Valuing a property in seconds, writing a listing, anticipating a vacancy, controlling a building's heating: AI now touches every step of real estate. The PropTech market (property technology) reaches 54.66 billion dollars in 2026 according to Precedence Research. But a JLL survey of more than 1,000 leaders shows the other side: 92% of real estate firms are testing AI, and only 5% achieve most of their objectives. For a Belgian company that sells, manages or operates buildings, the question is no longer whether to adopt AI, but how to make it deliver results.
Real estate was long run on instinct, a contact book and a spreadsheet. PropTech first brought listing portals, then management software. The current layer is analytical and generative: models that estimate a price, forecast a vacancy, write a listing or detect a consumption drift. We talk about AI in real estate when a system no longer just displays data, but produces a valuation, a forecast or content that a decision will rely on.
The underlying market is large. Precedence Research values global PropTech at 47.08 billion dollars in 2025 and 54.66 billion in 2026, with annual growth of 16.1% toward 209 billion by 2035. Residential accounts for a little over half the market, and managers and agents are the primary users. Two maturity levels coexist today.
The tool estimates, drafts, alerts and recommends. A professional validates and decides. This is the dominant level in 2026, the easiest to govern.
The agent chains tasks (tenant follow-ups, property shortlisting, rent adjustment) within set limits. Still rare in production, especially on high-stakes decisions.
The appeal of real estate is that many tasks are repetitive, documented and measurable: estimating, drafting, sorting, scheduling, monitoring. That is exactly where a model adds value, because the result can be compared with reality.
Automated valuation models (AVMs) compute a value from comparable sales, the property's features and the local market.
instant valuationWriting descriptions, multilingual translation, photo editing and virtual home staging from a simple fact sheet.
content in minutesSorting requests, answering tenants, scheduling maintenance, forecasting arrears and vacancy.
less adminDriving heating, ventilation and lighting by occupancy, detecting consumption anomalies.
optimized energyValuation is the most mature use. Zillow's Zestimate shows a median error rate of 1.83% on homes actively listed for sale in the United States: for half of them, the estimate falls within 1.83% of the final sale price. Performance drops off-market, on homes not for sale, where the median error exceeds 7% for lack of recent data. The lesson holds everywhere: a valuation model is only as good as the data feeding it.
In Belgium, where public sales now run online through Biddit, the platform of the Royal Federation of Belgian Notaries that has become the most-used form of public sale in the country, transaction data is gradually becoming structured. But a model remains a starting point, not a signed valuation opinion. It guides a negotiation, it does not replace it.
This is probably the most concrete source of value. In the European Union, buildings account for 40% of energy consumption and 36% of energy-related greenhouse gas emissions, and about 75% of the stock is considered energy inefficient. Finely driving heating, ventilation and lighting by actual occupancy cuts both the bill and emissions.
The JLL survey confirms it on the usage side: energy and emissions management platforms exceed 80% adoption, making them one of the most deployed use cases. For a Belgian owner, this is also a direct answer to tightening building energy performance requirements.
The striking figure in the JLL survey is not the enthusiasm, but the gap. 92% of occupiers and 88% of investors and owners are piloting AI, each running on average five projects in parallel among 56 identified use cases. Yet only 5% report having achieved most of their objectives.
The brake is not the technology. More than half of respondents cite incompatibility with existing systems, and the majority say they are poorly prepared on data and organization. In other words, value comes less from the model than from data quality and integration into processes. An AVM plugged into incomplete data, or a management tool that writes nowhere, will never deliver the expected return.
A valuation model used to decide a mortgage touches the assessment of a person's creditworthiness, classified as high risk by the EU AI Act (Annex III): human oversight, documentation and bias control become mandatory. Rental management, in turn, handles personal data of applicants and tenants, governed by the GDPR for collection, retention and any scoring. Before automating, you need to know where the data runs and who sees what.
AI in real estate inherits specific risks. Bias: a model trained on historical data can reproduce territorial discrimination. Data quality: off-market, an estimate can drift by more than 7%. The showcase effect: many tools claim AI without real capability. Hallucination: a generated listing can invent a non-existent feature, with legal risk. The rule stays the same: validate on your own data, keep a human on high-stakes decisions, document.
Valuation, multilingual listings and property shortlisting save time. In a trilingual market, generating and translating content has immediate value, provided it is reviewed.
Sorting requests, scheduled maintenance and energy control cut operating costs and meet building performance requirements.
Real estate data is sensitive: identities, incomes, leases, values. Entrusting it to an opaque remote service complicates compliance. A controlled deployment, compliant with the GDPR and the EU AI Act, on infrastructure you control, stays simpler to govern and more durable. A model that guides a sale, a rent or a loan must be traceable, supervised and correctable by design.
Vacancy, time to sell, energy bill, tenant response time. A quantified use case, not the technology.
Transaction history, sensor readings, digitized leases. Without clean data, no reliable valuation or forecast.
What the tool does alone, what requires a human. Oversight for high-risk uses, GDPR for tenant data.
Compare against actual results, correct bias, document. Scale only what proves its value.
Molderez Consult helps Belgian real estate players move from pilot to result: choice of use cases, data quality, GDPR and EU AI Act compliance, on controlled infrastructure.
Discuss my use case