AI Insight roundtable backs smarter data use

Firms that don’t embrace AI technology are in danger of being left behind as competitors capitalise on the benefits it presents to clients and tenants. This year’s Mipim convention in Cannes, France, provided the backdrop for a gathering of senior real estate technology innovators and users to discuss how AI adoption is already affecting their businesses, and the challenges and opportunities in maximising the potential from future developments in this area.

The discussion formed part of Property Week‘s ongoing AI Insight campaign, which aims to raise awareness of both the opportunities and risks posed by AI.

The roundtable took place over a three-course lunch at La Cantine Provençale, supported by our campaign sponsors –real estate sustainability software-as-a-service (SaaS) platform Deepki and data-driven intelligence specialist Nimbus. The conversation flowed, with all participants eager to learn from each other’s experiences.

All agreed just how much and how quickly AI had improved, even in recent months.

Faisal Butt, founder and managing partner of Pi Labs, a venture capital firm that invests in technology for real estate and real assets, explained how the shift over the past two years to large language models (LLMs), such as ChatGPT and Google Gemini, had exponentially increased the development of specific use case solutions.

According to Dan Hughes, founder and chief executive of Alpha Property Insight and founder of the Real Estate Data Foundation, 2026 is likely to prove “a year of AI hype”, and organisations and individuals are jumping into using AI without thinking about what they actually want to achieve from it. However, this is not necessarily a bad thing, he said.

“By trying it [AI], you find out where the problems are, and once you’re aware of those, you can get the benefits of it,” he explained. “Culturally, we’re often told that we [the property sector] are slow and counter-cyclical and low risk – that’s exactly what we should be for buildings, but it’s exactly what we shouldn’t be for technology.”

This cultural mindset and the fact existing data across the sector is vertically siloed and not joined up are the two main hurdles blocking the progress of AI use in the industry, Hughes added.

Referring to the cultural mindset challenge, Nimbus co-chief executive Paul Davis pointed out that there was a big difference in the industry’s approach to AI compared with other forms of technology in the past. “There’s an interesting dynamic at play because people in their personal lives are very much on AI; therefore, that adoption in the work environment becomes much easier,” he said.

The data challenge

In response to the data hurdles that Hughes identified, Deepki chief product officer Olivier Pin highlighted that the consistent use of AI required not only the right data but the correct handling of that data.

“If you want to be able to use AI consistently, then you have to have the right level of data, and that means also structuring how you’re going to store this data, how you’re going to manage it in the long run,” he said. “There will only be a few major algorithms, so it is data quality and industry knowledge that will make all the difference.”

The roundtable explored the issue of data quality at length. According to Stacey Patten, head of residential investments, Europe, at investment management firm Invesco, the company – which has hundreds of residential assets – has spent the past year focusing on this. “We can’t do anything with the information until we trust it,” she said, adding: “The benefit of the AI tools that we do have is that it can go in and highlight the inconsistencies, so it takes far less time to check that the data might be incorrect.”

Richard Williams, chief executive of real estate asset management company Ringley Group, said the firm was also “spending a lot of time making sure our data is correct”.

He added: “A lot of our buildings are old and [so are] the outputs they have – if they even have them. In some cases, a BMS [building management system] that was put in 10 years ago was probably 15 or maybe 20 years old – trying to connect with that to get any useful information out of it and to then interpret it via AI is inherently probably incorrect.”

The data challenge goes “all the way back through the supply chain”, said Chris Holmes, founder and chief executive of HermeticaBlack, an energy and infrastructure asset developer and manager. “How do you bridge that gap between construction firms and developers and all the technology that could be available?” he asked.

Speaking about LaSalle Investment Management’s efforts to maximise use of its property data, the firm’s head of sustainability for Europe, Brett Ormrod, explained how it was ensuring consistency with measurement surveys across existing buildings – using International Property Measurement Standards 1 and 2 for external and internal measurements, respectively.

David Maguire, global head of product, sustainability, at CBRE, said the firm was already benefiting from AI’s rapidly improving ability to recognise and complete patterns. “We’re able to take data that we previously couldn’t have used, or that would have prevented us from making decisions or insight – pulling in other data sources like climate information, weather trends, other building characteristics – and actually making decisions around BMS operations or design based on less-good-quality data than we had before,” he said.

Maguire stressed that this was, naturally, checked by the firm’s advisory teams, but he said the accuracy was impressive. “If you’d asked me five years ago ‘what’s the priority?’ I’d have said ‘data, data, data’. I think now it’s insights and action based on the data we have.”

According to Butt, this is a pattern he is seeing with the property portfolio companies that Pi Labs works with. “To train a new AI product, you need 10 or 100 times the amount of data they have access to, so they’re using AI to create synthetic data, which mimics the data they have,” he said.

Clear use cases

Pin identified two clear use case classes for AI in the industry: automated hardware producing data that can be governed automatically; and decision-helpers that digest data, make recommendations and sometimes also automate some of the actions. For Williams, AI use can be put into two “buckets”: operational efficiency and strategic assistance or, as Davis summarised, “saving money and making money”.

Hughes said the key to choosing the correct AI tools was to “get better at understanding what we’re trying to achieve”.

He added: “If you’re looking at an office building, you want it to attract more rents, get higher value. You want the people in it to be productive, you want them to be happy, to stay there. You want them to be sustainable and financially friendly. You want it to be performing well. Some people want prestige around it. All those things haven’t really changed.

“The question is: how do you achieve it? And that’s where I think we work backwards to start looking at how we then achieve those things.”

Ormrod explained how LaSalle Investment Management undertook a comprehensive net zero energy audit campaign and concluded that “the lowest no-cost [to tenants] intervention we could make is actually just getting our buildings to work more efficiently”, but that there was a “massive gap” between perceived and actual performance.

To address this, over the past 18 months the company scanned its 300-plus buildings across Europe for the correct BMS that could work with AI systems. As a result, it now has 12 BMS AI projects live with another seven in the learning phase. He said this year the firm hopes this will save more than 2,500 tonnes of carbon dioxide and over £200,000 in energy savings, most of which will go to tenants.

At asset level, Ormrod said, “we expect energy savings of between 10% and 35% across heating, cooling and ventilation”.

The experts around the table reported that AI tools were enabling better use of a number of existing technologies.

Maguire cited building information modelling (BIM) and digital twins as two good examples. He recalled both being “a very noisy part of the industry going back five, six years”, but that adoption was generally quite low. AI, he said, is now enabling the outputs from such technology to be used more easily, effectively
and quickly.

Butt talked about AI technology that compares virtual “as-built” models of what is happening on site with the BIM model to deliver a “quality assurance function”.

This allows developers to assess if their building is being built the way it is supposed to be and to detect any defects quickly.

In a similar way, better use of SaaS tools is being enabled by AI, Holmes said. He also highlighted how AI can help lower the costs related to infrastructure and onsite enabling works by identifying potential ground risk and utility diversions.

Patten added that she was excited about the predictive ability of AI enabling further operational benefits in the future, citing the Renters’ Rights Act coming into force in May as an opportunity. “Wouldn’t it be great if you could predict when your tenants might move out, depending on how long they’ve been there, depending on what their rent is versus the market?” she said.

“AI will give you the ability to use LLMs to make better insight decisions to allow you to then be able to solve a lull or void in your building. There are so many use cases for what the technology can do and that’s just scratching the surface of one subsector.”

It is these use-case-based applications that are driving take-up and progress of AI technology generally and across the industry. Butt likened this evolution to the way that apps such as Uber and Airbnb were developed to use Apple’s iOS platform for iPhone.

Choosing the right tools for the right reasons, however, is just one part of the AI puzzle, Hughes cautioned, noting that it is essential AI is applied “in a sensible way with the right guardrails and the right governance in place”.

According to Williams, Ringley ensures that reports with AI input are read and checked by three people before they are sent out. Pin questioned whether this could lead to “supervision fatigue” and erosion of trust in teams.

“I see a lot of executives saying ‘yes, it’s saving me some time, but in the end it doesn’t save that much time because I’m doing so much supervision; I’m second-guessing everything that’s being built by the people I manage, because I’m wondering if they build it themselves or they just use AI’,” he said. “This is the importance of domain-trained AI: technology built and fed by industry expertise to ensure truly trustworthy, actionable output.”

Davis agreed: “In today’s mindset you go on the assumption that it [the AI output] is wrong […] but when you find the output is good, you have to retain that scepticism to check that you don’t make mistakes yourself.”

Tech oversight

Getting the level of oversight is easier at smaller companies, Holmes said, adding that at HermeticaBlack, ownership is encouraged at multiple levels so that not everything falls to the senior leadership team.

Meanwhile, Maguire said working at a large business meant all AI-related policies and protocols were set by central digital, tech and legal teams, which can be frustrating for those eager to use the newest tools. He also highlighted an upcoming “huge problem in how we train and upskill the next generation of professionals” when the entry-level work is automated.

William and Butt countered that different types of skills would be needed. While Williams said Ringley had increased headcount on the tech side to drive AI, Butt said many of the landlords and investors that invested in Pi Labs’ fund were seriously thinking about having a non-voting AI as a member of their investment committee. “It’s the smartest ‘person’ in the room, so why wouldn’t you want to have that AI as your intellectual sparring partner on whether you should do this deal or not?” he said.

As to the dangers of historical human biases influencing AI-assisted decisions, Davis pointed out that the evolution of quantum computing would enable AI to draw on deeper information and said he believed any inherent bias would “slowly disappear over time”.

Another problem that could emerge, however, is legal responsibility if AI recommendations get things wrong. Everyone agreed it was only a matter of time before this would be tested in court.

However, Davis was confident the evolution of LLMs would, over time, fix issues related to ‘hallucinations’ as users become more adept at using AI to check the accuracy of the data it provides. In fact, AI will ensure there are fewer mistakes, he said, citing driverless cars as an example. “In these conversations, you’re almost assuming people don’t make mistakes and [questioning whether] AI can catch up; but obviously in the case of driving, we all know crashes happen and people aren’t perfect and therefore the AI is now overtaking or has overtaken us,” he explained.

Despite the unanswered questions, a comment from Maguire summed up the sentiment around the future impact of AI in the industry: “The transformation that AI will drive in our business is unbelievable. We will be unrecognisable as a business within five years.”