By Arunabh Dastidar July 7, 2026 12:06 pm
reprintsEvery quarter, the same scene repeats across institutional real estate. A firm signs an enterprise AI deal or rolls out a company-wide subscription to the latest model. The pitch promises automated underwriting, instant lease abstraction and clean data. A big check goes out the door.
Six months later, the results are underwhelming. The bill is bigger than expected, and most of the team is still doing the real work in Excel.
Ninety-five percent of organizations are getting zero return on generative AI, despite $30 billion to $40 billion in enterprise investment, according to MIT’s 2025 State of AI in Business report.
The usual explanations blame the technology or the people using it, but both miss the point. The real bottleneck is context. Until a firm solves that, no model will change the outcome, no matter how powerful it is.
The reason projects fail is the fact that the tools cannot retain memory, don’t adapt to specific workflows, and fail to improve based on user feedback. In plain terms, a generic chatbot forgets what you told it last week, never learns how your firm works, and never gets better.
That is why tools with no memory of your business stall out while those that build up that memory stick. Hence the winners are not the ones with the flashiest technology, but the ones that “embedded themselves inside workflows, adapting to context,” according to the MIT study.
Anyone can scan a lease — the documents are not the hard part — but the real knowledge is in how experienced people read those documents to make decisions. That interpretive layer is the company brain. It gets captured over thousands of real deals, tested against the mistakes that actually show up in commercial leases, and sharpened through feedback in a live environment. The history is the asset, and that history is made of decisions far more than documents.
In other words, adoption does not track how big the model is. It tracks how much it knows about your business, meaning a consistent, growing record of how your firm actually reads a deal, plus the ability to learn from it.
Commercial real estate shows this gap at its widest. About 88 percent of investors, owners and landlords have started piloting AI, according to JLL’s 2025 Global Real Estate Technology Survey. Yet, only 5 percent report achieving all their AI goals even as spending climbs. The main obstacle is so-called data debt: fragmented information scattered across systems and stripped of context.
A lease abstract’s value is in what it highlights, instead of a simple summary of terms. Two firms can review the same deal and reach different conclusions for perfectly valid reasons. One may care more about tenant rollover while another may focus on replacement cost. One may view a submarket as early-cycle. Another may see the same submarket as overbuilt.
That is why a CBRE report on an asset reads differently from a JLL report on the same property. Each firm decides what to pull from the data, what to highlight, and what story to tell. A general purpose model cannot make that type of sophisticated distinction.
When prompting a chatbot directly proved too blunt, the industry’s next move was to wrap an expensive model in an “agent” and expect precision to follow. A lot of money went into that idea.
But it’s not build versus buy. An agent built on a generic model is more sophisticated than a raw prompt, but sophistication and accuracy are not the same thing. If the intelligence at the center was never shaped for the work, no amount of orchestration around it will produce judgment.
Therefore the future is a flexible layer that grows and adjusts as the work evolves, while holding to the same output standards every time.
This logic also reframes the adoption debate. Employees are already using generative AI roughly three times more than their leaders assume, according to a study by McKinsey. People generally don’t avoid AI because they resist change, but they certainly resist tools that slow them down or make them explain the same background every time. A system with a company brain removes that burden as it already understands the work, and gets better with use.
So the firms that win with AI in real estate will be the ones that build the deepest context and let the system keep learning from it as opposed to those that license the biggest model.
Arunabh Dastidar is the co-founder and CEO of real estate investment platform Leni.