Perspectives · 16 Nov 2026 · 5 min read
Why do real estate AI pilots stall?
By Sinclair Dobbs, CEO and Co-founder

Last year JLL surveyed more than 1,000 senior corporate real estate decision makers across 16 markets. 92% were piloting AI or planned to start within the year, up from under 5% two years earlier. Only 5% said they had achieved most of their programme goals. Among investors, owners and landlords, 88% had started piloting. The pattern holds outside property too: MIT's Project NANDA reported that 95% of organisations saw no measurable return from generative AI.
92% are piloting AI. 5% have achieved most of their goals.
Share of corporate real estate teams, JLL surveys 2023 to 2025
Every one of those teams can buy the same models, the 5% included. The difference sits in what the model is given to read: one agreed version of each number, or several competing ones.
What do people actually ask AI at work?
Surveys record intentions. Usage logs record behaviour. We audited a full year of AI assistant use inside a real estate operator, more than 30,000 interactions, and sorted every request into buckets.
Portfolio questions were ~15% of what people asked
Approximate share of requests to an AI assistant over one year
Writing and editing
~65%Spreadsheet help and general questions
~20%Questions about the portfolio
~15%
Most of the use was help with wording. The questions the business runs on, such as how an asset is performing, what a lease says or what goes in the investor report, formed the smallest bucket. Teams built the investor report by hand and then asked the assistant to tidy the sentences.
Across the year, around one request in ten was sent again because the first answer failed. The failures fell into five groups:
- Finding the file. The person knew the document existed and the assistant returned others.
- Reading the file. Large linked workbooks were read in pieces, and the answer came back wrong.
- Choosing the value. Two documents gave different figures and the assistant quoted one in silence.
- Knowing the house method. How the firm calculates a yield lived in one email and a few heads.
- Producing the report. Requests to update a recurring report were refused.
Why is a fund's number a coordination problem?
Take the area of a single building. The investment paper says 88,400 sq ft. The measured survey, done after completion, says 88,750. Both figures are correct, because contractors tend to build slightly over plan. The team knows the survey takes over once the building completes, and that rent, value and the loan all follow from it.
That knowledge is an agreement between people. The asset manager, the finance team and the valuer settled it at some point, and it now lives in habit, in an email thread, or in a folder with "outdated" in its name. Multiply it by every field in the investor report: which of three yields is quoted, which forecast is current, which letting date the lender has been given.
Development projects show it clearly. The right retention figure for a reporting period often arrives in an email from the project manager. The cost tracker and the financial model each hold an older value, and the team simply knows which to use.
An AI assistant searching the shared drive is the one colleague who was absent from all of those conversations. It finds every version, ranks them by how closely the text matches the question, and returns one with a citation. Where two documents give different letting dates for the same unit, it picks one and carries on.
Where Damus fits
Damus builds and maintains a single source of truth for institutional real estate. It reads the documents where they already live, such as email or SharePoint, and cites every value to its source. Conflicts go to your team to settle, and each decision is logged. Reports, models and AI assistants then read from that one agreed record.
Sources
- JLL, Reality check: The true pace and payoffs of AI adoption in corporate real estate, Global Real Estate Technology Survey 2025.
- JLL news release, Real estate's AI reality check, 28 October 2025.
- MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, as reported by Virtualization Review.
