The pattern is familiar by now. A firm buys seats for a chat assistant, a few analysts try it on a memo, the output reads well, and then nothing changes. Six months later the same team is still rebuilding the same model in the same spreadsheet, and the assistant has quietly become a fancier way to rewrite emails.
It is easy to read that as AI being overhyped. It is usually something more specific: the assistant never had access to anything true about your portfolio.
A chat window has no idea what you own
Ask a generic model to underwrite a deal and it will produce something that looks like underwriting. It does not know your rent roll, your debt terms, your comp set, or the way your shop treats replacement reserves. So it invents a reasonable-sounding version of all of it. The output is plausible and unusable, which is the worst combination, because someone has to check every line before it can go anywhere near an investment committee.
The fix is not a better prompt. It is context. A tool is only as useful as the data it is standing on.
Where the data actually lives
In most CRE shops the answer is: everywhere. Property performance sits in the property management system. Actuals sit in accounting. Deal status lives in someone's tracker, or in a partner's head. Rent rolls and T-12s arrive as attachments, in whatever format the broker felt like exporting. Market data comes from a subscription nobody has logged into since renewal.
None of that is unusual, and none of it is a failure of discipline. It is what happens when a business grows deal by deal. But it does mean there is no single place an AI tool can look to answer a question as simple as "which of our assets are running below underwritten NOI this quarter."
What has to be true first
Before AI produces work you would sign your name to, three things have to hold:
- The data has to be somewhere the tool can reach. Files, systems of record, market data, all in one connected layer rather than four exports and a hope.
- The tool has to know the domain. A rent roll is not a spreadsheet, it is a rent roll. A T-12 has a shape. Software that starts from CRE structure stops guessing.
- The output has to be a real artifact. Not a chat reply that evaporates, but an app, a model, a dashboard your team can open again next month and trust to be current.
The unglamorous part is the whole job
This is why we built Playgrounds around the data layer first and the AI second. Every account starts with over $50,000 worth of CRE property data already loaded, so there is something real to work against on day one. Your own files can be dragged straight in and map themselves. Your systems can be connected so an agent keeps them current. And if you would rather not do any of that yourself, our team onboards it for you.
Only then does the interesting part work: you describe what you need in a sentence, and you get a working app with its own database behind it, running on your live numbers.
The teams getting value out of AI in CRE right now are not the ones with the cleverest prompts. They are the ones who fixed the plumbing first.
Free to start, no credit card. Drop in a rent roll and build something real in minutes.
Sign up free → How the data layer works