Quarterly, a familiar scenario repeats itself in the institutional real estate sector: a company enters into a general AI contract or introduces a company-wide subscription to the latest model. The promise is automated underwriting, instant lease abstraction, and clean data. A substantial amount is invested. Six months later, the results are sobering. Costs exceed expectations, and the majority of the team continues to do the actual work in Excel.
According to the '2025 State of AI in Business Report' from MIT, ninety-five percent of organisations achieve no return on generative AI, despite between US$30 billion and US$40 billion in corporate investments. Typically, the technology or the users are blamed, but both explanations miss the core problem. The real bottleneck is context. Until a company solves this problem, no model will change the results, regardless of its performance.
The Problem of Lack of Adaptability
The reason projects fail is that the tools cannot retain memory, adapt to specific workflows, or improve based on user feedback. Simply put, a generic chatbot forgets what it was told last week, never learns how your company operates, and never gets better. This is why tools without a memory of your business stagnate, while those that build this memory succeed. The winners are consequently not those with the most conspicuous technology, but those that, according to the MIT study, are “embedded in workflows and adapted to context.”
Anyone can scan a lease – the documents themselves are not the difficulty. Crucial knowledge lies instead in how experienced individuals interpret these documents to make informed decisions. This layer of interpretation forms the company's intelligence. It is captured across thousands of real deals, tested against actual errors found in commercial leases, and sharpened by feedback in a live environment. History is the asset, and this history consists far more of decisions than of documents.
The Value of Contextualisation
Adoption depends not on the size of the model, but on how much it knows about your business. This means a consistent, growing record of how your company actually evaluates a deal, as well as the ability to learn from it. The commercial real estate market shows this gap most clearly. Approximately 88 percent of investors, owners, and landlords have already started AI pilot projects, according to the 'JLL 2025 Global Real Estate Technology Survey'.
Nevertheless, only 5 percent report achieving all their AI goals, despite increasing expenditure. The main obstacle is so-called data debt: fragmented information scattered across different systems and without context. The value of a lease abstraction lies in what it highlights, not just a simple summary of terms. Two companies can review the same deal and come to different conclusions for entirely valid reasons. One company might be more interested in tenant turnover, while the other focuses on the reinstatement value. One might view a submarket as early-cyclical, while another considers the same submarket to be overdeveloped.
This is why a CBRE report on an asset reads differently from a JLL report on the same property. Each company decides what data it selects, what it highlights, and what story it tells. A general model cannot make such complex distinctions. When it became clear that directly prompting a chatbot was too imprecise, the industry’s next step was to wrap an expensive model in an “agent” and expect precision. A lot of money flowed into this idea. However, it's not about
- —In-house development versus acquisition.
- —An agent based on a generic model is more sophisticated than a raw prompt.
- —But sophistication and accuracy are not the same thing.
- —If the central intelligence has never been tailored to the specific task, no orchestration can produce judgment.
The future therefore lies in a flexible layer that grows and adapts as work evolves, while consistently delivering the same quality standards. This logic also reframes the debate about adoption. Employees are already using generative AI about three times more often than their executives believe, according to a McKinsey study. People generally do not avoid AI because they resist change, but they certainly resist tools that slow them down or force them to explain the same background each time. A system with company intelligence relieves this burden, as it already understands the work and improves with use. Therefore, the companies that succeed with AI in the real estate sector will be those that build the deepest context and continuously allow the system to learn from it, rather than those that license the largest model.
Arunabh Dastidar is co-founder and CEO of the real estate investment platform Leni.
