Special Report

This article was published in our October 2026 magazine

European real estate no longer needs to be persuaded that artificial intelligence matters. The evidence suggests that the industry has already moved beyond that argument. JLL’s 2025 Global Real Estate Technology Survey found that 88% of investors had begun piloting AI and were pursuing an average of five use cases at once. In 2023, fewer than 5% had started pilots.

Yet the same research exposed a gulf between activity and achievement: only 5% of occupiers said they had met all their programme goals, while more than 60% of investors remained strategically, organisationally or technically unprepared to scale AI beyond experiments.

That gap - between using AI and building a business around it - is the defining feature of adoption in 2026. It explains why AI can feel entirely normal at desk level while still looking immature when viewed across an organisation.

Employees draft, search, summarise and analyse with generative tools every day. Far fewer businesses have clean and connected data, agreed governance, redesigned workflows, reliable measurement and enough people who understand both property and technology.

Beyond real estate

The contrast is not unique to real estate. Eurostat says 20% of EU enterprises with at least 10 employees used AI in 2025, up sharply from 13.5% in 2024 and 7.7% in 2021. Size matters: 55% of large EU enterprises were users.

Property’s largest investors therefore sit inside a much broader corporate acceleration, but the sector brings particular complications. Real estate information is often scattered through leases, valuation files, building systems, emails and spreadsheets. Assets are local, transactions are infrequent, outcomes take years to observe and apparently similar buildings can behave very differently.

RICS research reveals an equally uneven picture. Its 2026 global study of commercial property and construction companies found that around 37% of commercial property respondents planned a moderate or significant rise in AI investment, but only about 4% reported widespread or full integration. In construction and project management, investment intentions reached roughly 47%, compared with widespread use of about 6%.

Professionals’ experiences vary in scale and sophistication, but the central conclusion is strikingly consistent. Access to a powerful model is becoming ordinary. Competitive advantage comes from the information supplied to it, the workflow around it, the questions asked of it and the quality of the people responsible for the result.

The first gains are easy to understand because they are measured in time. James Cooper, global head of analytics engineering and AI at CBRE Investment Management, says AI has reduced the time needed to synthesise leasing data by about 25%.

“We use AI across a range of activities, from analysing large volumes of market data and news flow to extracting information from leases and supporting portfolio analysis,” he says. “The goal is not simply to automate tasks, but to help our professionals spend less time gathering and organising information and more time applying their expertise, judgement and market knowledge.”

Saving time

At PATRIZIA, meanwhile, Dr Marcelo Cajias says activities that once took weeks can now be completed in a fraction of the time, with more than 120 AI-driven workflows supporting internal processes. Jamie Ley, managing director of communications agency Forma Global, adds that some research and briefing work that previously occupied several hours can now be completed in 15 minutes.

Academic evidence makes such claims plausible, while cautioning against treating every impressive result as universally transferable. In an experiment involving 453 college-educated professionals completing writing tasks, Shakked Noy and Whitney Zhang found that ChatGPT cut average completion time by 40% and increased assessed quality by 18%. A separate field study of more than 5,000 customer-support agents, published in the Quarterly Journal of Economics, found a 15% average increase in issues resolved per hour. Gains were much larger for less experienced workers and smaller for the most skilled.

Those studies did not examine investment managers, surveyors or asset managers, so they cannot be pasted directly onto real estate. They do, however, support two themes running through the interviews conducted for this report. Generative AI can shorten information-heavy tasks substantially and its effect differs according to the user. The second point matters because businesses commonly report adoption as though providing a licence meant providing a capability.

Gains differ between users

Jenny Wang, COO and head of asset management at Sunrise Real Estate, sees the difference at close quarters. “I am often surprised by how personal AI adoption is,” she says. “Two people in the same company, doing very similar jobs, can be using AI in completely different ways depending on their level of curiosity, understanding and even friend circles.

“Everyone has access to ChatGPT, Copilot, Claude and so on, but access to the tools is only a small part of the equation. The bigger difference comes from how people use them, how well the tools are integrated into workflows and whether the business gives people the right structure and guidance.” Successful adoption, in her view, is more a management and organisational challenge than a technological one.

Cooper’s explanation of the benefit also moves beyond simple labour saving. “Productivity and scale” have been the largest gains, he says. “AI enables our teams to process and analyse information at a scale that would have previously taken significantly longer. That allows investment professionals to focus more of their time on strategic thinking, risk assessment and investment decisions rather than manual data collection and analysis.”

Tobias Kassner, GARBE Industrial’s head of research and an executive-board member, describes the result as a shorter distance between raw information and decision-ready insight. “AI can accelerate the first stages of research, structure complex material and produce usable working drafts much faster,” he says. “This gives our experts more time for interpretation, challenge, stakeholder dialogue and the decisions that cannot be delegated to a model.”

This distinction is essential. If every competitor can generate a summary in seconds, speed alone ceases to distinguish one firm from another. Kassner says the initial acceleration quickly becomes a new baseline as clients, partners and rivals adopt the same tools.

“The early acceleration creates enthusiasm, but as clients, partners and other stakeholders adopt the same tools, the entire market moves faster and the relative advantage diminishes,” he says. “Sustainable value therefore depends less on speed alone and more on proprietary data, better questions, sound judgement and the ability to translate faster analysis into better decisions.”

In other words, productivity is the gateway to adoption, not its highest purpose. The more consequential ambition is to make the organisation’s accumulated knowledge easier to retrieve and apply: connecting market evidence, asset performance, past decisions and local expertise in ways that improve capital allocation.

Proprietary data

As models become widely available, several interviewees expect differentiation to migrate towards proprietary data. Cooper argues that unique datasets, domain expertise and the ability to convert information into action will become the source of advantage.

“As AI models become widely available, competitive advantage will increasingly come from unique datasets, domain expertise and the ability to turn information into actionable insight,” he says. “The firms that can successfully combine those elements will have a significant advantage.”

Wang makes much the same point: the models are being commoditised, while clean, structured and connected information remains difficult to create. “I see people still focused on using AI for relatively small tasks, like writing emails or summarising reports,” she says.

“Those are useful, but they probably represent only a fraction of what AI can actually do. The models themselves are becoming increasingly commoditised, so the real advantage is having structured, clean and connected data across your teams and workflows, and then using AI to make that information genuinely useful. People can get to the right answer faster, make better-informed decisions and spend less time digging through folders, files and spreadsheets just to find one number.”

Underused information

The appeal is obvious in a sector that generates enormous volumes of underused information. Leases can be abstracted and compared; property and spatial data can improve location decisions; historic asset performance can be linked to market movements; building data can identify energy or maintenance patterns; planning, infrastructure and labour information can be combined with supply-chain evidence.

GARBE is exploring the combination of geospatial data, satellite imagery, energy availability, infrastructure, labour markets and logistics patterns. “The competitive advantage will come from connecting trusted data with sector expertise and embedding the resulting insight in investment, location, development and asset-management decisions,” Kassner says. Such applications are much closer to real-estate intelligence than to a general-purpose chatbot.

Greatest benefit

PATRIZIA offers a glimpse of what sustained investment can produce. Cajias says the group has invested in data, analytics and machine learning since 2017. Its ImpACT Solutions now include tools for location and market intelligence, risk assessment and performance monitoring. More than 50 virtual machines process hundreds of thousands of requests each year and analyse over 20 million property, market and spatial data points.

“The greatest benefit has been the ability to scale expertise across the organisation,” Cajias says. “More than 120 AI-driven workflows support internal processes, while activities that once took weeks can now be completed in a fraction of the time.”

That phrase - scaling expertise - captures a more useful conception of AI than replacement. A high-quality analytical method need not remain confined to the small team that created it; it can be turned into a repeatable service available across markets and sectors. Institutional knowledge can become easier to find and investment professionals can spend less time assembling evidence and more time testing what it means.

For Cajias, the next opportunity lies in connecting information that is still held separately. “The industry may be underestimating the potential of AI-enabled investment intelligence,” he says, “particularly its ability to connect market data, institutional knowledge, operational information and investment expertise into intelligent systems.”

But proprietary data is not automatically good data. Real-estate records may be incomplete, inconsistent, duplicated, trapped in documents or governed by unclear ownership. Kassner says AI does not create this problem but it does make its consequences more visible.

“The biggest obstacle is turning individual experiments into repeatable, governed processes,” he says. “Access to reliable, well-structured data is fundamental, but data availability, ownership and quality remain challenging in practice. This is not an AI-specific problem. AI simply makes the consequences of fragmented or inconsistent data more visible.”

Unified technology strategy

Poor input can now generate a polished but misleading answer with extraordinary speed. Successful deployment therefore requires definitions, permissions, audit trails and traceable sources as well as technical infrastructure.

This helps explain JLL’s readiness gap. Its research says 60% of investors lack a unified technology strategy across real estate functions and asset types. Among occupiers, 70% do not have a change management framework for AI and half are insufficiently resourced in digital and AI talent. The obstacle is not an absence of promising demonstrations. It is the unglamorous work needed to make demonstrations reliable, secure and repeatable.

It is tempting to speak of the property industry moving along a single adoption curve, but the evidence suggests several curves are developing at different speeds. General purpose tools spread horizontally through an organisation: almost anyone can use them to search, summarise, translate or prepare a first draft. Specialist applications move vertically through a function and require much deeper work. Lease abstraction needs document standards and checking. Portfolio optimisation needs consistent asset data and agreed assumptions. Predictive maintenance depends on building systems. Location intelligence requires spatial information, infrastructure and local interpretation.

Kassner describes the same problem in more formal terms. “AI has not eliminated the need for validation,” he says. “Producing a plausible answer is easy. Producing a reliable, traceable and decision-ready answer is much harder.”

Human review

Models can become stuck in prompt-correction loops, repeatedly supplying generic or unhelpful responses despite more context. “AI cannot compensate for incomplete data, unclear assumptions or weak source material,” Kassner adds. “In real estate, where decisions involve capital, locations, tenants and long time horizons, domain expertise and human review remain essential.”

The caution is particularly important in real estate because errors can travel into decisions involving large sums, long holding periods, tenants, planning authorities and regulated processes. An answer can be statistically neat and commercially naive. A valuation model may miss an unusual lease event; a location score may encode yesterday’s infrastructure assumptions; a document summary may omit a clause whose importance is obvious only in context.

This does not invalidate AI. It changes the design of the workflow. Sources should be visible where possible. Material outputs should be reviewed by someone competent to challenge them. Uses should be proportionate to risk. Confidential information needs a secure environment, and responsibility must remain with an identifiable professional. These principles also fit the direction of European regulation and professional standards: governance is becoming part of normal practice, not an optional brake imposed after innovation.

The hype around autonomous agents therefore receives a cool response. Kassner regards the idea of a universal AI autopilot running the property value chain as overblown. “Real estate is contextual, relationship driven and exposed to regulation, capital-market risk and local conditions,” he says. “AI agents will become useful for defined workflows, but the promise of a universal autopilot for real estate ignores where the real complexity lies.”

Wang is equally sceptical of systems imagined as human replacements. “The more immediate opportunity is using AI to remove manual work, organise information, set reminders and help people make better decisions faster,” she says. “AI should accelerate work, improve the quality of outputs and, just as importantly, increase employees’ job satisfaction by freeing them up to focus on the more interesting and valuable parts of their jobs.”

Human expertise

None of the interviewees expects human expertise to become less important. Their argument is subtler: AI changes which parts of expertise carry the greatest value. Roles based heavily on collecting, formatting and summarising information will change first. Analysts, researchers and portfolio professionals are likely to spend less time building the information base and more time interpreting it, testing scenarios, spotting weak assumptions and explaining decisions.

Cooper expects comfort with digital and AI tools to become a core workplace skill across roles. “We continue to recruit for critical thinking, commercial judgement and collaboration,” he says, “but increasingly expect people to be comfortable working alongside AI-enabled technologies.”

He expects data-focused roles to change most. “Investment analysts, researchers and portfolio professionals will spend less time gathering and structuring data and more time interpreting information, testing scenarios and making decisions,” he says. “AI will elevate the importance of judgement and insight rather than replace them.”

Kassner agrees. “The most valuable profiles will combine real estate expertise with sufficient data and technology understanding,” he says. “Continuous development of existing teams is therefore at least as important as external recruitment.”

The academic productivity findings add an intriguing complication. In the customer-support study, AI benefited novice and lower-skilled workers most. That suggests tools may spread established knowledge and help less experienced employees move up the learning curve faster.

In property, a well-designed system might help a junior analyst locate precedents, understand a market or produce a first analysis more quickly. But faster output is not the same as deeper professional formation. If junior staff no longer perform the foundational tasks through which previous generations learnt, employers will need to be more deliberate about teaching context, scepticism and accountability.

Lower headcount?

That is why the employment story is unlikely to be a simple count of jobs lost or saved. Tasks will be redistributed inside jobs. Some routine production will shrink; other tasks will expand. Organisations may be able to handle more work without equivalent growth in headcount, but they will also need people to structure data, govern tools, redesign processes and translate technical capability into property decisions.

The human boundary is clearest where responsibility and relationships matter. “Real estate remains fundamentally a people business,” Cooper says. “Building client relationships, negotiating complex transactions, understanding local market nuances and making investment decisions in uncertain market conditions will continue to require human judgement. AI can inform those decisions, but it cannot replicate the experience and accountability that professionals bring.”

Kassner adds stakeholder alignment and the contextual understanding gained from sites, tenants and local authorities. “Most importantly, it will not take responsibility for a capital allocation or development decision,” he says. “AI can improve the evidence base, but people remain accountable for the choice.”

Cajias reaches the same conclusion. “While AI can analyse data, identify patterns and suggest actions, it cannot replace judgement, accountability, negotiation or trust,” he says. “Investment decisions ultimately require conviction, experience and responsibility.”

The recurring claim that property is a people business can sound defensive, but here it is concrete. AI can recommend; it cannot accept fiduciary responsibility. It can identify comparable transactions; it cannot know how much trust exists across a negotiating table. It can rank locations; it cannot fully absorb political mood, an occupier’s unspoken priorities or what a physical inspection reveals. Accountability is not merely another variable waiting to be automated.

What next?

The next phase of AI in European real estate will probably look less dramatic than the first but matter more. The novelty of instant drafts and conversational search is fading. Boards will ask which pilots have improved revenue, risk, service or decision quality. Employees will expect tools to connect with their real work rather than sit beside it. Regulators, clients and investment committees will ask how outputs were produced and who checked them.

The strongest companies will not necessarily be those that deploy the greatest number of applications. They will be those that select problems carefully, connect trusted information, redesign workflows and keep responsibility clear. They will use AI to distribute their best knowledge without mistaking generated fluency for experience.

The model itself is becoming less special. What surrounds it - proprietary evidence, knowledge, trust and judgement - becomes more valuable. The real test is whether a business can turn a technology almost everyone can access into decisions, services and insight that competitors cannot easily copy.

Plausibility is not proof. Generative systems can produce confident, polished answers that are generic, incomplete or wrong. Material outputs need source checks and review by somebody who understands the decision.

Poor data does not become good insight. Fragmented records, inconsistent definitions and unclear ownership limit what a model can deliver. AI often exposes an existing data problem rather than solving it.

A pilot is not a capability. Licences and experiments do not demonstrate business value. Repeatable benefits require workflow integration, secure infrastructure, governance, training and measurement.

Autonomy should be bounded. Agents may handle defined steps, but high-stakes work needs clear approval points. AI can move a task closer to the finish line but should not be near the send button.

Responsibility stays human. A system cannot own a capital-allocation decision, negotiate trust or accept professional accountability. The human reviewer is part of the product, not a temporary workaround.