significantly reduced capex costs. AI would clearly help companies improve predictive maintenance. This also aligns with sustainability. When we start accessing more electricity data, supplier data, and measurements like gas and water usage, we will be able to measure the optimal state of a property more accurately. With the right data inflow, it's fairly easy to set up an AI agent or workflow to capture anomalies, such as unusual electricity consumption on a Sunday. Streamlining and automating this can only benefit the real estate sector.
So, predictive analytics in terms of maintenance, lease management, and data analysis within the real estate industry's data are crucial. Operational efficiency, benefiting both occupiers and asset owners, is also key. I'm particularly focused on the AI researcher concept, which is still a bit early but will become more relevant around 2026-2027. On a personal note, keeping up with new model releases is a challenge. Recently, a few Chinese models have been released, and they seem very capable in reasoning and usability. This will be very interesting to follow in the near term.
Q: Lastly, could you share your thoughts on the upcoming London PropTech Show and its significance for the industry?
A: The PropTech Show is one of the core events you want to attend in Europe for the PropTech scene. It captures the overlay of commercial real estate, AI, startups, and technology. It's a place where interested parties can see what's next and understand the trends in the real estate space that we're all passionate about.
Thank you for diving into this insightful conversation with Jacob Nielsen on PropTech Power Talks. As AI continues to reshape the landscape of real estate, we hope these discussions inspire you to explore and innovate within your own ventures. Stay tuned for more thought-provoking insights and trends in the world of PropTech. Until next time, keep pushing the boundaries of what's possible!
To watch the podcast, click: https://www.youtube.com/watch?v=Y-wtdJtOP8U
