Generative AI models may attract the most attention in realms like marketing automation and financial technology. But a recent analysis by A/O PropTech reports that investments in built-world AI start-ups are growing faster both in terms of the volume and value of investments. These investments have seen a significant uptick with efforts to prepare for new sustainability mandates.
As the name suggests, A/O PropTech is a venture capital investment firm specializing in the built environment. For the moment built-world, AI is still in its early days, owing to the predominance of unstructured data that is harder to calibrate and align than other domains. One additional caveat is that A/O PropTech takes a wide view of the built world, including real estate management and insurance of physical assets that others might lump into financial services.
Climate technology is the fastest-growing AI application, including climate risk, ESG reporting, and energy management. The largest deals are focused on the most mature segments, including real estate transactions, property management, and construction.
Getting beyond the AI wash
Catriona Hyland, research analyst at A/O PropTech, said they initiated the recent study to see how the recent hype in generative AI was impacting built-world startups. For now, generative AI has not taken off in the sector owing to the numerous challenges in making sense of unstructured data. In the short run, generative AI may play a much more significant role in augmenting data sets to provide a foundation for other tools.
She said they tried to tease out some of the issues of AI wash that plagued previous research into AI startups. A 2019 study by MMC found that 40% of startups reportedly using AI did not. Further analysis revealed this was primarily due to third parties touting the AI capabilities that the firms never claimed nor corrected. Her team developed a large language model for analyzing data from Pitchbook and Crunchbase for the analysis. Hyland explained:
I think quite often AI is used as a kind of buzzword when you have a startup where AI forms the core component of that business. But I think in the built world, and probably in most industries, it is used very much as a tool rather than as the kind of centerpiece.
London leads deals
The study found that over the past ten years, AI-enabled Built World startups in Europe and North America have received $18.6B in venture funding, of which close to half was in the last two years ($8.6B). And in both 2020 and 2022, venture deals in AI-enabled Built World startups overtook FinTech AI funding, reaching over 600 deals globally in 2021 alone.
Also, London saw the highest number of deals, while the San Francisco Bay Area saw more capital. London also saw more deals than Paris, Berlin, Dublin, and Tel Aviv combined. When asked why Hyland explained:
When we looked at the breakdown of the types of deals that were happening in London, a lot of them were quite early stage and focused a lot more on areas like real estate transactions and the financial aspects of the built world. This makes sense when considering London’s position within Europe as a kind of financial hub.
Structuring the unstructured
In the short run, they expect more of a focus on using AI to make sense of structured data rather than generating new building designs. Innovations in computer vision are proving incredibly important for scanning construction sites, tracking progress, and automating insurance claims processing. But other kinds of generative AI are still in the early stages.
Jess Clemans, Investor at A/O PropTech, explained:
The innovation around the large language models and image generation models was hyped in media quite an awful lot, but we haven’t seen a lot of that filter into real estate technology just yet. I think we’re on the cusp of a lot of that being utilized, but it’s the very early days of this. What we’ve seen is a lot of other innovative AI go into built-world technology, but we’re interested to see where this is going and try and get a better picture of how we think these new innovations are going to feature in the next generation of technology.
What’s challenging is that buildings fundamentally abide by the laws of physics and have a lot of regulation and logic around how they are structured and how they have to be built. You can’t leave it entirely up to a machine to determine the output. We’ve seen companies using this approach. And the output was often very illogical bathrooms attached to kitchens, hallways that don’t lead anywhere or windows not connected to bedrooms.
What we’ve seen as a slightly more successful approach in this space is mixing generative AI models with human readable rule systems that the AI has to comply with. So, I think we’re going to see a bit of a merging of algorithmic approaches with generative AI approaches in the space.
One big challenge is that developers are still trying to figure out how to specify things like construction processes and building codes in a way that AI can understand. Today most construction data is stored in paper records or PDF documents. Considerable work will be required to catch up. In addition, building codes can vary between cities and countries. Technical and subjective aspects need to be considered as to what local governments will approve.
There’s no actual structured rule system. This is in such an infancy that every company we’ve seen doing this has invented its own system for assigning rules to elements. And it’s generally specific. So, we’ve seen companies doing this for electrical or plumbing system design. And they’ve had to attach it in a slightly different way to someone doing architectural floor plans or architectural detail design.
But then, the final piece of the puzzle that’s really interesting and intricate is basic human logic. For instance, if you decide to place an electrical railing in the ceiling, you might not want to attach it directly to the edge of the wall because fixing screws is challenging for a human hand to get in there. But a machine would never understand that. That’s just an example of some of the finer intricacies that are not part of building code standards that are not part of typical construction documents, but a part of human logic that has to be translated into this going forward.
The generative aspects of AI are getting all the press this year. But better data translations and alignment may provide more value in the short run. Improving workflows and processes for building and operating physical infrastructure will require finding better ways to make sense of data captured for other reasons.
The recent crop of generative AI applications arose from Google’s efforts to build a better translator between French and English. These transformer models could also be important in translating documents, designs, and 3D data captures into digital twins.
The UK was one of the innovators in building information modeling (BIM) technology for organizing data about the built environment. It will be interesting to see how this lead may pan out with the UK government’s ten-year plan to make the UK a global AI superpower.